EAAI - 2568

• (2025). A novel weight-optimized machine-learning hybrid model for daily river runoff prediction. EAAI, 162, 112396. (www)

  @article{ , abstract = {The daily runoff process has been characterized as nonlinear and unsteady due to the impacts of watershed precipitation and evaporation, vegetation coverage rate, reservoir operations and other human activities. In recent years, machine-learning (ML) models have been widely applied in the daily runoff predictions, but the robustness and effectiveness of individual ML model is always limited. A novel weight optimization scheme has been introduced to ML models to obtain accurate predictions of daily river runoff. Variational modal decomposition method is adopted in the dataset preprocessing, and the runoff prediction performance of various classic ML models, including Genetic Algorithm-Back Propagation neural network (GA-BP), Long Short-Term Memory network (LSTM), Elman neural network (Elman) and Genetic Algorithm-Support Vector Machine (GA-SVM) are subsequently evaluated. A particle swarm optimization (PSO) based weight optimization strategy is proposed to combine different types of ML models, thus more accurate and robust results could be obtained. The ten-fold cross-validation method has been adopted and the performance of the optimized hybrid models are further evaluated for different schemes. A case study at Hankou hydrological station demonstrates that root mean square error (RMSE) and mean absolute percentage error (MAPE) is improved by 35.7 %, 75.8 % respectively for the optimized hybrid model. The present study shares useful insights to the comprehensive optimization of various ML models in the intelligent management of water resources.}, archive = {J_EAAI}, author = {Zhonglian Jiang and Jianglong Ying and Zhen Yu and Xiao Chu and Chengqiang Yu}, doi = {10.1016/j.engappai.2025.112396}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112396}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel weight-optimized machine-learning hybrid model for daily river runoff prediction}, volume = {162}, year = {2025}, }
• (2025). Intelligent assessment of habitat quality based on multiple machine learning fusion methods. EAAI, 162, 112395. (www)

  @article{ , abstract = {Evaluating habitat quality can help balance the relationship between economic development and biodiversity conservation, and it serves as a foundation for constructing an ecological security pattern. However, research on the intelligent construction of habitat quality is limited. This study develops a comprehensive framework to assess habitat quality based on optimized machine learning methods. The findings of the research are as follows: (1) From the perspective of human-machine interactive interpretation, ensemble learning is used to enhance the performance of basic classifiers, resulting in a classification map with high precision and recall. (2) The particle swarm optimization (PSO) algorithm can improve the goodness of fit of the Extreme Gradient Boosting (XGBoost) inversion model by 4–5 %. (3) The habitat quality inversion method based on XGBoost-PSO has high credibility and application value, with its texture structure being the result of both expert experience and image information interaction. (4) The model demonstrates certain application potential in downscaling; under the seven-band perspective, the blue and near-infrared bands are the most important, while in the four-band perspective, green and near-infrared bands take precedence.}, archive = {J_EAAI}, author = {Kui Yang and Dongge Cui and Chengrui Wang and Qi Tang and Linguang Miao}, doi = {10.1016/j.engappai.2025.112395}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112395}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent assessment of habitat quality based on multiple machine learning fusion methods}, volume = {162}, year = {2025}, }
• (2025). Health state estimation of retired batteries based on physical constraints. EAAI, 162, 112390. (www)

  @article{ , abstract = {With the growing use of retired lithium-ion batteries, accurately monitoring their health status has become increasingly important. This study proposed a method to estimate the health of retired batteries by embedding their capacity degradation characteristics directly into the loss function of a Bidirectional Long Short-Term Memory (BiLSTM) network, combined with a Physically Informed Neural Network (PINN) model. The model is developed by incorporating the dynamics of the solid electrolyte interface (SEI) membrane, which evolves as the lithium-ion poles of the retired battery move. By combining these dynamics with the governing equations of motion, a partial differential equation (PDE) is derived. This approach integrates physical constraints, data-driven learning, and PDEs into a composite loss function. The proposed method is validated on two different datasets under varying operating temperatures. The results show that the PINN-BiLSTM model achieves a Root Mean Square Percentage Error (RMSPE) of 0.024, representing a 9.67 % improvement over the PINN-LSTM. This adaptive PINN method offers highly accurate health state predictions across temperature variations, thus supporting the sustainable use of retired batteries in secondary applications and helping to mitigate energy scarcity.}, archive = {J_EAAI}, author = {Fei Xia and Qianwen Dong and Lin Xia and Zhenyi An and Ziyang Xia and Chunyang Gong}, doi = {10.1016/j.engappai.2025.112390}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112390}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Health state estimation of retired batteries based on physical constraints}, volume = {162}, year = {2025}, }
• (2025). An optimisation approach guided by crack variation mechanism in the informer prediction model. EAAI, 162, 112381. (www)

  @article{ , abstract = {Structural health monitoring (SHM) faces a fundamental challenge in reconciling predictive performance with physical interpretability for infrastructure diagnostics. Conventional deep learning (DL) approaches neglect essential mechanisms governing crack width variation—including thermal gradients, hysteretic responses, and phase-shifted correlations—limiting their reliability in real-world applications. To bridge this gap, we propose a mechanism-guided optimization (MGO) framework that integrates domain knowledge into the Informer architecture through physics-informed enhancements: auto-correlation modeling for capturing temperature-crack hysteresis, static gated fusion for multi-feature integration, and adaptive elastic net regularization for feature selection. Validated on cable-stayed bridge monitoring data, our framework achieves significant mean absolute error reductions (MAE) (5 %–60 %) and root mean square error reductions (RMSE) (10 %–55 %) versus baseline Informer across all cracks and prediction horizons, with diebold-mariano (DM) tests confirming statistical superiority in most cases. Crucially, it demonstrates superior precision relative to six state-of-the-art benchmarks across all evaluation scenarios. The ordinary least squares (OLS)-enhanced variant further delivers volatility reduction, while sensor failure tests establish quantifiable robustness benchmarks through MAE progression from 0.013 mm to 0.391 mm. This work establishes an interpretable, physics-grounded paradigm that explicitly links environmental drivers to structural degradation.}, archive = {J_EAAI}, author = {Xujia Liu and Youliang Ding and Fei Xu and Yichao Xu and Kang Yang}, doi = {10.1016/j.engappai.2025.112381}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112381}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An optimisation approach guided by crack variation mechanism in the informer prediction model}, volume = {162}, year = {2025}, }
• (2025). Lightning risk assessment of offshore wind farms considering tropical cyclone impacts based on an improved gradient boosting algorithm. EAAI, 162, 112376. (www)

  @article{ , abstract = {In recent years, global climate warming has led to a significant increase in both the frequency and intensity of tropical cyclones (TCs). The development of TCs is often accompanied by frequent lightning activities. The risk of lightning strikes to high offshore wind turbines is substantially elevated. This study evaluates the lightning risk faced by offshore wind farms influenced by tropical cyclones. Firstly, TC paths are analyzed in both spatial and temporal dimensions by linking them with lightning data to examine the distribution of TC-related lightning, and the lightning strike characteristics of offshore wind turbines are investigated. Secondly, a Bayesian Optimization (BO)-based eXtreme Gradient Boosting (XGBoost) model for lightning risk assessment is proposed, incorporating characteristics of TC lightning and offshore wind farms as input variables. The proposed BO-XGBoost model outperforms XGBoost, Bidirectional Long Short-Term Memory (Bi-LSTM), Support Vector Machine (SVM) and Neural Network (NN), achieving a precision of 98.9 % and a recall of 98.9 % on the test set. Additionally, SHapley Additive exPlanations (SHAP) value analysis indicates that TC lightning characteristics and offshore wind farm characteristics significantly impact the model output, enhancing the accuracy of the model. The assessment outcomes provide a theoretical basis for future offshore wind farm planning and guidance for lightning protection measures in offshore wind farms.}, archive = {J_EAAI}, author = {Qibin Zhou and Kehan Chen and Xiaoyan Bian and Shangjie Chen and Gaopeng Lu}, doi = {10.1016/j.engappai.2025.112376}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112376}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightning risk assessment of offshore wind farms considering tropical cyclone impacts based on an improved gradient boosting algorithm}, volume = {162}, year = {2025}, }
• (2025). Multi-objective deployment optimization for integrated sensing and communication-enabled unmanned aerial vehicle swarm. EAAI, 162, 112368. (www)

  @article{ , abstract = {With the convergence of mobile communication, sensing, and computational networks in sixth-generation technology, the integration of sensing and communication with unmanned aerial vehicles (UAVs) is promising. This paper focuses on the contribution of artificial intelligence in optimizing the deployment of UAV swarms for multi-objective target detection applications in sixth-generation networks. Specifically, the artificial intelligence contribution lies in the development of an improved multi-objective particle swarm optimization (IMOPSO) algorithm for solving a complex multi-objective deployment problem. The problem aims to simultaneously optimize communication rate, sensing quality, and energy consumption in the deployment of UAV swarms. To address this, the proposed IMOPSO incorporates chaotic initialization, Lévy flight mutation, dynamic mutation rate, and an elimination mechanism based on opposition-based learning. These innovations are designed to enhance the algorithm’s ability to explore the solution space effectively, overcome premature convergence to local solutions, and improve solution quality. In terms of engineering applications, the IMOPSO is applied to the deployment of UAV swarms for target detection, demonstrating its ability to enhance communication and sensing performance while reducing energy consumption in practical scenarios. Through extensive simulations, we show that the IMOPSO outperforms traditional optimization methods and other baseline algorithms, achieving superior results across all optimization objectives. Specifically, the IMOPSO achieves approximately 5% higher transmission data rate, 9% better sensing quality, and 19% lower energy consumption compared to baseline algorithms across multiple test scenarios. Furthermore, the solutions obtained are not only closer to the optimal front but also more concentrated, indicating higher-quality results.}, archive = {J_EAAI}, author = {Hongjuan Li and Haiyuan Chen and Miao Wang and Jiahui Li and Hui Kang and Yuzhuo Guan and Xu Lin}, doi = {10.1016/j.engappai.2025.112368}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112368}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective deployment optimization for integrated sensing and communication-enabled unmanned aerial vehicle swarm}, volume = {162}, year = {2025}, }
• (2025). Complete information extraction for monocular depth estimation using a dual framework. EAAI, 162, 112337. (www)

  @article{ , abstract = {This paper aims to address the problem of efficient extraction of complete multi-scale information for supervised monocular depth estimation. Most of the existing depth estimation methods are based on Convolutional Neural Network (CNN). By gradually exploring the contextual and semantic features, they have achieved good results in scene depth estimation. However, with the expansion of the receptive field, global information limited by the local induction bias is gradually suppressed, resulting in the performance cannot be further improved. Recently, Transformer-based methods have been widely used to model the global correlation between features. Nevertheless, since the Transformer networks are not spatially aware enough, they usually lose local details and have no clear mechanism for reusing features when processing images. The Transformer networks perform self-attention mechanism at each location and cannot directly obtain information from other locations for features. Therefore, we propose a novel dual framework called as Transformer-CNN, which includes the Transformer-branch and the CNN-branch for monocular depth estimation. Specifically, the Transformer-branch is able to model the global contextual information and the CNN-branch can capture local spatial relationships in images. However, simply fusing these two independent branches may result in insufficient feature aggregation. To this end, we design a Parallel Feature Interaction Module (PFIM), which contains a Self-Attention Module (SAM) and a Cross-Attention Module (CAM), so as to highlight features from the Transformer-branch and the CNN-branch respectively and extract complementary information between the two branches. Meanwhile, in order to make full use of the low-level features with low quality in the scene, we propose a Low-level Information Acquisition Module (LIAM) to capture texture-related information and preserve texture details in the CNN-branch. Finally, to address the lack of multi-scale contextual information in Vision Transformer (ViT), we introduce a Wide Area Multi-scale Decoder (WAMD), which incorporates the multi-scale feature representations into the decoder part via a Wide Area Attention (WAA). Extensive experiments on benchmark datasets collected in the outdoor and indoor environments demonstrate the competitive results of the proposed method, compared with the state-of-the-art monocular depth estimation methods.}, archive = {J_EAAI}, author = {Bin Li and Dazheng Zhou and Xianjie Gao and Mingliang Zhang}, doi = {10.1016/j.engappai.2025.112337}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112337}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Complete information extraction for monocular depth estimation using a dual framework}, volume = {162}, year = {2025}, }
• (2025). Beyond the horizon: A comprehensive analysis of artificial intelligence-based weather forecasting models. EAAI, 162, 112335. (www)

  @article{ , abstract = {The field of Artificial Intelligence (AI)-based weather forecasting is growing rapidly, with continuous progress in model development, techniques, and performance improvements. This paper provides a comprehensive overview of AI-based weather forecasting models, focusing on their current status, challenges, and directions for further development. A review of more than 40 models, primarily proposed after 2015, underscores the importance of critically examining various aspects of AI-based forecasting. Unlike previous reviews that targeted only a limited number of models or features, this study addresses a complete set of aspects and analyzes existing challenges from multiple perspectives. These aspects include the Machine Learning (ML) and Deep Learning (DL) methods used, datasets, predictand parameters, overfitting, and capability for forecasting extreme weather, lead time, spatiotemporal scale, performance criteria, overfitting, data assimilation, data-driven models, and the analysis of state-of-the-art (SOTA) models such as FengWu, ClimaX, Pangu-Weather, FourCastNet, GraphCast, GenCast, and Artificial Intelligence Forecasting System (AIFS) from various viewpoints. The review also discusses current challenges, including limited historical data and data quality, small-scale weather forecasting, model explainability, uncertainty, extreme weather prediction, physical constraints, temporal adaptation, and generalization, and outlines potential future directions.}, archive = {J_EAAI}, author = {Saeid Haji-Aghajany and Witold Rohm and Piotr Lipinski and Maciej Kryza}, doi = {10.1016/j.engappai.2025.112335}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112335}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Beyond the horizon: A comprehensive analysis of artificial intelligence-based weather forecasting models}, volume = {162}, year = {2025}, }
• (2025). A domain adaptation method to defend chinese textual adversarial attacks via prompt-tuning. EAAI, 162, 112330. (www)

  @article{ , abstract = {The textual adversarial attack aims to fool existing models into making erroneous predictions by adding strategic perturbations to normal data without affecting the user’s understanding. Recently, methods based on Pre-trained Language Models (PLMs) and Large Language Models (LLMs) have shown promising performance in various Natural Language Processing (NLP) downstream tasks. However, due to significant deviations between the original and perturbed texts, these methods struggle to achieve satisfactory results in defending against textual adversarial attacks, especially in Chinese, which has unique syntactic structures. To address this issue, we propose a domain adaptation method for defending against Chinese textual adversarial attacks through a prompt-tuning model, which effectively mitigates the discrepancy between different domains. Specifically, the original and perturbed texts are treated as the source and target domains, respectively, with the textual adversarial defense task framed as a cross-domain classification problem. The soft prompt-tuning model trained in the source domain is iteratively adapted to uncover the true label information in the target domain. The graph attention network is incorporated to integrate Chinese syntactic structure information with semantic features. Through a voting mechanism on predicted labels generated by the iterative model, soft prompt-tuning is further optimized for cross-domain classification tasks. Extensive experimental results demonstrate the superior effectiveness of our method in Chinese textual adversarial defense tasks compared to baseline methods, including the state-of-the-art fine-tuning approaches for PLMs and LLMs.}, archive = {J_EAAI}, author = {Yi Zhu and Zhenglong Li and Yun Li and Yunhao Yuan and Jipeng Qiang}, doi = {10.1016/j.engappai.2025.112330}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112330}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A domain adaptation method to defend chinese textual adversarial attacks via prompt-tuning}, volume = {162}, year = {2025}, }
• (2025). Zero-shot learning augmented slow feature analysis for semantic-aware industrial process fault detection. EAAI, 162, 112321. (www)

  @article{ , abstract = {Slow Feature Analysis (SFA) has shown considerable success in the field of industrial process fault detection. Nonetheless, due to its unsupervised nature, SFA relies solely on the normal training data and overlooks the incorporation of prior process knowledge, which consequently diminishes its efficacy in early fault detection. To mitigate this limitation, this paper introduces the concept of Zero-Shot Learning (ZSL) and proposes an improved SFA approach, referred to as ZSL-SFA. This novel method leverages fault semantic representations as auxiliary knowledge to enhance fault detection sensitivity in industrial process monitoring. The ZSL-SFA framework implements a dual-model collaborative monitoring system: (1) a primary SFA model is developed using normal operational data to capture the dynamic characteristics of the process; and (2) a semantic encoding mechanism, grounded in expert knowledge, is devised to build the auxiliary model, where a probabilistic attribute learner adaptively extracts semantic information from fault attribute descriptions, facilitating effective fault knowledge transfer through similarity analysis. The monitoring outcomes from both the primary and auxiliary models are integrated using a Bayesian fusion strategy, culminating in a comprehensive ZSL-SFA monitoring system. The main advantage of this method is its ability to fully exploit prior process knowledge to enhance the basic SFA model without the need for additional labeled fault samples. Experimental validations on the Tennessee-Eastman process simulation platform are performed to indicate that the proposed ZSL-SFA method surpasses the basic SFA method in terms of fault detection performance.}, archive = {J_EAAI}, author = {Wenjie Yang and Xiaogang Deng and Lumeng Huang and Yuping Cao}, doi = {10.1016/j.engappai.2025.112321}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112321}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Zero-shot learning augmented slow feature analysis for semantic-aware industrial process fault detection}, volume = {162}, year = {2025}, }
• (2025). Multimodality based deep learning method for cancer-related T-cell receptor sequence prediction. EAAI, 162, 112318. (www)

  @article{ , abstract = {T-cell receptor sequences (TCR-seq) are closely related to cancers, and in particular, cancer-related TCR-seq are crucial in cancer diagnosis and treatment. Current prediction methods for cancer-related TCR-seq often focus solely on the sequence structure, neglecting its spatial structure. Therefore, we propose a multimodal deep learning method based on parallel and residual structures (MDPR) for the detection of cancer-related TCR-seq. MDPR can effectively integrate the spatial and sequence structure of TCR-seq for accurately identifying cancer-related sequences. First, we introduce a TCR-seq encoding method based on atomic three-dimensional spatial coordinates, allowing for more effective extraction of the spatial structural features of TCR-seq. Second, we use high-dimensional word vectors instead of the amino acid feature vectors traditionally used by other researchers. Third, we pretrain the spatial feature extraction module and then conduct joint training with the sequence feature extraction module. This approach allows the model to better consider the relationship between the two modalities, thereby improving prediction accuracy. Finally, MDPR achieved an area under the curve (AUC) of 0.971 after ten rounds of three-fold cross-validation on the dataset. The AUC of MDPR is 5% higher than that of the previous best method. In short, we propose an artificial intelligence method called MDPR, and apply it to the biomedical field. MDPR can be obtained from https://github.com/biomg/MDPR .}, archive = {J_EAAI}, author = {Junjiang Liu and Shusen Zhou and Mujun Zang and Chanjuan Liu and Tong Liu and Qingjun Wang}, doi = {10.1016/j.engappai.2025.112318}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112318}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodality based deep learning method for cancer-related T-cell receptor sequence prediction}, volume = {162}, year = {2025}, }
• (2025). A combined perspective self-supervised contrastive learning framework for human activity recognition integrating instance prediction and clustering. EAAI, 162, 112317. (www)

  @article{ , abstract = {Human activity recognition (HAR) is a significant application for wearable devices, which primarily identifies current human activities by analyzing sequential sensor data. The real-time data recording of wearable devices enables the collection of vast amount of unlabeled data. Utilizing this data for self-supervised contrastive pre-training of HAR models presents a feasible solution to the decline in recognition performance due to limited labeled data. However, traditional contrastive learning frameworks are primarily designed based on positive and negative sample pairs in the image domain. The relatively simple sequence data of HAR is prone to generating incorrect negative pairs, thus pre-training HAR models solely in this manner is unsatisfactory. Given the phenomena described above, this paper proposes an Instance Prediction and Clustering Self-supervised Contrastive Learning Framework (IPCSC) for HAR, considering the characteristics of human activity data. IPCSC circumvents negative sample pairs, instead extracting contrastive information at the instance perspective by prediction tasks among various augmented views of samples and integrating clustering concepts for contrastive learning from a holistic perspective. The primary objective is to enable the model to discern critical information within human activity data and distinguishable features between different activities, thereby improving the model’s pre-training efficacy and enhancing its downstream activity recognition performance. Numerous experimental analyses demonstrate that IPCSC outperforms other self-supervised methods, achieving an average F1-Score performance improvement of 5.65%, 4.11%, and 7.99% over supervised baselines on the UCI-HAR, MobiAct, and MotionSense datasets, respectively, with only 1% of the labeled data.}, archive = {J_EAAI}, author = {Zhixuan Yang and Kewen Li and Zongchao Huang and Zhifeng Xu and Xinyuan Zhu and Yuan Xiao}, doi = {10.1016/j.engappai.2025.112317}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112317}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A combined perspective self-supervised contrastive learning framework for human activity recognition integrating instance prediction and clustering}, volume = {162}, year = {2025}, }
• (2025). A self-supervised method for learning path-augmented knowledge graph embedding. EAAI, 162, 112315. (www)

  @article{ , abstract = {Knowledge graphs (KGs) consist of factual triples that describe relations between entities in the real world. Knowledge graph embedding (KGE) aims to map entities and relations into constantly low-dimensional vectors, which is important for lots of downstream tasks (e.g., KG completion and information retrieval). Current KGE methods primarily rely on explicit structural patterns, neglecting latent contextual semantics behind those structures and resulting in sub-optimal performance. While some methods incorporate additional data (e.g., textual descriptions), such dependencies limit applicability due to additional data requirements. Furthermore, most KGE models suffer from limited supervision with sparse labeled triples, restricting their capacity to learn comprehensive semantic features. Inspired by the simple but effective self-supervised language model word2vec, one interesting question is: Can KGE be performed as a simple self-supervised language model ? To achieve this, we innovatively propose a self-supervised KGE framework that learns entity and relation embeddings by adapting word2vec’s skip-gram objective to path sequences extracted from KGs. Our framework employs separate embedding spaces for entities and relations with an entity-relation mapping mechanism for effective interaction between the two embedding spaces. Further, to enhance the training efficiency, we introduce a markov chain-based negative sampling strategy, which generates semantically meaningful negative samples by preserving the structural contexts along KG paths. Our framework, which is the first attempt to follow the context-based self-supervised idea of language models to conduct KGE tasks, addresses the constraints of label-dependent supervised KGE techniques and obviates the requirement for external information, while simultaneously enabling effective extraction of the implicit contextual semantics inherent in triple structures. Experiments on two widely-used KGE datasets show state-of-the-art performance, demonstrating our framework’s ability to learn semantically rich representations solely from graph structure.}, archive = {J_EAAI}, author = {Tong Shen and Fu Zhang and Jingwei Cheng}, doi = {10.1016/j.engappai.2025.112315}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112315}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A self-supervised method for learning path-augmented knowledge graph embedding}, volume = {162}, year = {2025}, }
• (2025). A single-cell RNA sequencing data imputation method based on non-negative matrix factorization and multi-kernel similarity network fusion. EAAI, 162, 112313. (www)

  @article{ , abstract = {Artificial intelligence-based single-cell RNA sequencing (scRNA-seq) technology is widely used in cell type identification and disease research, but its data often contain a large number of missing values and zero values due to technical limitations and biological differences. These zero values not only affect downstream analysis, but also make it difficult to distinguish technical zero values from biological zero values. Therefore, this paper proposes a scRNA-seq data interpolation method (sc-MKNMF) based on non-negative matrix factorization and multi-kernel similarity network fusion for the first time. This method improves the accuracy of cell clustering by accurately filling some zero values. First, sc-MKNMF uses gene-cell dual-level analysis to distinguish technical zero values from biological zero values, and then calculates the similarity network of multi-kernel fusion of genes and cells respectively. Then, this method uses non-negative matrix factorization combined with similarity network to construct the objective function, and introduces sparse regularization terms to ensure the similarity between genes and cells and improve stability. In addition, sc-MKNMF is also equipped with an efficient optimization algorithm to promote its convergence by continuously updating the objective function. Finally, the verification and comparative experiments on 12 scRNA-seq datasets show that the sc-MKNMF method outperforms other advanced data interpolation methods. In addition, the extension of sc-MKNMF to the two tasks of cell trajectory inference and differentially expressed gene analysis showed significant improvement and excellent versatility.}, archive = {J_EAAI}, author = {Pei Liu and Cheng Chen and Hao Liu and Jin Gu and Xinya Chen and Ying Su and Zhiyuan Cheng and Xiaoyi Lv and Chen Chen}, doi = {10.1016/j.engappai.2025.112313}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112313}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A single-cell RNA sequencing data imputation method based on non-negative matrix factorization and multi-kernel similarity network fusion}, volume = {162}, year = {2025}, }
• (2025). A novel object detection model for sugar beet cercospora leaf spot in field scenarios based on large kernel decomposition and spatial channel interaction attention. EAAI, 162, 112311. (www)

  @article{ , abstract = {Cercospora leaf spot (CLS) is a widespread disease that seriously threatens beet yield and sugar quality. Timely detection enables farmers to take early control measures and reduce economic losses. Although artificial intelligence (AI)-based methods are replacing manual inspection in agriculture, CLS detection in complex field environments remains highly challenging due to subtle early-stage symptoms and severe occlusions caused by overlapping leaves and weeds. To address these challenges, this paper presents Cercospora Leaf Spot–You Only Look Once (CLS–YOLO), an enhanced detection model built upon You Only Look Once version 11 (YOLOv11), incorporating novel modules specifically designed for accurate CLS detection under challenging field conditions. To improve the detection of weak and early-stage symptoms, we design the Multi-Scale Large Kernel Decomposition (MSLKD) module, which enhances feature extraction for subtle lesions. Furthermore, we develop the Spatial-Channel Interaction Attention (SCIA) module to mitigate detection errors arising from occlusion and fragmented disease patterns by refining multi-scale feature representations. Experimental results demonstrate CLS–YOLO achieves superior performance, reaching an mAP@0.5 of 73.6% ± 0.2% and an mAP@0.5:0.95 of 40.6% ± 0.3% over five independent runs, outperforming twelve mainstream object detection algorithms while maintaining lightweight efficiency. To validate generalization capability across scenarios, crops, and diseases, we conducted comparative experiments on two public crop disease datasets, where our method achieved superior overall performance. In summary, this study provides an effective AI-driven solution for precise crop disease detection, contributing to the practical advancement of intelligent agriculture.}, archive = {J_EAAI}, author = {Hualong Dong and Yi Lu and Yurong Qian and Xuefei Ning and Ting Chen and Ke Tang}, doi = {10.1016/j.engappai.2025.112311}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112311}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel object detection model for sugar beet cercospora leaf spot in field scenarios based on large kernel decomposition and spatial channel interaction attention}, volume = {162}, year = {2025}, }
• (2025). ElecBench: A large language model benchmark in electric power domain. EAAI, 162, 112310. (www)

  @article{ , abstract = {Large language models (LLMs) have made substantial advancements in the field of natural language processing, necessitating the development of new benchmarks to accurately track their progress. In this paper, we introduce ElecBench, the first benchmark specifically designed for the electric power domain. ElecBench comprises 24 datasets spanning different scenarios, covering general electric power knowledge and four specific business applications, with a total of 34,030 data entries. Furthermore, we evaluate the performance of a series of open-source Chinese LLMs on ElecBench. Our experiments demonstrate that ElecBench serves as an effective benchmark for electric power scenarios and highlight that existing LLMs require further optimization to gain domain-specific knowledge and achieve better performance.}, archive = {J_EAAI}, author = {Sai Zhang and Qiaochu Huang and Qiang Zhang and Xiao Liang and Weiwei Liu and Kunlun Gao and Fei Zhou and Congcong Shi}, doi = {10.1016/j.engappai.2025.112310}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112310}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ElecBench: A large language model benchmark in electric power domain}, volume = {162}, year = {2025}, }
• (2025). Feature-enhanced edge-INverse attention network for skin lesion segmentation. EAAI, 162, 112306. (www)

  @article{ , abstract = {Skin cancer, particularly melanoma, is one of the most aggressive and deadly forms of cancer with its incidence rising globally. Early detection is crucial for improving survival rates, but the traditional dermatoscopy method is a highly time-consuming and subjective process. To resolve this issue, we propose a novel Feature-Enhanced Edge-INverse attention network (FEEINnet) model that helps to segment the skin lesion region more accurately. FEEINnet consists of three sub-networks: Feature Enhanced Mechanism (FEM) learns and extracts the fine-grained enhanced features from informative channels, the Edge Attention Mechanism (EAM) helps to precisely identify the edges of the lesion region and the INverse Attention Mechanism (INAM) generates inverse attention maps which emphasize the less confident or ambiguous regions thereby increasing the segmentation accuracy iteratively. These three sub-networks collectively help to improve feature extraction, enhance boundary detection, and refine segmentation maps, even in challenging scenarios with varying lesion sizes, shapes and pigmentation. FEEINnet consistently outperforms existing models, achieving a F1-score of 95.55%, 95.53%, and 94.52%; Intersection over Union (IoU) of 92.76%, 92.43%, and 91.34%; and Structural Similarity Index Measure (SSIM) of 94.63%, 93.51%, and 91.85% on the Human Against Machine 10000 (HAM10000), Pedro Hispano Hospital ( P H 2 ), and International Skin Imaging Collaboration 2018 (ISIC2018) datasets, respectively. The obtained results demonstrate that the proposed model has a greater ability to segment complex skin lesions more accurately.}, archive = {J_EAAI}, author = {Shivamm Warambhey and Aravindkumar Sekar}, doi = {10.1016/j.engappai.2025.112306}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112306}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature-enhanced edge-INverse attention network for skin lesion segmentation}, volume = {162}, year = {2025}, }
• (2025). A severity indicator for unsupervised fault severity shift detection of heating, ventilation, and air conditioning sub-system faults using a novel adaptive feature weighing method. EAAI, 162, 112305. (www)

  @article{ , abstract = {Fault-free functioning of Heating, Ventilation, and Air Conditioning (HVAC) systems is essential for reducing energy waste in modern-day buildings. Hence, data-driven approaches for HVAC fault detection have gained popularity. Faults become more severe with time. Fault detection reveals the presence of an anomaly, but it does not convey how critical the fault severity is. Fault severity indication provides this essential context, enabling urgent resource allocation to more severe faults, adding practical significance. However, faults being rare, obtaining substantial data at different severity levels to train supervised Machine Learning models is a realistic challenge. Therefore, we propose a method for estimating fault severity in an unsupervised setting. We define a robust Severity Indicator (SI) that reflects the shift in the severity levels of a fault. First, we define a healthy domain boundary for fault-free data using One-Class Support Vector Machines. SI scores are then computed using a novel adaptive feature weighing algorithm that assigns weights to individual features, adaptively, for every fault. We focus on detecting the shift in severity, rather than quantifying it. The study of the robustness of SI for different faults in HVAC subsystems, chillers, and air handling units (AHUs) yields consistently promising results. Our comparative analysis shows that our method outperforms the unweighted approach and existing state-of-the-art techniques for fault severity estimation. Notably, our method excels in detecting low-severity faults, addressing a common limitation in current methods.}, archive = {J_EAAI}, author = {Ramnath V. Prabhu Bam and Rajesh S. Prabhu Gaonkar and Clint Pazhayidam George}, doi = {10.1016/j.engappai.2025.112305}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112305}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A severity indicator for unsupervised fault severity shift detection of heating, ventilation, and air conditioning sub-system faults using a novel adaptive feature weighing method}, volume = {162}, year = {2025}, }
• (2025). Propeller fault-detection method for electric-propulsion aircraft using motor signals and generative model-based semi-supervised learning. EAAI, 162, 112301. (www)

  @article{ , abstract = {Failures in propulsion components, such as propellers, can critically affect flight safety; thus, early failure detection, preferably before flight, is essential. Traditional fault-diagnosis methods typically rely on additional sensors or operational data, which may not be available or practical in all situations. This study addresses these challenges by introducing motor-electric-signal-based fault diagnosis that is independent of airframe configuration and can detect faults, even when the aircraft is not in operation. However, difficulties arise owing to poor class variance in motor-electric-signal data and the challenge of obtaining fault data. To overcome these issues, a semi-supervised learning model based on a modified variational autoencoder-generative adversarial network (VAE-GAN) is proposed, which predicts faults using only normal motor-electric-signal data. Additionally, a new preprocessing method and patch-based ensemble inference technique are introduced to improve the poor class-variance characteristics of the data, thereby enhancing the prediction performance. This work demonstrates that propeller faults can be successfully diagnosed using motor-electric signals without the need for additional sensors or fault-data acquisition.}, archive = {J_EAAI}, author = {Sanga Lee and Dohyeong Kim and Minkyun Noh and Shinkyu Jeong and Jikang Kong and Youngjun Yoo}, doi = {10.1016/j.engappai.2025.112301}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112301}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Propeller fault-detection method for electric-propulsion aircraft using motor signals and generative model-based semi-supervised learning}, volume = {162}, year = {2025}, }
• (2025). A deep Q-network-driven multi-objective evolutionary algorithm for distributed heterogeneous hybrid flow shop scheduling with worker fatigue. EAAI, 162, 112292. (www)

  @article{ , abstract = {The hybrid flow shop scheduling problem (HFSP) is a prominent challenge in advanced manufacturing systems. Existing research often overlooks the impact of workers in production shops or treats worker fatigue as a static parameter, failing to capture its nonlinear accumulation and recovery effects on processing efficiency. However, with the advent of Industry 5.0, there has been a growing emphasis on the critical role of human factors in production scheduling. As a result, designing an effective algorithm for HFSP that considers human factors has become a prominent research focus. In this paper, an extended distributed heterogeneous hybrid flow shop scheduling problem with the dynamic effects of worker fatigue (DHHFSP-WF) is investigated. To address this problem, a Deep Q-Network-based multi-objective optimization algorithm (DQNMOEA) is designed to minimize makespan, total energy consumption (TEC), and total worker idle time (WIT). In DQNMOEA, a four-dimensional vector encoding scheme considering worker allocation represents solution, and a reconstruction strategy ensures initial population quality and diversity. Moreover, an improved order crossover, two-point crossover, and a segment-based recombination mutation method are proposed to enhance the global search performance of the algorithm. Then, a problem-specific local search strategy is designed for each layer of the vector, allowing the Deep Q-Network (DQN)-based adaptive decision-making mechanism to perform local perturbations on the current non-dominated solutions in the most suitable dimensions. Finally, seven algorithms are adopted to make a comparison on 36 sets of instances, the experimental results indicate that DQNMOEA exhibits competitive performance in solving DHHFSP-WF.}, archive = {J_EAAI}, author = {Jianlin Zhang and Longbin Ma and Wu Zhao and Jie Cao and Zuohan Chen and Tianpeng Xu}, doi = {10.1016/j.engappai.2025.112292}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112292}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep Q-network-driven multi-objective evolutionary algorithm for distributed heterogeneous hybrid flow shop scheduling with worker fatigue}, volume = {162}, year = {2025}, }
• (2025). Hierarchical binary diffing with inductive code representation learning with graph sampling-and-aggregate. EAAI, 162, 112279. (www)

  @article{ , abstract = {Binary program diffing, or simply binary diffing, is a type of program analysis technique that quantifies the similarity between two binary programs to derive their differences. In particular, binary diffing is an essential technique for uncovering vulnerabilities and potential attack vectors in industrial control systems, where patch deployment is complicated by closed and restricted environments. Studies on binary diffing can be broadly categorized into dynamic analysis-based, static analysis-based, and neural network-based approaches. Each category of existing studies has its shortcomings, including limited coverage, low accuracy, and issues with on-demand learning. In this paper, we propose the binary diffing with sampling-and-aggregate, a hierarchical binary diffing model that generates inductive code representations based on graph sampling-and-aggregate. Our model sequentially produces instruction-level embedding, block-level embedding, and function-level embedding from the inter-procedural control flow graph of a given program, and then performs hierarchical code diffing based on these embeddings. We formally define the detailed models and present the algorithm of hierarchical binary diffing. Additionally, we conduct a thorough analysis of this algorithm, deriving several advantages. We implemented a prototype and evaluated it on a large-scale dataset in a cross-version, cross-optimization, and obfuscation settings. Our prototype showed F1-scores up to 0.96 and 0.968 in cross-version setting for function and basic block diffing, respectively. Also, our method demonstrated its robustness over several binary obfuscations. In conclusion, our proposal, which generates basic block- and function-level embedding by considering the control flow, has solid advantages on binary diffing and shows the robustness on the binary tampering.}, archive = {J_EAAI}, author = {Seungho Jeon and Kijong Koo and Daesung Moon and Jung Taek Seo}, doi = {10.1016/j.engappai.2025.112279}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112279}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical binary diffing with inductive code representation learning with graph sampling-and-aggregate}, volume = {162}, year = {2025}, }
• (2025). Nonlinear excavation load prediction of hydraulic excavator based on gated recurrent unit neural network. EAAI, 162, 112278. (www)

  @article{ , abstract = {Excavator arms are integral to the mining and construction industries, where real-time excavation load prediction is a critical element for the advancement of automated excavation technology. This study presents a novel Physics-guided Neural Network (PGNN) designed to predict the excavating force of hydraulic cylinders used in earthwork excavation. The PGNN model synergizes the physical load model of excavators with a Gated Recurrent Unit (GRU) neural network and is optimized using the Hyperband algorithm to attain both high-speed and precise forecasting. Through comparative experiments, the study validates the PGNN model's ability to achieve optimal response speed and precision in predicting excavation loads. Additionally, the predictive performance of the PGNN model is assessed via a Hardware-in-the-loop (HIL) test, conducted within the context of an actual excavation experiment. This research introduces a promising approach that seamlessly integrates physics-based modeling with machine learning techniques, facilitating real-time load forecasting for excavators. The findings pave the way for more efficient and precise excavation processes, with implications for the broader fields of mining and construction automation.}, archive = {J_EAAI}, author = {Jinshi Chen and Yue Yu and Dongyang Huo and Han Zhang and Jingyan Wang}, doi = {10.1016/j.engappai.2025.112278}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112278}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Nonlinear excavation load prediction of hydraulic excavator based on gated recurrent unit neural network}, volume = {162}, year = {2025}, }
• (2025). Plug-and-play fine grained neural cognitive diagnosis framework. EAAI, 162, 112276. (www)

  @article{ , abstract = {Cognitive diagnosis (CD) is a core task in intelligent education, which accurately assesses students’ mastery of specific knowledge concepts (KCs) by analyzing their answer records. However, existing methods mainly rely on explicit interaction data and use diagnostic models for automatic knowledge proficiency inference. These methods lack systematic optimization for fine-grained knowledge level representation, making it difficult to fully reflect students’ true learning status. To address this, this paper introduces a Plug-and-Play F ine Grained N eural C ognitive D iagnosis Framework (FNCD) with Knowledge-Level Constraint Awareness . The framework combines a knowledge proficiency evaluation module with students’ answer records and a Q-matrix to statically assess knowledge mastery. It uses a student similarity construction method based on random grouping to reveal latent learning pattern associations. Additionally, it employs a multi-scale relational learning strategy and a Top-k attention-enhanced graph network mechanism to dynamically adjust the student similarity relationship network, accurately modeling the complex learning relationships between students. Ultimately, a joint training mechanism is used to optimize the outputs of each module, significantly improving the rationality, interpretability, and accuracy of CD. The experimental results demonstrate that FNCD, as an artificial intelligence-driven plug-and-play module, can be effectively integrated into existing CD models to enhance the modeling of fine-grained knowledge mastery and improve diagnostic accuracy, showcasing the application potential of artificial intelligence in personalized education.}, archive = {J_EAAI}, author = {Xinjie Sun and Qi Liu and Kai Zhang and Weiyin Gong and Shuanghong Shen and Fei Wang and Yan Zhuang and Meikai Bao and Shijin Wang and Yuling Ma and Enhong Chen}, doi = {10.1016/j.engappai.2025.112276}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112276}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Plug-and-play fine grained neural cognitive diagnosis framework}, volume = {162}, year = {2025}, }
• (2025). A lightweight citrus detection and counting method based on deep learning model. EAAI, 162, 112268. (www)

  @article{ , abstract = {Picking robots have become an important development direction of smart agriculture, and the accurate detection, counting and lightweight deployment of fruits are the technical basis for realizing robot picking. However, due to complex weather conditions and the possible mutual occlusion between branches and citrus, it is challenging to accurately detect and count citrus in orchards. This study proposes a lightweight small target detection model for detecting and counting citrus, and deploys it on the citrus detection platform. The model first introduces FasterNet Block into the cross-stage partial feature fusion module of the backbone network to reduce the number of parameters and calculations while improving the detection accuracy of the network. Secondly, a multi-scale attention mechanism is added to the backbone network to enhance the feature extraction ability of the network. Finally, a bounding box loss function based on a dynamic non-monotonic focusing mechanism is used to increase the model convergence speed and further improve the model accuracy. Experimental results show that the model has an accuracy of 92.7%, an average precision of 91.7%, and a model size of only 5.37 megabytes. The lightweight model is applied to the citrus detection platform. Based on this application, a citrus counting method is proposed, which obtains a mean absolute error (MAE) of 0.92, a root mean square error (RMSE) of 1.28, a determination coefficient ( R 2 ) of 0.98, and a frame rate of 80.6 per second, which meets the requirements of real-time citrus detection and counting. This provides technical support for the subsequent deployment and counting research of picking robots.}, archive = {J_EAAI}, author = {Jiqing Chen and Mingchang Zhang and Bin Lu and Quan Chen and Zhiwu Jiang and Peilin Li and Jingyao Gai}, doi = {10.1016/j.engappai.2025.112268}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112268}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight citrus detection and counting method based on deep learning model}, volume = {162}, year = {2025}, }
• (2025). Stacked machine learning-based probabilistic seismic demand and risk assessment of railway embankments with variable slopes. EAAI, 162, 112245. (www)

  @article{ , abstract = {Accurate seismic risk assessment of railway embankments is critical for risk mitigation, seismic design, and emergency planning. However, conventional methods often suffer from computational inefficiency and complexity. This study proposes a novel machine learning (ML) framework to rapidly and accurately evaluate probabilistic seismic demand and risk for railway embankments. Latin hypercube sampling is utilised to generate representative soil parameter samples to construct numerical models for simulating dynamic responses under near-fault pulse-like ground motions. The peak permanent settlement (PPS) of the embankment surface is used as the key performance metric. Multiple ML models, including decision trees, random forests (RFs), extreme gradient boosting (XGBoost), artificial neural networks (ANNs), and a stacked ML model that integrates RFs, XGBoost, and ANNs, are trained and compared. The stacked ML model outperforms the other models and achieves the highest predictive accuracy for the PPS. SHapley Additive exPlanations are used to identify the velocity spectrum intensity (VSI) and the internal friction angle of the embankment as the most influential factors. Seismic fragility and risk curves are subsequently developed. The VSI and a power-law seismic hazard function are combined to estimate the annual exceedance probabilities for three seismic design criteria levels. The proposed ML framework significantly enhances the efficiency of seismic risk analysis while maintaining high precision, thereby providing a transformative approach for the seismic assessment of railway embankments.}, archive = {J_EAAI}, author = {Pan Si and Liang Tang and Shuang Tian and Xianzhang Ling and Yanfang Liu}, doi = {10.1016/j.engappai.2025.112245}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112245}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Stacked machine learning-based probabilistic seismic demand and risk assessment of railway embankments with variable slopes}, volume = {162}, year = {2025}, }
• (2025). Adaptive chemical industrial processes fault detection model based on sparse filtering-based improved mixed-gaussian probabilistic principal component analysis considering low-probability events. EAAI, 162, 112236. (www)

  @article{ , abstract = {Probabilistic Principal Component Analysis (PPCA) is widely used in process monitoring. However, its underlying assumption that data follows a Gaussian distribution limits its effectiveness in handling Low Probability Events (LPEs), which often deviate from this assumption. To address this challenge, we propose a novel method called Sparse Filtering-based Improved Mixed-Gaussian Probabilistic Principal Component Analysis (SFIMPPCA) for enhanced LPEs detection. First, a Sparse Filtering (SF) preprocessing technique with an incremental structure is employed to extract the most discriminative features. Second, to address the distortion caused by LPEs, a dynamic ratio correction mechanism based on statistical variability is introduced, followed by a newly designed Mixed-Gaussian Probabilistic Principal Component Analysis (MPPCA). Third, a Bayesian Optimization Algorithm (BOA) is applied to automatically adjust control limits, enhancing the accuracy and reliability of fault detection. The effectiveness of the proposed method is validated using the Tennessee Eastman (TE) process and the Tin Chemical Process (TCP). Experimental results demonstrate that the proposed method significantly improves performance under LPEs conditions, achieving a 10%–12% improvement in most cases.}, archive = {J_EAAI}, author = {Chuangyan Yang and Jiande Wu and Peng Li and Xun Lang and Mingxi Ai and Hancheng Wang}, doi = {10.1016/j.engappai.2025.112236}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112236}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive chemical industrial processes fault detection model based on sparse filtering-based improved mixed-gaussian probabilistic principal component analysis considering low-probability events}, volume = {162}, year = {2025}, }
• (2025). MAE-diff: Masked-AutoEncoder-guided diffusion framework for source-free domain adaptive medical image segmentation. EAAI, 162, 112219. (www)

  @article{ , abstract = {Domain gaps can often cause dramatic performance deterioration when applying medical image segmentation models trained on the source domain to the target domain. Although unsupervised domain adaptation methods can address the domain gap challenge to some extent, their reliance on accessing source images largely hampers their practical applicability, as source data are often inaccessible due to privacy concerns. Moreover, the low-quality characteristic of medical images can further degrade the domain adaptation performance of segmentation models. To address these issues, we propose the Masked-AutoEncoder-guided Diffusion (MAE-Diff) framework for source-free domain adaptive medical image segmentation. MAE-Diff mainly consists of a Masked AutoEncoder (MAE) Module for effective feature extraction and domain adaptation, and a Diffusion Module for effective segmentation of low-quality medical images. On source images, the MAE encoder is trained to extract image-specific features, and the Diffusion Module is trained to generate segmentation maps following a gradual denoising strategy, under the guidance of features extracted by the MAE encoder. Training on the target domain involves only fine-tuning MAE (trained on the source images) with target images, allowing MAE-Diff to adapt to the target domain distribution. Inference on target images can then be made by the source-based Diffusion Module, under the guidance of features extracted by the MAE encoder fine-tuned on the target images. Extensive experiments on three datasets demonstrate the effectiveness of the proposed framework for source-free domain-adaptive medical image segmentation. The code of MAE-Diff is available at https://github.com/xuss804/MAEDiff .}, archive = {J_EAAI}, author = {Shanshan Xu and Le Xu and Yeqing Yang and Lixia Tian}, doi = {10.1016/j.engappai.2025.112219}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112219}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MAE-diff: Masked-AutoEncoder-guided diffusion framework for source-free domain adaptive medical image segmentation}, volume = {162}, year = {2025}, }
• (2025). Dynamic-static siamese Takagi–Sugeno-kang fuzzy system with inductive-reflection deep fuzzy rule. EAAI, 162, 112199. (www)

  @article{ , abstract = {The first-order Takagi–Sugeno–Kang (TSK) fuzzy classifiers are famous for their high computational efficiency and strong interpretability, but they often struggle to learn from complex and large-scale datasets, and perform not very well compared to higher-order TSK fuzzy classifiers. To address this issue, in this paper we propose a novel Dynamic-Static Siamese TSK Fuzzy classifier with Inductive-Reflection Deep Fuzzy rule. It aims to enhance the model’s self-learning capabilities by utilizing Siamese network to integrate deep fuzzy knowledge and fine-grained knowledge without the need for a teacher model. The innovations of this study are as follows: (1) The deep fuzzy rules in the proposed classifier are enriched with an “Inductive-Reflection” process, which reduces constraints on traditional basic fuzzy rule and aligns rule acquisition more closely with general human thinking manners; (2) The proposed method includes a mechanism for self-learning and improvement from both deep and fine-grained fuzzy knowledge, eliminating the complexity of retraining a new teacher model; (3) An adaptive learning function is developed to effectively adjust the learning process, adapting to tasks with different complexities. Extensive experiments results on benchmark datasets, as well as two real-world datasets, demonstrate the effectiveness of the proposed classifier in terms of classification accuracy and weighted F1-score.}, archive = {J_EAAI}, author = {Xiongtao Zhang and Qihuan Shi and Yunliang Jiang and Qing Shen and Jungang Lou and Ruiqin Wang}, doi = {10.1016/j.engappai.2025.112199}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112199}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic-static siamese Takagi–Sugeno-kang fuzzy system with inductive-reflection deep fuzzy rule}, volume = {162}, year = {2025}, }
• (2025). A new design of arithmetic and logic unit for enhancing the security of future internet of things devices using quantum-dot technology. EAAI, 162, 112113. (www)

  @article{ , abstract = {The Internet of Things (IoT) is a network of interconnected devices that collect, monitor, analyze, and exchange data. This technology plays a crucial role in the smart city infrastructure by seamlessly interconnecting various nodes. The extensive application and recognition of IoT across multiple city domains, such as healthcare, transportation, energy, education, and agriculture, bring significant challenges, with security among the most pressing. Traditional hardware technologies like Complementary Metal Oxide Semiconductor (CMOS) and Very Large Scale Integration (VLSI) suffer from limitations such as high power consumption and insufficient scalability, which hinder secure and sustainable IoT deployment. Such limitations have prompted the need to seek other technologies that would serve the dual purpose of providing security as well as energy. Quantum-based technologies can become adequate candidates offering promising solutions to make IoT devices and sustainable systems more secured. Quantum-dot Cellular Automata (QCA) has been proposed as a nanotechnology with the potential of consuming ultra-low powers, less area, and high-speed operation. QCA enhances security through sustainable computing objectives by minimizing energy usage. To improve the future security and efficiency of IoT hardware, this paper suggests a QCA-based Arithmetic Logic Unit (ALU). This ALU can generate more than 12 logical and arithmetic operations. Designed together with the majority gates, XOR gates, multiplexers, and full adders, the ALU is simulated using the QCA-Designer 2.0.3. Simulated results indicate improvements in the number of cells and reduced occupied area relative to the earlier designs. These results indicate the potential of QCA technology in enabling secure, energy-efficient, and compact computing architecture applicable in the future IoT.}, archive = {J_EAAI}, author = {Maryam Zaker and Seyed Sajad Ahmadpour and Nima Jafari Navimipour and Muhammad Zohaib and Neeraj Kumar Misra and Sankit Kassa and Ahmad Habibizad Navin and Arash Heidari and Mehdi Hosseinzadeh and Omar I. Alsaleh}, doi = {10.1016/j.engappai.2025.112113}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112113}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new design of arithmetic and logic unit for enhancing the security of future internet of things devices using quantum-dot technology}, volume = {162}, year = {2025}, }
• (2025). GhostPointNet: A deep learning-based method for ghost point noise detection in four-dimensional (4D) millimeter-wave radar point clouds of underground mine. EAAI, 161, 112380. (www)

  @article{ , abstract = {The high dust concentration, multi-metal supports, and narrow winding tunnels in underground mines collectively lead to frequent ghost point noise in four-dimensional (4D) millimeter-wave radar point clouds, posing serious challenges for mining perception and localization. To address this, we propose a deep learning algorithm, named GhostPointNet, for 4D millimeter-wave radar ghost point detection in underground mining environments. From an artificial intelligence perspective, this model thoroughly considers the multi-modal features of 4D millimeter-wave radar and the environmental complexity of underground mines. It incorporates multi-parameterized spatial information inputs in both Cartesian and Spherical coordinates, coupled with “Double T-Net” adaptive alignment correction, while integrating non-spatial information such as radar power and Doppler data to achieve multi-modal representation and end-to-end discrimination between ghost points and real points. Experimental validation shows that GhostPointNet achieves excellent performance in underground mining scenarios with 92.45 % accuracy and 95.84 % F1-score, outperforming traditional filtering, clustering, and machine learning algorithms. From an engineering application perspective, GhostPointNet is specifically designed for ghost noise detection in underground mines. Even in complex scenarios such as mine tunnel intersections and turns, it preserves critical structural points. Its end-to-end neural network simplifies post-processing procedures, enhances operational efficiency, and provides stable and reliable perceptual support for subsequent tasks such as autonomous mine locomotive navigation and three-dimensional (3D) structure reconstruction. Experimental results demonstrate that this method surpasses baseline approaches in ghost point detection, real point preservation, and generalization capability, providing significant support for improving underground mining safety and efficiency.}, archive = {J_EAAI}, author = {Hu Liu and Zhenghua Zhang and Jing Yang and Jörg Benndorf and Xiaofei Wang and Jiaqi Dong and Zitao Lin and Guoliang Chen}, doi = {10.1016/j.engappai.2025.112380}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112380}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {GhostPointNet: A deep learning-based method for ghost point noise detection in four-dimensional (4D) millimeter-wave radar point clouds of underground mine}, volume = {161}, year = {2025}, }
• (2025). Hybrid ensemble learning for predicting peak deviatoric stress in soil-rock mixtures from triaxial test data. EAAI, 161, 112377. (www)

  @article{ , abstract = {Peak deviatoric stress ( q f ) is an important mechanical property index of soil-rock mixtures (SRM). However, determining it through experiments is a challenging and time-consuming task. This study presents a weighted averaging ensemble model for predicting the q f of SRM. 585 sets of consolidated drained triaxial test data for SRM were collected to construct the proposed models. Firstly, taking the extreme gradient boosting (XGBoost) as an example, the accuracy of the models established with different input parameter combinations was tested to determine the optimal inputs. Then, five artificial intelligence models were used to train the dataset, and the Bayesian optimization method was adopted for hyperparameter adjustment. Based on the prediction results, two base models (i.e. XGBoost and random forest (RF)) with good performance were selected from these five models, and a novel weighted averaging ensemble model was developed to predict the q f of SRM. The coefficient of determination (R 2 ), root mean squared error (RMSE) and mean absolute error (MAE) values of the ensemble model were 0.990, 220.0 kPa (kPa) and 118.4 kPa, respectively, for all datasets, indicating that this model has great potential in predicting the q f of SRM. In addition, the robustness, parameter patterns and the SHapley Additive exPlanations (SHAP) analysis of the ensemble model were also conducted in this study. The results of this study can be applied to rapid analysis of mechanical parameters for SRM in engineering, and are of great significance for geotechnical engineering design and disaster prevention.}, archive = {J_EAAI}, author = {Ruiliang Zhang and Xinhua Xue}, doi = {10.1016/j.engappai.2025.112377}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112377}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid ensemble learning for predicting peak deviatoric stress in soil-rock mixtures from triaxial test data}, volume = {161}, year = {2025}, }
• (2025). Graph attention neural network for advancing medical imaging by enhancing segmentation and classification. EAAI, 161, 112372. (www)

  @article{ , abstract = {Brain tumor classification and segmentation using Magnetic Resonance Imaging (MRI) scans remain challenging due to limited spatial relationships and the inability of conventional models, such as Convolutional Neural Networks (CNNs), to effectively capture structural dependencies. To address these limitations, this study introduces GANN-Med (Graph Attention Neural Network for Medical Imaging), a novel framework designed to improve the accuracy and reliability of brain tumor detection, segmentation, and classification. The framework utilizes the publicly available Brain Tumor MRI dataset and applies preprocessing followed by wavelet transform to extract both spatial and frequency features, generating wavelet-based node embeddings that represent fine textures and overall tumor structure. These features are modeled as a graph and processed through a U-Net (U-shaped Convolutional Network) architecture integrated with Graph Attention Networks (GATs), enabling adaptive attention to be directed to critical tumor regions. The segmented tumor regions are then classified using GraphSAGE (Graph Sample and Aggregation) with a pooling aggregator, which supports inductive neighborhood learning while preserving topological consistency. Unlike existing models such as mResU-Net (Modified Residual U-Net) and LG-GNN (Local-Global Graph Neural Network), GANN-Med uniquely combines wavelet-based multi-resolution graph embeddings with attention mechanisms in a unified architecture. Experimental results demonstrate a significant performance boost, including an 18.47 % increase in Dice Similarity Coefficient (DSC), notable improvements in Jaccard Index and sensitivity, a 21.04 % reduction in false positives, and an overall classification accuracy of 93.2 %, with a balanced accuracy of 91.8 % across glioma, meningioma, pituitary, and no tumor classes. Additionally, the model achieves minimal training and validation losses of 43.2 % and 41.6 %, respectively, highlighting its potential to minimize misdiagnosis and contribute to more accurate, early-stage clinical decision-making in neuro-oncology.}, archive = {J_EAAI}, author = {Meshari D. Alanazi and Khaled Kaaniche and Mohammed Albekairi and Turki M. Alanazi and Munid Alanazi and Ghulam Abbas}, doi = {10.1016/j.engappai.2025.112372}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112372}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph attention neural network for advancing medical imaging by enhancing segmentation and classification}, volume = {161}, year = {2025}, }
• (2025). Fuzzy scheduling in distributed heterogeneous printed circuit board assembly lines: Feedback assisted neighborhood-based search coupling with rapid evaluations. EAAI, 161, 112353. (www)

  @article{ , abstract = {Printed Circuit Board (PCB) production involves multiple heterogeneous and highly automated assembly lines running concurrently. These assembly lines involve complex processes with uncertain and fuzzy production times, presenting significant challenges for scheduling. Moreover, PCB production is notably affected by carryover sequence-dependent setup times (CSDST) and the need for product grouping, further complicating the scheduling process. To address these challenges, this paper formulates the scheduling problem as a fuzzy distributed heterogeneous flowshop group scheduling problem with carryover sequence-dependent setup times (FDHFGSP/CSDST). A novel and effective feedback-assisted population-based neighborhood search (FPBNS) algorithm, coupling with rapid evaluation of neighboring solutions, is proposed to solve this problem. The feedback strategy operates on two main levels: evaluating population quality and adjusting the search process. Additionally, an integrated heuristic method is employed to generate a high-quality and diverse initial population. To further enhance solution quality, a collaborative operation is designed to improve interaction between solutions. This comprehensive algorithm significantly enhances its adaptability in PCB manufacturing, providing an efficient solution to the complex scheduling challenges of real-world production environments. Comprehensive and systematic experiments were conducted to evaluate the effectiveness of the proposed algorithm and related methods, demonstrating its superior performance in solving the complex scheduling problem in PCB manufacturing.}, archive = {J_EAAI}, author = {Zhen-duo Han and Biao Zhang and Chao Lu and Lei-lei Meng and Wen-qiang Zou}, doi = {10.1016/j.engappai.2025.112353}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112353}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzy scheduling in distributed heterogeneous printed circuit board assembly lines: Feedback assisted neighborhood-based search coupling with rapid evaluations}, volume = {161}, year = {2025}, }
• (2025). A high-precision acupoint recognition and localization model for acupuncture robot end-effectors. EAAI, 161, 112348. (www)

  @article{ , abstract = {Accurate acupoint recognition is fundamental to therapeutic efficacy of acupuncture robots, presenting a crucial challenge in medical robotics: high-precision, real-time localization of sparse features on deformable skin surfaces. To address this, we propose the Acupoint Recognition Network (ACU-Net) for reliable acupoint identification to guide robotic acupuncture. From an artificial intelligence (AI) perspective, the core contribution of ACU-Net lies in its novel modules. The Convolutional Dilatation with Residual and Split (CDRS) module uses dilated convolution to expand the receptive field and overcomes the sparsity of acupoint features by capturing both anatomical contexts and subtle local gradients. The Cross-Scale Feature Fusion (CSFF) architecture integrates spatial and semantic information more effectively than standard Feature Pyramid Network (FPN). Formulated as a keypoint regression task, the network is jointly optimized using a weighted Object Keypoint Similarity (OKS)-based loss function. Evaluated on a physician-annotated, large-scale chest acupoint dataset, ACU-Net has achieved a mean Average Precision (AP 50-95 (B)) of 94.1 % for acupoint bounding boxes and 99.5 % for acupoint points (AP 50-95 (P)), outperforming mainstream Convolutional Neural Networks (CNNs) and Transformer models. Its average prediction error (0.282 cm) on volunteers represents a 30.9 % reduction compared to the second-best model of Real-Time Multi-person Pose Estimation (RTMPose). Robustness test on an independent forearm dataset confirmed its precision. Attention visualization showed strong focus on acupoint regions. Testing on chest acupoint under adverse imaging conditions of low illumination and information loss demonstrated its reliability with zero missed detection. This work significantly advances the development of acupuncture robotics for intelligent diagnosis and treatment.}, archive = {J_EAAI}, author = {Ailing Tan and Hao Mu and Wei Ma and Yu'e Lv and Haoyu Wang and Yong Zhao}, doi = {10.1016/j.engappai.2025.112348}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112348}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A high-precision acupoint recognition and localization model for acupuncture robot end-effectors}, volume = {161}, year = {2025}, }
• (2025). Eviformer: An uncertainty fault diagnosis framework guided by evidential deep learning. EAAI, 161, 112328. (www)

  @article{ , abstract = {Unpredictable signals are commonly encountered during equipment operation, and existing deep learning-based fault diagnosis methods often fail to accurately evaluate the uncertainty of diagnostic results, limiting the model's capacity to respond to unexpected signals. To address this challenge, this study introduces a modified Transformer model based on deep evidential learning—Eviformer. The proposed model first employs a Swin-Transformer (ST)-based distribution projector, which preserves the advantages of ST in extracting features from vibration signals and simultaneously projects these signals directly into a Dirichlet distribution with second-order probabilities. Furthermore, by incorporating novel evidence correction terms and a constraint factor to reconstruct the evidence constraint loss, more precise uncertainty quantification in diagnostic predictions is achieved. Using a gear-bearing vibration dataset, comparative experiments were conducted across various scenarios, including out-of-distribution gear faults, faults in unmonitored components, noise interference, and variable speed conditions. The results demonstrate that the proposed method can promptly issue uncertainty-based warnings when encountering vibration signals that significantly differ from the training set distribution, thereby offering essential support for maintenance decisions.}, archive = {J_EAAI}, author = {Jingjie Luo and Fucai Li and Xiaolei Xu and Wenqiang Zhao and Dongqing Zhang}, doi = {10.1016/j.engappai.2025.112328}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112328}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Eviformer: An uncertainty fault diagnosis framework guided by evidential deep learning}, volume = {161}, year = {2025}, }
• (2025). Multi-scale spatiotemporal feature fusion for super-resolution video reconstruction in dynamic scenes. EAAI, 161, 112327. (www)

  @article{ , abstract = {High-quality geospatial video reconstruction is vital for applications such as urban surveillance and disaster assessment. However, existing methods often suffer from motion blur and visual artifacts due to hardware limitations and dynamic environments with rapid motion, frequent background changes, and lighting variations. To address these challenges, we propose a novel video super-resolution method called Multi-Scale Spatiotemporal Feature Fusion for Video Super-Resolution. This approach leverages artificial intelligence techniques to extract rich spatiotemporal features using a multi-scale deformable convolutional pyramid with parallel branches. It integrates contextual residual learning and optical flow estimation to capture spatial details and compensate for missing temporal information. A hybrid spatiotemporal attention mechanism adaptively fuses key features across frames through dynamic weighting, suppressing background interference and enhancing target-region representation. The framework includes an explicit motion compensation module for precise frame alignment, followed by upsampling to generate high-resolution frames with improved perceptual quality. A multi-task joint loss function optimizes structural fidelity, motion consistency, and visual realism. Experimental results on the University of Alberta Detection and Tracking dataset and Video for Super-Resolution benchmark show the method achieves a peak signal-to-noise ratio of 30.62 dB and a structural similarity index of 0.92, surpassing state-of-the-art methods by up to 4.28 dB and improving perceptual metrics such as Learned Perceptual Image Patch Similarity and Flip Loss in Perception. Ablation studies confirm each module's contribution. The model runs efficiently at 24.8 ms per frame, demonstrating potential for real-time applications. This approach offers a robust and efficient solution for high-resolution geospatial video reconstruction.}, archive = {J_EAAI}, author = {mengqin Yang and yanhui Wang and yang Yang and chunhua Peng}, doi = {10.1016/j.engappai.2025.112327}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112327}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale spatiotemporal feature fusion for super-resolution video reconstruction in dynamic scenes}, volume = {161}, year = {2025}, }
• (2025). Explainable framework based on integrated sensor fusion and deep learning to estimate stability in high-speed micromilling of thin-walled titanium alloy. EAAI, 161, 112326. (www)

  @article{ , abstract = {A novel deep learning (DL) framework based on multi-sensor fusion based on vibration and audio signals with transfer learning has been developed and validated for chatter onset prediction in micromilling of thin-walled titanium (TC4) alloys. The DL framework has been trained and validated with individual sensor data, labelled using machined surface micrographs and cutting tool condition. A total of 879,840 spectrograms has been generated for training via signal segmentation and augmentation methods that yielded a validation accuracy of 89 % during training phase. The trained DL model has been tested on the unlabelled fused signal data to make the DL model adapt to the unseen patterns in the signals recorded during real-time machining. The proposed DL framework based on fused signals outperformed individual sensor-based detection methods by demonstrating an increment of 33 % in accuracy, with a mean absolute error of 0.027416 and R 2 of 0.987. Ablation studies confirm that fusing signals enhances the DL model's performance, yielding a 21 % and 29 % improvement in precision and recall, respectively. The optimal fusion strategy reduced chatter misclassification, with one instance of misclassification. The proposed framework presents a robust, explainable DL approach that can be readily used for adaptive control of cutting conditions in different manufacturing processes. The industry readiness of the DL model has been demonstrated by testing the pretrained model on turning process. The DL model shows 95.31 % prediction accuracy with only three misclassifications out of sixty-four test conditions.}, archive = {J_EAAI}, author = {Sethurao Gururaja and Kundan K. Singh}, doi = {10.1016/j.engappai.2025.112326}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112326}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable framework based on integrated sensor fusion and deep learning to estimate stability in high-speed micromilling of thin-walled titanium alloy}, volume = {161}, year = {2025}, }
• (2025). Liquid metal microfluidic cooling system for high-efficiency thermal management via learning-based genetic algorithm. EAAI, 161, 112324. (www)

  @article{ , abstract = {High heat flux density is a critical factor that limits the performance and reliability of miniaturized, high-power microelectronic systems. This study proposes a liquid metal (LM)-based microfluidic cooling system optimized through a data-driven computational framework based on an enhanced Genetic Algorithm (LC-GA), aiming to deliver an efficient thermal management solution for high-density integrated systems. By integrating LM near-junction cooling with microchannel heat dissipation in a silicon substrate, we developed a heterogeneous three-dimensional interconnect cooling architecture capable of optimizing thermal performance through algorithm-guided parameter tuning. To validate the proposed method, four distinct microchannel configurations were designed, fabricated, and experimentally tested. LM was introduced into the channels to conduct both experimental cooling tests and thermal performance simulations on a simulated heat source. The results demonstrate that this LM-based microfluidic cooling system, optimized through computational parameter determination, can effectively dissipate heat from chips with power consumption up to 800 W while maintaining stable thermal performance. Additionally, a response surface methodology combined with enhanced LC-GA was utilized for multi-factor sensitivity analysis and multi-objective optimization, enabling automatic determination of optimal design and operating parameters to balance thermal resistance and pressure drop. The optimized configuration reduced the maximum chip temperature to approximately 357.54 K, lowered the system pressure requirement, and improved the Performance Evaluation Criterion (PEC) to 2.327. This work provides a data-driven optimization approach that supports the development of high-performance integrated microsystems through algorithm-assisted thermal design.}, archive = {J_EAAI}, author = {Yucheng Wang and Antong Bi and Kaiyu Chen and Shenxin Yu and Wanping Gao and Wenyi Zhang and Yuwan Wu and Zhiqiang Li and Shaoxi Wang}, doi = {10.1016/j.engappai.2025.112324}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112324}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Liquid metal microfluidic cooling system for high-efficiency thermal management via learning-based genetic algorithm}, volume = {161}, year = {2025}, }
• (2025). HoloFluoNet: Live cell imaging intelligence based on fused holography and fluorescence for virtual staining, cell segmentation, classification, and viability analysis. EAAI, 161, 112323. (www)

  @article{ , abstract = {Fluorescence microscopy enables detailed visualization of subcellular structures but is limited by phototoxicity, photobleaching, and labor-intensive labeling. In contrast, digital holographic microscopy (DHM) offers label-free, quantitative phase imaging but lacks biochemical specificity. To integrate the strengths of both modalities, we propose HoloFluoNet, a deep learning-based framework that generates virtual fluorescence information from phase images acquired by DHM. Using a dual-mode imaging system, we simultaneously captured phase and fluorescence images of live cancer cells. The fluorescence data were used to construct biologically grounded supervision signals, including nuclei masks and multi-class cell viability labels. From a single-phase image, HoloFluoNet predicts a virtual nuclei mask, a distance map for boundary refinement, and a viability mask. The architecture incorporates multi-scale feature extractor, and attention mechanisms, optimized with novel inclusion and exclusion loss functions. Post-processing using the watershed algorithm ensures accurate segmentation of overlapping cells. The final registered image provides label-free virtual staining for nuclei localization, viability assessment, and class-specific morphological profiling. HoloFluoNet achieved a Dice score of 0.8685 and an Aggregated Jaccard Index (AJI) of 0.7883, outperforming conventional deep learning models such as U-Net, Fully Convolutional Network (FCN), Pyramid Scene Parsing Network (PSPNet), and DeepLab v3+. These improvements were statistically validated using one-way analysis of variance (p < 0.01). Ablation experiments confirmed the complementary roles of architectural modules and novel loss functions, while robustness tests under noisy and low-quality conditions revealed high stability to low contrast and moderate noise. With an inference speed of 0.123 s per image, the model enables real-time cellular analysis. By bridging structural and molecular imaging, HoloFluoNet provides an efficient, label-free alternative to fluorescence microscopy, with promising applications in artificial intelligence (AI)-assisted drug screening, cancer research, and live-cell monitoring.}, archive = {J_EAAI}, author = {Seonghwan Park and Jaeseong Lee and Jaewoo Park and Inkyu Moon}, doi = {10.1016/j.engappai.2025.112323}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112323}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {HoloFluoNet: Live cell imaging intelligence based on fused holography and fluorescence for virtual staining, cell segmentation, classification, and viability analysis}, volume = {161}, year = {2025}, }
• (2025). Method for ensuring quasi-maximum stability of a system with interval deviations by a robust controller parametric synthesis. EAAI, 161, 112320. (www)

  @article{ , abstract = {It is known that ensuring reliable stability of automatic control systems with interval and stochastic parameter deviations is critical for aviation and other critical areas. Therefore, this article proposes a method for synthesizing Proportional-Integral-Derivative controller parameters based on the coefficient approach to stability analysis (through the characteristic polynomial neighboring coefficients ratios) with an extension to interval uncertainties and the stability quasi-maximum degree integration into the multi-criteria optimization Non-Dominated Sorting Genetic Algorithm II. Scientific contribution and novelty lie in combining analytical sufficient stability conditions (including verification by Kharitonov polynomials and dominant polynomial analysis) with multi-objective optimization taking into account parameter errors and energy costs, which allows obtaining stable and economical Proportional-Integral-Derivative settings for discrete two-channel systems. The results are confirmed by numerical experiments on a helicopter turboshaft engine two-channel control system model with real on-board data. It is shown that the optimal parameter vector provides overshoot of ≈1.43 % and transient process time of ≈1.12 s and maintains stability with local changes in coefficients (±3 … 7 %) and in Monte-Carlo tests. At the same time, a comparative analysis with traditional and neural network controllers showed the proposed approach's advantage in key metrics.}, archive = {J_EAAI}, author = {Denys Baranovskyi and Serhii Vladov and Valerii Sokurenko and Oleksandr Muzychuk and Victoria Vysotska and Maryna Bulakh}, doi = {10.1016/j.engappai.2025.112320}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112320}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Method for ensuring quasi-maximum stability of a system with interval deviations by a robust controller parametric synthesis}, volume = {161}, year = {2025}, }
• (2025). A robust deep learning model selection with data augmentation for automatic detection of tessellated fundus images and explainable artificial intelligence based interpretation. EAAI, 161, 112316. (www)

  @article{ , abstract = {A robust deep learning system for automatically classifying retinal fundus images into two classes—normal and tessellated—is presented in this study. Visual Geometry Group – 16 is used as the base model, taking advantage of transfer learning to develop an efficient framework for fundus image classification. The approach uses nine different model architectures and makes use of a dataset of 352 fundus images that was increased to 4865 samples using sophisticated data augmentation techniques. Of these, Model_8 performed the best, achieving a loss of 0.129 % and an impressive accuracy of 99.39 %. The suggested approach ensures higher performance and dependability by combining rigorous data augmentation, efficient preprocessing, and model fine-tuning techniques. Furthermore, Explainable Artificial Intelligence was used to improve the interpretability of the model and visualize important aspects such as features or imposed pathologies in fundus images more clearly. The study offers promising support to ophthalmologists by offering precise automated diagnoses for the early identification and treatment of retinal disorders.}, archive = {J_EAAI}, author = {Kachi Anvesh and Shanmugasundaram Hariharan and Bharati M. Reshmi and Qiang Xu and Joan Lu and Vinay Kukreja and Murugaperumal Krishnamoorthy}, doi = {10.1016/j.engappai.2025.112316}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112316}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust deep learning model selection with data augmentation for automatic detection of tessellated fundus images and explainable artificial intelligence based interpretation}, volume = {161}, year = {2025}, }
• (2025). Camera-invariance correlation learning and inter-domain-specific distinct representation for person re-identification. EAAI, 161, 112314. (www)

  @article{ , abstract = {Unsupervised domain adaptation (UDA) in person re-identification (ReID) experiences performance degradation due to two primary factors: (1) the information changes across cameras among intra-domain identity samples within the source and target domains; and (2) the inter-domain gap when applied to an unexplored image domain, which makes it a challenging task. The majority of current methodologies predominantly address the inter-domain gap while overlooking the impact of the intra-domain issue. To overcome this limitation, this paper proposes an effective invariance representation ReID (IR-ReID) network that encompasses camera-invariance correlation learning (CICL) and Transformer-based, inter-domain-specific discriminative representation (IDSR), supported by curriculum learning, with the aim of concurrently addressing both of these pivotal issues. Specifically, a CICL module is designed to explore intra-domain distinct features of an individual captured by multiple cameras across the source and target domains. Secondly, an IDSR module is proposed to address the inter-domain gap issue. Thirdly, a curriculum learning-driven strategy is introduced for sample training, systematically guiding our network in learning samples progressively, ranging from easy to difficult instances. Meanwhile, a cluster-based constraint loss is presented to further optimize the clustering results. Finally, extensive experiments are conducted on three well-known datasets, which demonstrate the effectiveness and superiority of our proposed approach in comparison to the prevailing state-of-the-art methodologies.}, archive = {J_EAAI}, author = {Shangdong Zhu and Yunzhou Zhang and Peng Duan}, doi = {10.1016/j.engappai.2025.112314}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112314}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Camera-invariance correlation learning and inter-domain-specific distinct representation for person re-identification}, volume = {161}, year = {2025}, }
• (2025). Applications of machine vision technology for conveyor belt deviation detection: A review and roadmap. EAAI, 161, 112312. (www)

  @article{ , abstract = {Conveyor belt deviation is a frequent challenge in product transportation filed, and failure to promptly detect and rectify this anomaly not only significantly reduces transport efficiency but also poses a risk of serious safety accidents, leading to enormous economic losses. Traditional contact-based deviation detection technologies, with their inherent limitations of high costs and complicated maintenance, have struggled to meet the practical demands of long-distance conveyor belt inspection. In this context, non-contact machine vision technology has emerged as a prominent solution in the field of conveyor belt deviation detection, thanks to its notable advantages of a simple hardware structure and round-the-clock operational capability. Recently, with the rapid development of artificial intelligence theories, this research field has accumulated a series of effective solutions that have been proven through machine vision practical applications. This paper delves into the technical principles of the existing solutions, systematically summarizes them and objectively evaluates their strengths and weaknesses in practical applications. Based on this foundation, this paper also provides an insight on the future development trends of intelligent monitoring for conveyor belt deviation, aiming to offer valuable reference and guidance to technicians in related fields.}, archive = {J_EAAI}, author = {Jiaming Han and Ting Fang and Wensheng Liu and Chenxiao Zhang and Molin Zhu and Jibin Xu and Jie Ji and Xianhua He and Zhang Wang and Min Tang and Chong Dong and Long Ma and Xinlong Yang}, doi = {10.1016/j.engappai.2025.112312}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112312}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Applications of machine vision technology for conveyor belt deviation detection: A review and roadmap}, volume = {161}, year = {2025}, }
• (2025). Synthetic aperture radar target recognition with limited training data based on frequency-domain-assisted dual-stream attention hierarchical deformable convolutional networks. EAAI, 161, 112309. (www)

  @article{ , abstract = {In the field of synthetic aperture radar (SAR) automatic target recognition (ATR), the recognition performance of deep learning methods is often constrained by sample scarcity. To address this issue, this paper proposes a frequency-domain-assisted dual-stream attention hierarchical deformable convolutional network (DAHDF-Net). A frequency-domain-assisted dataset construction method is designed to extend the representational diversity of the sample feature space. Then, a parallel input architecture of frequency-space dual-stream features is constructed, and the adaptive dynamic fusion of frequency-domain stream features with spatial-domain stream features is realized by using a channel attention module in the frequency-domain stream. To enhance the expressiveness of the network on the basis of modulated deformation convolution, a hierarchical residual (H-residual) module is designed. Moreover, a triple hybrid loss is proposed to address the issues of high deformation, which makes classification difficult, and the fuzzy decision boundaries between categories. Experiments were conducted on the moving and stationary target acquisition and recognition (MSTAR) dataset and the FUSAR-Ship dataset. Among them, DAHDF-Net achieves recognition accuracies of 99.75 % and 97.48 % on all training sets and 10-way 25-shot under standard operating condition (SOC) of MSTAR, which demonstrated superior recognition accuracy compared to current state-of-the-art methods.}, archive = {J_EAAI}, author = {Shanliang Yuan and Hongyuan Gao and Xiaoyuan Gu and Jingya Ma}, doi = {10.1016/j.engappai.2025.112309}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112309}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Synthetic aperture radar target recognition with limited training data based on frequency-domain-assisted dual-stream attention hierarchical deformable convolutional networks}, volume = {161}, year = {2025}, }
• (2025). Multi-scale distance similarity entropy: A novel complexity measurement for gearbox fault diagnosis. EAAI, 161, 112308. (www)

  @article{ , abstract = {Entropy-based methods have gained increasing attention in fault diagnosis due to their capability in quantifying the intrinsic complexity and dynamic behavior of time series signals. Vibration signals from gearboxes are typically nonlinear, nonstationary, and transient, often spanning multiple time scales, thus requiring robust multi-scale analysis for accurate condition assessment. However, existing multi-scale entropy methods, prone to missing local variations and vulnerable to noise, often yield incomplete features and low diagnostic accuracy with high computational costs. To overcome these challenges, this paper proposes a novel complexity measurement method termed multi-scale distance similarity entropy, which integrates distance similarity entropy with a multi-scale coarse-graining process. By computing element-wise distances and mapping them through a Gaussian kernel function, multi-scale distance similarity entropy captures subtle nonlinear variations in vibration signals while effectively suppressing noise interference. It estimates the similarity distribution between adjacent vectors, enabling sensitive tracking of dynamic complexity changes in mechanical vibration data. Experimental validation on two gearbox datasets encompassing both single and compound faults demonstrates that multi-scale distance similarity entropy achieves superior classification accuracies of 98.63 % and 97.14 %, respectively. The method also shows strong robustness under random impulse noise and low signal-to-noise ratio conditions, along with high computational efficiency, particularly in small-sample scenarios. These characteristics make it highly suitable for online real-time fault diagnosis in complex and noise contaminated industrial environments involving diverse and compound fault patterns.}, archive = {J_EAAI}, author = {Tao Wang and Shin Yee Khoo and Zhi Chao Ong and Pei Yi Siow and Teng Wang}, doi = {10.1016/j.engappai.2025.112308}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112308}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale distance similarity entropy: A novel complexity measurement for gearbox fault diagnosis}, volume = {161}, year = {2025}, }
• (2025). JSSM-DTA: Joint sequence-structure modeling with multi-scale transformers for explainable drug-target affinity prediction. EAAI, 161, 112303. (www)

  @article{ , abstract = {Drug-Target Affinity (DTA) prediction is a pivotal task in drug discovery, aiming to quantify the strength of interactions between drug molecules and their biological targets, such as proteins. Despite significant advancements in deep learning-based methods, several challenges persist. Scale mismatches obstruct functional group-specific representations, leading to suboptimal feature extraction and loss of critical interaction patterns. Additionally, the inability to fully exploit molecular graph structures weakens hierarchical dependency modeling, while ineffective feature fusion hampers the integration of heterogeneous molecular representations, restricting the capture of intricate drug-target interaction dynamics. To address these challenges, this work introduces JSSM-DTA, a novel Joint Sequence-Structure Modeling framework that integrates Multi-Scale Transformers with intermodal feature attention to enhance DTA prediction. JSSM-DTA formulates a unified representation space through the Adaptive Convolutional Transformer (ACT) and the Multi-Scale Diffusion Transformer (MSDT), where ACT refines sequential feature extraction by capturing intricate local dependencies and long-range contextual relationships, while MSDT preserves hierarchical learning, optimizing both global topological organization and fine-grained interactions. Additionally, the Factorized Inter-layer Interaction Module constructs joint embeddings by seamlessly integrating heterogeneous representations, enabling robust feature fusion. The proposed model was evaluated on Davis, KIBA, Metz and BindingDB datasets, demonstrating superior predictive accuracy over state-of-the-art methods. Explainability analysis reveals key molecular substructures and binding interactions, enhancing transparency in decision-making and establishing JSSM-DTA as a benchmark for both accuracy and interpretability in DTA prediction.}, archive = {J_EAAI}, author = {Uma E. and Mala T.}, doi = {10.1016/j.engappai.2025.112303}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112303}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {JSSM-DTA: Joint sequence-structure modeling with multi-scale transformers for explainable drug-target affinity prediction}, volume = {161}, year = {2025}, }
• (2025). Cyclic translations between pathomics and genomics improve automatic cancer diagnosis from whole slide images. EAAI, 161, 112302. (www)

  @article{ , abstract = {Incorporating genomic characterization into histopathological image modeling brings substantial value to enhancing diagnostic accuracy and supporting the development of targeted and effective treatment strategies. However, prevailing multi-modal integration methods often assume the availability of both pathomics and genomics data in both training and testing phases, overlooking the challenge of data absence due to prohibitive costs. In this paper, we propose a multi-modal cyclic feature generation network (MCFGN) that facilitates cyclic translations between pathomics and genomics to acquire a unified representation of multi-modal data. This approach enables the use of pathological images alone as input to generate joint representations during the testing phase. First, we utilize a general-purpose, self-supervised vision encoder to embed histological image patches as distinctive visual tokens. Next, we hierarchically aggregate patch-level tokens to region-level and slide-level, generating improved whole slide image (WSI) representations. We build self-supervised Masked Autoencoders (MAE) to initialize the hierarchical aggregator. Finally, to incorporate genomic characterization into the learning process, we develop a novel cross-modal cyclic feature generation module to create an intermediate joint representation of pathological and genetic features for patient diagnosis. Evaluations have been conducted on two public datasets from The Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA) and Non-Small Cell Lung Cancer (TCGA-NSCLC) for various diagnostic tasks, including cancer subtyping and biomarker status prediction. Experiments indicate that our MCFGN model improves predictive performance in cancer diagnosis using histological slides, yielding an 8.7% improvement in area under the curve (AUC) for the cancer subtyping task and a 14.1% gain for biomarker prediction.}, archive = {J_EAAI}, author = {Xinyu Hao and Hongming Xu and Xiaofeng Wang and Tong Wang and Timo Hamalainen and Fengyu Cong}, doi = {10.1016/j.engappai.2025.112302}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112302}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cyclic translations between pathomics and genomics improve automatic cancer diagnosis from whole slide images}, volume = {161}, year = {2025}, }
• (2025). A multimodal industrial anomaly detection method based on mask training and teacher–student joint memory. EAAI, 161, 112299. (www)

  @article{ , abstract = {In recent years, the teacher–student framework has been applied to both single-modality detection and multimodal detection, which realizes anomaly detection based on the feature difference between the teacher model and the student model. However, current multimodal teacher–student models use the same teacher model to extract two-dimensional (2D) image and three-dimensional (3D) point cloud features. The point cloud features extracted by the teacher model pre-trained on images are not the optimal feature representation. To further improve the performance of the teacher–student framework on the multimodal anomaly detection task, this paper proposes M ultimodal T eacher- S tudent J oint M emory ( MTSJM ). MTSJM constructs a teacher–student joint memory bank for each modality, the feature distance between the test sample and the memory bank is used as the anomaly indicator. This distance reflects the feature differences between the test sample and the normal sample at multiple levels, including the teacher–teacher, teacher–student, and student–student levels. Then, this paper proposes a mask-based student model training method. While ensuring that the student learns the feature of normal regions, mask training increases the feature difference of non-normal regions between the student and the teacher. On the MVTec 3D Anomaly Detection (MVTec 3D-AD) dataset, the proposed MTSJM achieves effective anomaly detection performance, reaching 95.7% mean Image-level Area Under the Receiver Operator Curve (I-AUROC) and 97.2% mean Area Under the Per-Region Overlap (AUPRO). In addition, MTSJM achieves 99.3% I-AUROC and 99.6% Pixel-level AUROC (P-AUROC) on a real-world vehicle stamping part task, which further illustrates the applicability of MTSJM on the multimodal anomaly detection task.}, archive = {J_EAAI}, author = {Yi Liu and Changsheng Zhang and Xingjun Dong and Yufei Yang}, doi = {10.1016/j.engappai.2025.112299}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112299}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multimodal industrial anomaly detection method based on mask training and teacher–student joint memory}, volume = {161}, year = {2025}, }
• (2025). Fuse MetaFormer with convolutional neural networks for three-dimensional model classification. EAAI, 161, 112298. (www)

  @article{ , abstract = {Multimedia technology is widely applied to artificial intelligence and it is key to the performance of artificial intelligence systems, such as computer-aided design system, virtual reality system, the augmented reality system, medical image processing system, and game development system. With the development of multimedia technology, the number of three-dimensional (3D) models in network or database is becoming larger and larger. It is important to classify 3D models. In order to improve accuracy of three-dimensional model classification, a method of 3D model classification fusing MetaFormer with CNN (Convolutional Neural Networks) is proposed. 3D model is projected into two-dimensional (2D) views through the fixed-view projection, and representative views are selected by the clustering algorithm. Points sampled randomly from representative view are used to calculate its shape descriptors. View feature is extracted from representative view by MetaFormer. Shape feature is extracted from shape descriptors of representative view by Convolutional Neural Networks. View feature and shape feature of representative view are fused. At the same time, majority voting algorithm is used to determine category of 3D model based on the fusion of view features and shape features. Experiments are conducted on ModelNet10 dataset. Experimental results show that the proposed method achieves better results than others.}, archive = {J_EAAI}, author = {Xueyao Gao and Haomin Wu and Chunxiang Zhang and Yongzeng Xue}, doi = {10.1016/j.engappai.2025.112298}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112298}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuse MetaFormer with convolutional neural networks for three-dimensional model classification}, volume = {161}, year = {2025}, }
• (2025). Intelligent design of broadband metasurface based on spectrum prediction neural network and transition simulated annealing algorithm. EAAI, 161, 112297. (www)

  @article{ , abstract = {Recently, metasurfaces have demonstrated considerable potential for application across various fields, including wireless communication, Internet of Things, holographic imaging, and radar systems. With the rapid development of artificial intelligence and its application in other disciplines, metasurface design based on artificial intelligence techniques, particularly deep learning, has attracted widespread attention. In this paper, we propose a novel scheme that integrates deep neural networks with an enhanced simulated annealing algorithm to achieve intelligent design of broadband metasurfaces. Firstly, we construct and train a neural network capable of accurately predicting the amplitude and phase responses of an isotropic metasurface structure, achieving a Mean Absolute Error of 1.36° on the test set. Subsequently, we develop a novel simulated annealing algorithm that is integrated with the prediction neural network to form an intelligent optimization framework. This approach can effectively search for the solution space to accurately achieve arbitrary target phase response, with a very low Mean Squared Error calculated on the test set. Furthermore, we validate the effectiveness of the proposed model and algorithm through a practical application that demonstrates broadband dispersion characteristics within the frequency range of 9–11 Gigahertz (GHz). Additionally, the prediction model can employ transfer learning strategy to adapt to various operating frequency bands, significantly enhancing its generalization capabilities. The proposed intelligent design methodology introduces a novel pathway for rapidly designing metasurface in complex application scenarios, serving as a valuable reference for the engineering design of devices in electronics, optics, and wireless communication.}, archive = {J_EAAI}, author = {Yin Zhang and Yuxin Dai and Yiyu Fan and Jun Yu}, doi = {10.1016/j.engappai.2025.112297}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112297}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent design of broadband metasurface based on spectrum prediction neural network and transition simulated annealing algorithm}, volume = {161}, year = {2025}, }
• (2025). Industrial fabric defect-generative adversarial network (IFD-GAN): High-fidelity fabric cross-scale defect samples synthesis method for enhancing automated recognition performance. EAAI, 161, 112296. (www)

  @article{ , abstract = {—High-quality defect samples and datasets are the cornerstone for enhancing the performance of deep learning detection algorithms for fabric defects. Constraints, such as the complexity of fabric textures, the diversity and sporadic occurrence of defects, the insufficient scale, and the significant class imbalance of datasets, impede improvements in detection accuracy and generalization of detection models, thereby limiting industrial application. To address the issues above, a two-stage generative model for fabric defect images called Industrial Fabric Defect-Generative Adversarial Network (IFD-GAN), based on contrastive learning mutual information mechanism, was introduced. The Pixel-level Defect-Background Stripper (DBS) was devised for precise localization, and the Longitude-Latitude Self-Attention Mechanism (LLSA) was proposed to efficiently focus on defect foreground details, facilitating decoupling and efficient extraction of defect features. The combination of the two, along with the structural similarity loss function, collectively regulates the coordination and consistency between defect and background textures. An industrial-grade fabric defect dataset was collected for IFD-GAN training and generative testing. The model's effectiveness in generating defect images was evaluated based on dimensions such as fidelity and diversity. IFD-GAN is used to enhance this dataset, and the system compared the performance of advanced object detection models trained on the datasets before and after the enhancement. Extensive experimental results show that IFD-GAN can accomplish high-fidelity generation of high-resolution cross-scale fabric defect images in an unsupervised environment, significantly contributing to the creation of more balanced and diverse large-scale industrial fabric defect datasets and the enhancement of defect detection networks in precision, robustness, and generalizability.}, archive = {J_EAAI}, author = {Shun Xu and Sheng Cheng and Shuyang Jin and Xudong Hu and Weitao Wu and Zhong Xiang}, doi = {10.1016/j.engappai.2025.112296}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112296}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Industrial fabric defect-generative adversarial network (IFD-GAN): High-fidelity fabric cross-scale defect samples synthesis method for enhancing automated recognition performance}, volume = {161}, year = {2025}, }
• (2025). Sample-efficient planning-control framework for autonomous underwater vehicle docking using lightweight cross Q-learning. EAAI, 161, 112293. (www)

  @article{ , abstract = {Autonomous Underwater Vehicles (AUVs) are essential for long-duration missions in underwater sensor networks, but their operation is constrained by limited onboard energy and low-bandwidth communication. Docking stations are deployed to support energy replenishment and data transfer, requiring AUVs to autonomously plan return trajectories and achieve precise docking. Traditional motion planning and control methods for AUV docking often decouple global motion planning from local control or rely on accurate dynamics models, while existing deep reinforcement learning (DRL) approaches frequently neglect sample efficiency. This work presents a novel DRL framework that tightly integrates motion planning and control while emphasizing sample efficiency to enable reliable and efficient AUV docking. The docking task is formulated as an Markov Decision Process (MDP), in which a structured reward function is defined to guide the agent towards goal-directed behavior in an interpretable and sample-efficient manner. A lightweight Cross Q-Learning (CrossQ-Lite) algorithm is developed by removing target networks and incorporating batch normalization, achieving stable learning with approximately 92.8% fewer parameters while maintaining comparable performance. Simulation results demonstrate that, under the proposed MDP formulation, CrossQ-Lite achieves a minimum of 1 . 39 × higher sample efficiency and improved training stability compared to state-of-the-art DRL baselines. The trained model reliably completes AUV docking tasks, providing a practical and high-performance solution for autonomous underwater sensor networks.}, archive = {J_EAAI}, author = {Kaixin Zhang and Yuelong Xu and Minghao Zhao and Yu Jiang}, doi = {10.1016/j.engappai.2025.112293}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112293}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sample-efficient planning-control framework for autonomous underwater vehicle docking using lightweight cross Q-learning}, volume = {161}, year = {2025}, }
• (2025). Fault diagnosis of wind turbine based on dual-channel feature aggregation network with attentional mechanism. EAAI, 161, 112291. (www)

  @article{ , abstract = {Wind turbines are subjected to alternating stresses and shock loads during operation, and the collected vibration signals are nonlinear, non-stationary and contain noise, resulting in subtle fault features that are difficult to extract. In order to extract discriminative features from vibration signals under variable speed operating conditions, a dual-channel feature aggregation network (DCNet) with attention mechanism is developed in this paper. First, a Parallel Patch-Aware Convolutional Module (PPCM) is constructed to extract feature information of different scales and levels from Time–Frequency Representations (TFR). Specifically, convolutional operations are performed in both the spatial and frequency domains, endowing it with local–global capturing capabilities and efficiency. Then, to improve the network operation rate, Haar Wavelet Downsampling (HWD) is embedded into the DCNet architecture. The core idea is to reduce the spatial resolution of features through Haar wavelet transform while preserving as many discriminative features as possible. Additionally, Channel Prior Convolutional Attention (CPCA) is introduced to enable DCNet to focus on more critical feature parts by dynamically allocating channel and spatial attention weights, thereby suppressing redundant feature interference. Finally, in the Deep Feature Fusion Module (DFFM), a cross-attention mechanism is adopted for global interaction, fusing features from different channels to achieve feature enhancement. To obtain good diagnostic results under variable speed conditions, we apply a label smoothing algorithm to assist model training. The experimental results show that the proposed model has high diagnostic accuracy, can generalize effectively, and it still maintains high diagnostic accuracy under noise and variable speed working conditions.}, archive = {J_EAAI}, author = {Haiyu Guo and Xingzheng Guo and Xiaoguang Zhang and Fanfan Lu and Chuang Liang}, doi = {10.1016/j.engappai.2025.112291}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112291}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis of wind turbine based on dual-channel feature aggregation network with attentional mechanism}, volume = {161}, year = {2025}, }
• (2025). A new deep learning framework for intelligent aerial monitoring of power transmission line insulators. EAAI, 161, 112290. (www)

  @article{ , abstract = {Transmission line insulators are critical components for maintaining the integrity and efficiency of power transmission lines. Monitoring the health, performance, and efficiency of this network has always been of interest since the past. Traditional and field inspection methods are often tedious, costly, and prone to human error. In order to monitor the condition of transmission lines and overcome traditional challenges, this study proposes an intelligent monitoring framework using unmanned aerial vehicle (UAV) imaging and advanced deep learning models to identify and classify insulator defects. The dataset used in this study consists of 870 original images, 600 healthy images, 140 damaged images, and 130 images of flashover insulators collected from UAV flights and online sources. To increase the diversity of the training data, a data augmentation technique was used, increasing the number of images to 1952 samples and dividing the data into 80 % training sets and 20 % test sets. While the current evaluation is performed on this dataset, real-world experiments will be considered in future studies to further validate the model. In this approach, three scenarios were considered to evaluate and detect these errors. The first scenario used an independent convolutional neural network (CNN) for feature extraction and classification, achieving an accuracy of 82.54 %. The second scenario combined CNN with transfer learning (TL) techniques to improve feature representation, achieving an accuracy of 95.99 %. Scenario Three integrated a hybrid CNN model with a long short-term memory network, achieving an average accuracy of 99.67 % in fault detection and classification. The results demonstrate the superiority and robustness of the proposed hybrid model, providing a reliable and cost-effective solution for transmission line insulator monitoring. This study highlights the significant potential of combining UAV imagery with deep learning algorithms, reassuring the audience of the effectiveness of the proposed solution.}, archive = {J_EAAI}, author = {Reza Rahimi Nejadbougar and Ebadat Ghanbari Parmehr and Alireza Afary and Samira Mavaddati}, doi = {10.1016/j.engappai.2025.112290}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112290}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new deep learning framework for intelligent aerial monitoring of power transmission line insulators}, volume = {161}, year = {2025}, }
• (2025). Spatial-temporal generative network based on deep long short-term memory autoencoder for hand skeleton data sequences reconstruction and recognition. EAAI, 161, 112289. (www)

  @article{ , abstract = {Convolutional neural networks attract the highest research focus in the developing field of Hand Gesture Recognition (HGR). Nevertheless, these approaches presented a challenging task in adapting to time-series data. In skeleton-based HGR, extracting spatial–temporal information remains a challenge. In recent times, recurrent neural networks have exhibited exceptional performance in detecting desired hand gestures by processing of varied length time-series data. Although they outperform traditional methods when huge training data is accessible, their effectiveness significantly diminishes when data availability is constrained. In this study, we introduce an unsupervised data augmentation network known as the Spatial-Temporal Generative Network (STGN), which reconstructs both the spatial and temporal information of the input sequences by leveraging a Deep Long Short-Term Memory Auto-Encoder (DLSTM-AE) network. Consequently, the DLSTM-AE combined with different Long Short-Term Memory (LSTM) network variations, forming an integrated network that can be trained end-to-end for HGR. Through experimentation conducted on the LeapGestureDB dataset (Leap Motion-based Gesture Dataset) and RIT dataset (Rochester Institute of Technology Hand Gesture Dataset), we prove that data reconstruction using STGN had a prominent effect on improving the accuracy of recognizing time-series based hand gestures. For all experiments, the best recognition results are achieved in the augmented dataset. Accuracies were improved on all tested LSTM networks from 2 to 10%. For reproducible research, the code is available at: https://github.com/AMEURsafa/STGN .}, archive = {J_EAAI}, author = {Safa Ameur and Mohamed Ali Mahjoub and Anouar Ben Khalifa}, doi = {10.1016/j.engappai.2025.112289}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112289}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatial-temporal generative network based on deep long short-term memory autoencoder for hand skeleton data sequences reconstruction and recognition}, volume = {161}, year = {2025}, }
• (2025). Neural dynamic fluid reconstruction technique for four-dimensional imaging of combustion flame based on deep learning. EAAI, 161, 112288. (www)

  @article{ , abstract = {Three-dimensional optical diagnostic techniques based on the principles of tomographic imaging enable the acquisition of rich three-dimensional information in experimental flow fields through reconstruction calculations. However, for tasks involving the reconstruction of three-dimensional flow fields at high temporal resolutions, existing methods incur high computational costs, low reconstruction efficiency, and struggle to achieve high spatiotemporal resolution measurements. This paper proposes the Neural Dynamic Fluid Reconstruction Technique (NDFRT) based on deep learning. NDFRT incorporates the time dimension into the reconstruction scope to achieve the four-dimensional reconstruction of dynamic flow fields using neural networks. NDFRT has the following technical advantages: (1) ultra-high spatiotemporal reconstruction resolution; (2) good computational efficiency, with the reconstruction parameter scale only half that of traditional voxel-based methods; (3) the ability to perform three-dimensional frame prediction of dynamic fluids. We validated the proposed method using numerical simulation and experimental jet flame reconstruction and compared it with the traditional algebraic reconstruction technique (ART). Experimental results demonstrate that NDFRT outperforms traditional ART methods in terms of computational efficiency, reconstruction resolution, and reconstruction accuracy.}, archive = {J_EAAI}, author = {Fuhao Zhang and Zhiyin Ma and Can Gao and Gang Xun and Qingchun Lei and Xuesong Li}, doi = {10.1016/j.engappai.2025.112288}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112288}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural dynamic fluid reconstruction technique for four-dimensional imaging of combustion flame based on deep learning}, volume = {161}, year = {2025}, }
• (2025). BeltDiff: Diffusion-based self-labeled generation system for conveyor belt damage detection. EAAI, 161, 112287. (www)

  @article{ , abstract = {The damages of conveyor belt are abnormal occurrence of industrial production line, generally resulting in serious economic losses. While utilizing deep-learning-based methods in belt damage detection task, models training is challenging due to the prevalent scarcity of both numbers and diversity in dataset, leading to the restricted performance and limited accuracy in practical applications. To address this issue, we propose a novel self-labeled generation system based on diffusion model, for creating effective and large-scale conveyor belt detection datasets. Specifically, the training of our system is divided into three stages for progressive task decomposition and interpretable user control. Among the generation process, we focus on the textual prompt alignment and shape guidance to generate more reasonable and high-quality damaged belt images. Meanwhile, larger scale datasets are created via our method on two scenes, including base scene with limited samples, and brand new scenes with no damaged samples, which compose 9636 and 7500 images respectively. By evaluating the quality of damaged images, and the relations between generated and real images, our method demonstrates its effectiveness on generating damaged belt image pairs with annotations. Further, for verifying the validity of our generative datasets, we implement numerous experiments of 4 types of popular detection models training on different settings. The results demonstrate that our datasets support effective accuracy improvement, comparing with the base datasets and the classic augmented datasets, specifically increased by up to 14.4% in the brand new scenes with no damaged samples.}, archive = {J_EAAI}, author = {Peixian Zhuang and Yuanxiu Cai and Xi Liu and Xianchao Zheng and Fuheng Xiao and Jiangyun Li}, doi = {10.1016/j.engappai.2025.112287}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112287}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {BeltDiff: Diffusion-based self-labeled generation system for conveyor belt damage detection}, volume = {161}, year = {2025}, }
• (2025). A robust deep learning based model for denoising phonocardiogram signals in clinical environments. EAAI, 161, 112286. (www)

  @article{ , abstract = {A cardiac auscultation, being a very popular first line cardiovascular diseases screening method, typically produces phonocardiogram signals contaminated by background sounds, e.g., speech and movement of other patients and doctors/nurses, phone ringing and door knocking/opening/closing, and internal physiological body noises, e.g., lung sounds, muscle contraction and motion artifacts, as it is usually done in a noisy environment and it is not possible to suppress all the internal physiological body noises. So, a quality of the recording is finally decreased. Moreover, this can strongly affect a subsequent analysis/classification of the captured recording and can also lead to its incorrect classification. In this paper, we propose a novel deep convolutional architecture that integrates the bidirectional long short-term memory layers with the transformer blocks involving multi-head attention mechanism, which is particularly designed for denoising phonocardiogram signals. The performance of the proposed artificial intelligence-based approach is evaluated through synthetic data generated by using one public phonocardiogram dataset and two public real-world hospital ambient/respiratory sound databases as well as by deploying two public and renowned real-world clinical environment datasets containing phonocardiogram signals contaminated by various environmental and physiological noises recorded during the clinical auscultations. The deployed complex evaluation strategy assures a higher robustness and generalizability of the proposed approach. As also a result of that, the proposed model outperforms the current state-of-the-art approaches and provides the significant improvements in all the assessed signal quality and noise suppression metrics and over all the involved benchmark datasets. Finally, it should be noted here that the model achieved an average estimated SNR of 14.984 dB and 11.924 dB for the PASCAL and CirCor22 dataset respectively, which represent the renowned real-world clinical environment datasets.}, archive = {J_EAAI}, author = {Maros Jakubec and Eva Lieskovska and Peter Pocta}, doi = {10.1016/j.engappai.2025.112286}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112286}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust deep learning based model for denoising phonocardiogram signals in clinical environments}, volume = {161}, year = {2025}, }
• (2025). Transformer-based end-to-end point restoration network for filtering temporarily-static objects from point cloud. EAAI, 161, 112284. (www)

  @article{ , abstract = {The precision of map-based localization systems is pivotal for advanced navigation and autonomous vehicle technologies. This study addresses the challenge of dynamic environments where temporarily-static objects, such as vehicles, introduce occlusion and temporal alterations to the mapping landscape, leading to inaccuracies in mapping and localization. We propose a transformer-based deep learning model that removes temporarily-static objects and refines occluded areas in raw point cloud data, thereby enhancing both mapping and localization. Our method leverages a transformer-based point restoration network trained on a novel dataset that does not require human-driven manual annotation and predicts point clouds free of temporarily-static objects. The approach was validated through experiments demonstrating improved qualitative and quantitative results. The results display the importance of our method by removing temporarily-static objects and refining occluded areas. The proposed method achieves a 26.4 % reduction in Chamfer Distance L1 (CD-L1), 33.1 % reduction in Chamfer Distance L2 (CD-L2), and a 30.2 % relative improvement in the harmonic mean of precision and recall (F-Score) over the baseline network, demonstrating significantly enhanced model for filtering. This research contributes to artificial intelligence (AI)-enabled autonomous navigation by offering a systematic solution to a common problem in dynamic environments, setting a new filtering technology for mapping systems.}, archive = {J_EAAI}, author = {Sabir Hossain and Xianke Lin}, doi = {10.1016/j.engappai.2025.112284}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112284}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer-based end-to-end point restoration network for filtering temporarily-static objects from point cloud}, volume = {161}, year = {2025}, }
• (2025). Cross-view gait recognition based on discriminative global–local feature representation and learning. EAAI, 161, 112282. (www)

  @article{ , abstract = {Performance of gait recognition can be heavily influenced by deformation of walking silhouettes due to variations of view angle. Cross-view gait recognition is still one of the most challenging problems in pattern recognition community. In this paper, we propose a new cross-view gait recognition algorithm by discriminative global–local feature representation and learning. First, we design cross-view convolutional neural network for learning global gait feature representations underlying binary walking silhouettes sequences. Second, we propose to extract local gait features by partial spatial–temporal sequence and local feature extractors, which can represent the motion characteristics and temporal changes of different human body parts. Both global and local features in our proposed method are extracted from different gait representations to provide spatial–temporal information from diverse aspects, and they are fused on feature level to further reduce the sensitivity to variations of view angle and improve the performance of gait recognition. Finally, we introduce deep metric learning framework and propose a robust, effective, and view-related loss function, named view triplet loss function, to learn discriminative gait features. Compared with triplet loss, the proposed loss function explores hierarchical relationship of the variations of view angles, which achieves better intra-subject compactness and inter-subject separability. Experimental results on CASIA-B, OULP and OUMVLP gait datasets demonstrate the feasibility and advantage of our proposed algorithm.}, archive = {J_EAAI}, author = {Muqing Deng and Yi Zou and Zhi Zeng and Xiaoreng Feng and Yanjiao Wang and Yuan Liu}, doi = {10.1016/j.engappai.2025.112282}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112282}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-view gait recognition based on discriminative global–local feature representation and learning}, volume = {161}, year = {2025}, }
• (2025). A novel clustering-ensemble learning model for day-ahead photovoltaic power forecasting. EAAI, 161, 112281. (www)

  @article{ , abstract = {Accurate day-ahead photovoltaic power forecasting (PPF) is essential for grid scheduling and transaction planning. However, time series prediction models based on historical meteorological and photovoltaic power data often struggle to capture long-term dependencies. Existing research typically employs multivariate regression models to establish mapping relationships between numerical weather prediction (NWP) and photovoltaic power data, yet still exhibits deficiencies in weather pattern recognition and consideration of temporal patterns. Therefore, this paper proposes a novel clustering-ensemble learning model to enhance predictive performance. In the proposed deep time-series clustering algorithm, key features from NWP data are extracted for clustering, addressing the inefficiencies of traditional clustering methods in handling high-dimensional data. A weighted Warp-Euclidean distance is proposed to capture sequence trend homogeneity and spatial similarity, overcoming the limitations of Euclidean distance in time-series data. The ensemble learning model combines the advantages of two gradient boosting tree models to handle nonlinear features and high-dimensional data, while introducing lag features to enhance temporal dynamic learning capability. This approach enhances the stability and mitigates the limitations associated with relying primarily on weather-related features. This study utilizes NWP and measured photovoltaic data from two stations in Hebei Province, China. Results demonstrate superior performance compared to other clustering algorithms and prediction models in day-ahead PPF tasks, achieving R 2 scores of 0.952 and 0.943 on two public photovoltaic datasets, representing improvements of 0.020 and 0.017 over optimal benchmark models. The proposed framework provides theoretical guidance for the stable operation of grid-connected photovoltaic systems.}, archive = {J_EAAI}, author = {Feifei Yang and Xueqian Fu and Zhengshuo Li and Dawei Qiu and Hamed Badihi}, doi = {10.1016/j.engappai.2025.112281}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112281}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel clustering-ensemble learning model for day-ahead photovoltaic power forecasting}, volume = {161}, year = {2025}, }
• (2025). Artificial intelligence-based phonocardiogram signal classification using segment-specific multi-domain features for cardiovascular and arterial disease. EAAI, 161, 112280. (www)

  @article{ , abstract = {Heart diseases are the leading cause of death globally, and early diagnosis coupled with timely treatment can save lives. The human heart produces sounds that reflect the functions of its valves and chambers, with murmurs often indicating abnormal heart behavior. The phonocardiogram (PCG) analysis offers a quick, easy, and affordable diagnostic method for detecting such conditions. The PhysioNet/2016 dataset, containing annotated heart sound recordings, has advanced automated heart sound classification research. Despite the dataset encompassing nine classes of heart diseases, it has primarily been used for binary classification tasks. Multi-class classification is challenging due to overlapping acoustic features among certain heart conditions and signal variability. To address these issues, we implemented segment specific feature extraction across six domains — time, frequency, amplitude, wavelet, spectrum, signal processing — and evaluated various machine learning classifiers. Our findings revealed that the Random Forest classifier outperformed others, leading us to integrate it into our model. On the testing dataset, it attained 94% accuracy, 93% sensitivity and 93% specificity for binary classification Our model achieved 89% accuracy, 80% sensitivity and 98% specificity for nine-class classification including both valve diseases and artery-related conditions.}, archive = {J_EAAI}, author = {Syeda Sana Bukhari and Shahab U. Ansari and Khurram Khan Jadoon and Raja Hashim Ali}, doi = {10.1016/j.engappai.2025.112280}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112280}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence-based phonocardiogram signal classification using segment-specific multi-domain features for cardiovascular and arterial disease}, volume = {161}, year = {2025}, }
• (2025). Probabilistic forecasting of non-ferrous metal prices based on outlier treatment algorithms, quantile regression based deep learning and two-phase multi-objective optimization. EAAI, 161, 112277. (www)

  @article{ , abstract = {Accurate forecasting of non-ferrous metal prices plays a crucial role in helping enterprises optimize production, control costs, and manage risks, as well as in enabling countries and industries to formulate resource strategies, ensure economic security, and promote sustainable development. However, existing research still has limitations in outlier detection, uncertainty analysis, and model optimization, making it difficult to meet the forecasting needs in complex market environments. To fill these gaps, this study proposes a novel combined probabilistic forecasting framework for non-ferrous metal prices. First, this study utilizes outlier treatment algorithms to achieve robust outlier detection and correction. Second, multiple quantile regression-based deep learning models are established for probabilistic forecasting. Subsequently, a two-phase multi-objective optimization strategy is proposed. Specifically, the first phase is designed to optimize the upper quantile half-interval and lower quantile half-interval of each quantile regression-based deep learning model, while the second phase is used to optimize the combined weights of the aforementioned single model to synergistically enhance interval reliability and prediction accuracy. Finally, silver and aluminum futures prices are used as illustrative case studies. The experimental results show that the proposed model outperforms the benchmark method in terms of interval coverage and prediction accuracy. This study provides a scientific basis and reference for industrial production optimization, investment decision-making and resource management planning.}, archive = {J_EAAI}, author = {Pei Du and Fangrui Gui and Lijie Shan and Qianyi Xing and Jianzhou Wang}, doi = {10.1016/j.engappai.2025.112277}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112277}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Probabilistic forecasting of non-ferrous metal prices based on outlier treatment algorithms, quantile regression based deep learning and two-phase multi-objective optimization}, volume = {161}, year = {2025}, }
• (2025). Identifying nonlinear roll damping and restoring parameters via physics-informed neural network. EAAI, 161, 112275. (www)

  @article{ , abstract = {To effectively characterize a ship’s roll motion, developing a robust dynamic model that incorporates accurate damping and restoring parameters is essential. Given that only limited free-decay measurements have been measured, an efficient physics-informed neural network-based approach has been developed to simultaneously determine the nonlinear damping and restoring parameters associated with ship rolling. Rather than relying on a large dataset to train a neural network for parameter identification, the proposed method incorporates the physical equations of roll motion into the residual networks, ensuring that the identified parameters adhere to their physical interpretation. Three numerical cases, encompassing computational fluid dynamics (CFD) analyses, and laboratory experiments, are presented to show the superior performance of the developed method. Key findings indicate that the proposed method effectively identifies the investigated parameters with high precision while maintaining consistent performance across various initial roll angle conditions. Compared to traditional methods, the proposed approach offers significant advantages, including reduced reliance on large datasets, automated parameter identification without manual tuning, and the ability to incorporate physical constraints directly into the learning process. These features make the method particularly suitable for real-world applications where data is sparse or costly to obtain.}, archive = {J_EAAI}, author = {Shuai Cong and Jinwei Sun and Qianying Cao and Changhong Zhi and Shixuan Liu}, doi = {10.1016/j.engappai.2025.112275}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112275}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identifying nonlinear roll damping and restoring parameters via physics-informed neural network}, volume = {161}, year = {2025}, }
• (2025). Anomaly detection in offshore open radio access network using long short-term memory models on a novel artificial intelligence-driven cloud-native data platform. EAAI, 161, 112274. (www)

  @article{ , abstract = {The Radio Access Network (RAN) is a critical component of modern telecommunications infrastructure, currently evolving towards disaggregated and open architectures. These advancements are pivotal for integrating intelligent, data-driven applications aimed at enhancing network reliability and operational autonomy through the introduction of cognitive capabilities, as exemplified by the emerging Open Radio Access Network (O-RAN) standards. Despite its potential, the nascent nature of O-RAN technology presents challenges, primarily due to the absence of mature operational standards. This complicates the management of data and intelligent applications, particularly when integrating with traditional network management and operational support systems. Divergent vendor-specific design approaches further hinder migration and limit solution reusability. These challenges are compounded by a skills gap in telecommunications business-oriented engineering, which remains a key barrier to effective O-RAN deployment and intelligent application development. To address these challenges, Boldyn Networks developed a novel cloud-native data analytics platform, specifically designed to support scalable Artificial Intelligence (AI) integration within O-RAN deployments. This platform underwent rigorous testing in real-world scenarios, and applied advanced AI techniques to improve operational efficiency and customer experience. Implementation involved adopting Development Operations (DevOps) practices, leveraging data lakehouse architectures tailored for AI applications, and employing sophisticated data engineering strategies. The platform successfully addresses connectivity challenges inherent in real-world offshore wind farm deployments using Long Short-Term Memory (LSTM) models for anomaly detection in network connectivity. After integrating the LSTM models into the network control, more than 90 percent of connectivity issues were reduced in runtime. This marks a step toward autonomous, self-organizing, and self-healing networks.}, archive = {J_EAAI}, author = {Abdelrahim Ahmad and Peizheng Li and Robert Piechocki and Rui Inacio}, doi = {10.1016/j.engappai.2025.112274}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112274}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Anomaly detection in offshore open radio access network using long short-term memory models on a novel artificial intelligence-driven cloud-native data platform}, volume = {161}, year = {2025}, }
• (2025). Structural design through reinforcement learning. EAAI, 161, 112273. (www)

  @article{ , abstract = {This paper introduces the Structural Optimization gym (SOgym), a novel open-source Reinforcement Learning (RL) environment designed to advance machine learning in Topology Optimization (TO). SOgym enables RL agents to generate physically viable and structurally robust designs by integrating the physics of TO into the reward function. To enhance scalability, SOgym leverages feature-mapping methods as a mesh-independent interface between the environment and the agent, allowing efficient interaction with the design variables regardless of mesh resolution. Baseline results use a model-free Proximal Policy Optimization agent and a model-based DreamerV3 agent. Three observation space configurations were tested. The Topology Optimization game ( TopOpt game) inspired configuration, an interactive educational tool for designing structures to minimize compliance under volume constraints, performed best in terms of performance and sample efficiency. The 100 million parameter DreamerV3 model (DreamerV3-100M) produced structures with compliance values approximately 54 % higher than those from traditional optimization methods and a 0 % disconnection rate, an improvement over supervised learning approaches that often struggle with disconnected load paths. When comparing the learning rates of the agents to those of engineering students from the TopOpt game experiment, the DreamerV3-100M model shows a learning rate approximately four orders of magnitude lower, an impressive feat for a policy trained from scratch through trial and error. These results suggest RL's potential to solve continuous TO problems and its capacity to explore and learn from diverse design solutions. SOgym provides a platform for developing RL agents for structural design challenges and is publicly available to support research in the field.}, archive = {J_EAAI}, author = {Thomas Rochefort-Beaudoin and Aurelian Vadean and Niels Aage and Sofiane Achiche}, doi = {10.1016/j.engappai.2025.112273}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112273}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Structural design through reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). Single-channel selection for epileptic seizure identification through a custom machine learning model. EAAI, 161, 112272. (www)

  @article{ , abstract = {Epilepsy affects millions of people worldwide, triggering undesirable motor and sensory events that drastically impair the quality of life of those affected. Detecting this condition through a monitoring system can be extremely useful in identifying epileptic seizures without relying exclusively on a specialist. Our goal is to identify these seizures using personalized supervised machine learning models – eXtreme Gradient Boosting – for each patient, employing only one electroencephalogram input channel. Our methodology explores the relationship between the selected input channel and the topographic location of epileptic seizures modeled for the patient. The results revealed notable accuracy performances for the three patients investigated: 100%, 99%, and 88%, after applying the proposed time filter that checks for the presence/absence of a seizure every 3 s. Furthermore, it was possible to observe that the results are consistent with the affected area in each patient, demonstrating the effectiveness of the method in selecting the single channel. This approach demonstrates the feasibility of detecting convulsive seizures using only one input channel and underscores the need to extend the methodology to more patients.}, archive = {J_EAAI}, author = {Jusciaane Chacon Vieira and Ignacio Sanchez-Gendriz and Luiz Affonso Guedes}, doi = {10.1016/j.engappai.2025.112272}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112272}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Single-channel selection for epileptic seizure identification through a custom machine learning model}, volume = {161}, year = {2025}, }
• (2025). IMobileTransformer: A fusion-based lightweight model for rice disease identification. EAAI, 161, 112271. (www)

  @article{ , abstract = {Rice blast, sheath blight, leaf scald, bacterial leaf blight, and brown spot severely threaten rice yield. To address the limitations of current deep learning methods in rice disease recognition, particularly their insufficient integration of local and global features, this study proposes an Improved MobileTransformer (IMobileTransformer) model. The proposed architecture synergistically combines MobileNet’s strengths in local feature extraction and lightweight architecture with Transformer’s superior capability in global information processing. Specifically, the model is designed with three functional branches: a) a MobileNet branch utilizing inverted residual structure with depthwise separable convolution layers to reduce parameters and computational complexity, b) a Transformer branch modified from Swin-Transformer architecture, where the Multilayer Perceptron (MLP) layer is enhanced by splitting input feature channels through an Inception-based structure to maintain global feature extraction efficiency while minimizing computational overhead, and c) a feature fusion branch that concatenates reshaped outputs from both branches through channel-wise stacking, enabling effective integration of local and global representations. Experimental results show that compared to classical models such as MobileNetV3-Large, EfficientNet-B0, Vision Transformer Base/16 (ViT-B/16), Shifted Window Transformer (Swin-Transformer), Tiny Vision Transformer (TinyViT), Mobile Vision Transformer (MobileViT), LocalViT-S, IMobileTransformer achieves a recognition accuracy of 99.62% for rice diseases, with improvements of 1.71%, 0.91%, 38.09%, 4.17%, 1.99%, 1.5% and 0.42%, respectively, providing an effective solution for rice disease recognition.}, archive = {J_EAAI}, author = {Yang Lu and Haoyang Zhou and Peng Wang and Erzhi Wang and Gongfa Li and Tongjian Yu}, doi = {10.1016/j.engappai.2025.112271}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112271}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {IMobileTransformer: A fusion-based lightweight model for rice disease identification}, volume = {161}, year = {2025}, }
• (2025). Personalized and safe medication recommendation based on convolutional neural network and transformer architecture. EAAI, 161, 112267. (www)

  @article{ , abstract = {With the accumulation of electronic health records (EHRs), artificial intelligence (AI) based medical services such as medication recommendation (MR) has aroused widespread concern. However, existing drug recommendation models suffer from inadequate patient representation and adverse drug–drug interactions (DDIs). To address these challenges, we propose an AI-based personalized and safe medication recommendation method based on convolutional neural network and transformer architecture (CT-PASMR). Specifically, convolutional neural network (CNN) and transformer are combined in parallel (CAT) to model local relationships in a patient’s single visit and long-term dependencies in sequential EHR data, respectively. Subsequently, graph attention networks (GATs) are deployed to capture drug co-occurrences and adverse DDIs with various weights, generating safe drug representations. Moreover, a joint loss function is introduced to balance accuracy and safety in CT-PASMR. Finally, the experimental results on MIMIC (Medical Information Mart for Intensive Care)-III and MIMIC-IV datasets demonstrate that CT-PASMR achieves competitive performance on seven evaluation metrics such as DDI rate, Jaccard index and F1 score, compared with nine state-of-the-art (SOTA) baseline models. Ultimately, ablation studies and further analysis confirm the efficacy of CATs and GATs in providing personalized and safe medication recommendations. Code, qualitative results, and trained weights will be available at the link: https://github.com/gefengru/CT-PASMR .}, archive = {J_EAAI}, author = {Fengru Ge and Xiaomei Yu and Xue Li and Xingxu Fan and Yanjie Zhao}, doi = {10.1016/j.engappai.2025.112267}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112267}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Personalized and safe medication recommendation based on convolutional neural network and transformer architecture}, volume = {161}, year = {2025}, }
• (2025). Contrastive learning with similarity enhancement for dimensionality reduction. EAAI, 161, 112266. (www)

  @article{ , abstract = {Dimensionality reduction aims to reduce the number of dimensions while preserving crucial information. As a self-supervised learning approach, contrastive learning provides a novel perspective for dimensionality reduction. However, most contrastive learning methods focus on optimizing similarity, which have limitations in reducing feature redundancy. Moreover, the dependence on negative pairs introduces computational overhead. To address these problems, we propose Contrastive Learning with Similarity Enhancement for Dimensionality Reduction (CLSDR), which integrates neighborhood embedding into the contrastive learning framework. Specifically, CLSDR uses k -nearest neighbors sampling to construct positive pairs. We design a multi-level loss function that captures the diversity of data while maintaining the consistency of local and global features. In addition, we propose a nonlinear similarity optimization mechanism based on logarithmic smoothing, which adjusts the gradient of similarity loss, improving the stability during the training process. Experimental results demonstrate that CLSDR significantly outperforms several state-of-the-art methods. Especially, on the Street View House Numbers dataset, CLSDR achieves 66.7% and 62.3% Top-1 accuracy with two classifiers in 64-dimensional embedding space, which have 10.6% and 38.4% improvement over the best competing approach. Thus, CLSDR exhibits strong robustness and scalability across different dimensions.}, archive = {J_EAAI}, author = {Qi Yang and Changpeng Wang and Linlin Feng and Lizhen Ji and Jiangshe Zhang}, doi = {10.1016/j.engappai.2025.112266}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112266}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Contrastive learning with similarity enhancement for dimensionality reduction}, volume = {161}, year = {2025}, }
• (2025). A multi-modal approach for detecting drivers’ distraction using bio-signal and vision sensor fusion in driver monitoring systems. EAAI, 161, 112265. (www)

  @article{ , abstract = {According to a report by the World Health Organization (WHO), approximately 1.3 million people lose their lives annually owing to traffic accidents. The majority of road traffic accidents stem from driver negligence. Recently, there has been a growing interest in utilizing deep learning and machine learning technologies to enhance the safety and efficiency of road traffic, with the aim of addressing issues arising from driver inattentiveness. Most studies focus on detecting abnormal driver behavior using driving sensors or driver images; however, they often overlook physiological factors such as the driver's bio-signals. Considering that the driver's state, including fatigue, stress, and concentration, can significantly affect driving safety, it is crucial to build models that consider biometric information. Therefore, this study proposes a multi-modal transformer model called Bio-Vision Transformer (BiViT) that comprehensively considers both driver bio-signals and images. The BiViT model uses a vision transformer to extract features from driver images and employs a time-series transformer to capture features from the driver's bio-signals. In addition, the interactions between the extracted features are modeled, and the joint fusion method is employed as the feature-fusion approach. To validate the proposed model, performance comparisons and analyses were conducted using commonly used models in image analysis. The experimental results demonstrated that the proposed BiViT model exhibited high performance, with an accuracy of 0.91 and a harmonic mean of precision and recall (F1-score) of 0.91, surpassing the performance of the comparison models.}, archive = {J_EAAI}, author = {Byeongjoon Noh and Myeongseok Park and Yechan Han and Jaeyun Kim}, doi = {10.1016/j.engappai.2025.112265}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112265}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-modal approach for detecting drivers’ distraction using bio-signal and vision sensor fusion in driver monitoring systems}, volume = {161}, year = {2025}, }
• (2025). Graph neural network with generative adversarial training for node classification on class imbalanced data. EAAI, 161, 112264. (www)

  @article{ , abstract = {Node classification in class-imbalanced graph data remains a critical challenge, as traditional graph neural networks (GNNs) either assume class-balanced graph structures or inadequately address class imbalance. This often results in predictive bias, where majority classes are favored while minority classes are underrepresented. To overcome this limitation, this study introduces a novel graph neural network with generative adversarial training (GNN-GAN), where the GNN extracts latent features from input node attributes balanced through data synthesis using a conditional generative adversarial network (GAN) and data fusion strategy. The GNN and GAN are trained synchronously to ensure GAN synthesizes samples that match real data distribution, while GNN adjusts to the quality of synthesized data in a timely manner. A data fusion strategy combines synthetic and real samples to mitigate class imbalance and maintain classification accuracy. Experiments on several benchmark graph datasets demonstrate that the GNN-GAN consistently outperforms state-of-the-art baselines. A comprehensive ablative study further validates the advantages of the synchronized training procedure, offering insights into the model's robustness across graph datasets with varying structures and imbalance ratios.}, archive = {J_EAAI}, author = {Xiaoqi Zhou and Peixin Shi}, doi = {10.1016/j.engappai.2025.112264}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112264}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph neural network with generative adversarial training for node classification on class imbalanced data}, volume = {161}, year = {2025}, }
• (2025). Diffusion-based feature absorption approach for improving lightweight object detectors in adverse weather conditions. EAAI, 161, 112263. (www)

  @article{ , abstract = {Although cutting-edge lightweight detectors have achieved desirable performance in favorable weather conditions, they again failed to accurately identify objects within low-quality scenes captured in inclement environments, especially during rainy nighttime. These detectors face difficulties learning beneficial information for detection because objects are obscured by rain and low-light conditions. To address the aforementioned challenges, we introduce an innovative and effective Diffusion-based Feature Absorption Learning (Diff-FAL) approach, to reinforce the performance of lightweight detection models in nighttime with rain interference. By developing an unsupervised training strategy based on the diffusion model, the proposed approach assists lightweight detectors absorb useful features from degraded images. To this end, our Diff-FAL framework comprises three distinct subnetworks: a Feature Optimization (FO) subnetwork, a Feature Mutation (FM) subnetwork, and a Lightweight Detection (LD) subnetwork. The FO subnetwork is designed to produce sharp, detailed features from rainy night images and the FM subnetwork is developed to transfer those features to the LD subnetwork. Comprehensive experiments on various datasets confirmed the superiority of our model, outperforming the compared method by up to 28.01% and 30.16% in accuracy on two published datasets: the rainy nighttime (RNT) and real rainy images (rRain) datasets, respectively, while maintaining high-speed performance during inference.}, archive = {J_EAAI}, author = {Quoc-Viet Hoang and Trung-Hieu Le}, doi = {10.1016/j.engappai.2025.112263}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112263}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Diffusion-based feature absorption approach for improving lightweight object detectors in adverse weather conditions}, volume = {161}, year = {2025}, }
• (2025). A multi-modal model for removal of crack image shadow based on language prior information. EAAI, 161, 112262. (www)

  @article{ , abstract = {Computer-vision based crack detection highly depends on the quality of images and it remains to be a challenging task due to the effects of shadow. The utilization of deep learning algorithm, guided by prior information, has been demonstrated effective in the removal of shadows. However, it has limitations in handling images with shadowed cracks. A deep learning method incorporating language prior information and a multi-modal model is proposed in this study to improve the removal of shadow effects in restoring the crack images. Natural language processing (NLP) is used to extract the language prior information from image text description to enhance the crack recognition, and OTSU method is used to obtain the image prior information. A novel multi-modal model, named as the Text-based Shadow Removal Network (TSRNet), for the removal of shadow in crack image, is proposed to have better crack restoration capability. Bayesian optimization approach is also employed to optimize the network architecture improving the prediction precision of TSRNet. The proposed framework for shadow removal is verified using an open-source shadowed concrete crack image dataset and a new dataset from experiment. Results indicate that the language prior information can enhance the TSRNet model with better performances of shadow removal for real-world crack images compared with other existing models.}, archive = {J_EAAI}, author = {Gang Liu and Xuming Li and Qingshan Yang and S.S. Law and Changjun Deng}, doi = {10.1016/j.engappai.2025.112262}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112262}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-modal model for removal of crack image shadow based on language prior information}, volume = {161}, year = {2025}, }
• (2025). Development and implementation of a hybrid visual prediction algorithm for robotic smart tomato harvesting. EAAI, 161, 112261. (www)

  @article{ , abstract = {This paper introduces new algorithms designed specifically for smart tomato harvesting, which combines predictions from two independently trained YOLOv8 (You Only Look Once version 8) models, each specialized on different datasets to detect tomatoes and classify their ripeness stages—ripe, unripe, and semi-ripe—while also computing their center points. To enhance detection accuracy under varying field conditions, multiple fusion strategies were developed, including a hybrid algorithm that integrates union and confidence-weighted summation methods. The Hybrid Algorithm achieved the best performance, surpassing the original models and other fusion techniques. It attained F1 scores of 0.697 and 0.694 at confidence thresholds of 0.5 and 0.9, respectively, compared to the original models' F1 scores of 0.670 and 0.648 at confidence 0.5, and 0.481 and 0.497 at confidence 0.9. This corresponds to an improvement of 4.0 % and 7.6 % at confidence 0.5, and 44.3 % and 39.6 % at confidence 0.9, demonstrating the hybrid algorithm's stability and superiority, particularly at higher thresholds. Furthermore, the system utilizes a high-resolution RGB (Red, Green, Blue) camera for real-time image capture, enhancing model performance in complex agricultural environments. This study validates the effectiveness of confidence-based fusion in developing robust and accurate vision systems for precision agriculture.}, archive = {J_EAAI}, author = {Giuseppe Carbone and Angad Singh Gurtatta and Dmitry Malyshev}, doi = {10.1016/j.engappai.2025.112261}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112261}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Development and implementation of a hybrid visual prediction algorithm for robotic smart tomato harvesting}, volume = {161}, year = {2025}, }
• (2025). Multi-modal multi-level feature representation learning for flow pattern identification of oil-water two-phase flow. EAAI, 161, 112260. (www)

  @article{ , abstract = {To tackle the difficulty of the traditional experimental methods in real-time monitoring and identification of the flow process, this paper introduces a novel flow pattern identification method of the vertical oil-water two-phase flow based on multi-modal multi-level feature representation. The one-dimensional electromagnetic signals are encoded into two-dimensional feature spaces to explore their structural complexity, evolutionary probability laws and nonlinear characteristics in multimodal domain, thereby generating a multi-modal representation of the electromagnetic signals. Subsequently, a multi-modal multi-level feature fusion network is developed for flow pattern identification network, which flexibly leverages effective information across different modalities and levels, thereby enhancing the identifying accuracy. Experimental results demonstrate that the proposed method achieves high accuracy on the constructed multi-modal dataset, proving its feasibility and effectiveness in identifying the flow pattern of the oil-water two-phase flow in vertical pipes.}, archive = {J_EAAI}, author = {Weihang Kong and Yaohan Chi and He Liu and Hongbao Tang and He Li}, doi = {10.1016/j.engappai.2025.112260}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112260}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modal multi-level feature representation learning for flow pattern identification of oil-water two-phase flow}, volume = {161}, year = {2025}, }
• (2025). Defining and evaluating decision and composite risk in language models applied to natural language inference. EAAI, 161, 112253. (www)

  @article{ , abstract = {Despite their impressive performance, large language models (LLMs) are known to pose important risks. One such set of risks arises from misplaced confidence, whether over-confidence or under-confidence, that the models have in their inference. While the former is well studied, the latter is not, leading to an asymmetry in understanding the comprehensive risk of the model based on misplaced confidence. In this paper, we address this asymmetry by defining two types of risk (decision and composite risk), and proposing an experimental framework consisting of a two-level inference architecture and appropriate metrics for measuring such risks in both discriminative and generative LLMs. The first level relies on a decision rule that determines whether the underlying language model should abstain from inference. The second level (which applies if the model does not abstain) is the model’s inference. This framework has direct implications for error-sensitive LLM-based engineering applications where reliable decision-making is critical, such as healthcare and finance. Through detailed experiments on four natural language commonsense reasoning datasets using both an open-source ensemble-based transformer model and a generative LLM, we demonstrate the practical utility of our evaluation framework. Our results show that the framework can get an LLM to confidently respond to an extra 20.1% of low-risk inference tasks that other methods might misclassify as high-risk, and skip 19.8% of high-risk tasks, which would have been answered incorrectly.}, archive = {J_EAAI}, author = {Ke Shen and Mayank Kejriwal}, doi = {10.1016/j.engappai.2025.112253}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112253}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Defining and evaluating decision and composite risk in language models applied to natural language inference}, volume = {161}, year = {2025}, }
• (2025). Advanced machine learning techniques for predicting wear performance in graphene oxide particulate interpenetrating polymer network composites. EAAI, 161, 112252. (www)

  @article{ , abstract = {This research investigates the wear behavior of hybrid polymeric composites made from synthetic glass and natural cotton fibers, reinforced with varying proportions of Graphene Oxide (GO) (0 %, 1 %, 3 %, 5 %, 7 %, 9 %). The effect of fiber arrangement and Graphene Oxide (GO) incorporation on wear rate and Coefficient of Friction (CoF) was evaluated using the Pin-On-Disk method, with analysis based on Taguchi's L 32 Orthogonal Array. The optimal parameters were found at 6 min, 5 % GO, 300 revolutions per minute (rpm) speed, 20 mm (mm) track diameter, and 10 N (N) load, achieving a minimum wear rate of 0.612 × 10 −4 cubic millimeters per newton-meter (mm 3 /N-m) and a CoF of 0.151. Predictive modeling was performed to predict the wear rate and coefficient of friction using supervised machine learning algorithms, including Linear Regression, Decision Tree, and Random Forest, to forecast material behavior. Performance evaluation using Confusion Matrix, Distribution Analysis, and various metrics showed that the Decision Tree model excelled, achieving near-perfect predictive power with a Mean Squared Error (MSE) of 0 and an R-squared value of 0.9999. The model demonstrated 100 % accuracy, with precision, recall, and F1-scores all equal to 1. This research demonstrates the effectiveness of combining natural and synthetic fibers with GO, along with the predictive power of machine learning in optimizing material properties.}, archive = {J_EAAI}, author = {Eastus Russel and S. Madhu and Judy S and Edwin Geo Varuvel and G.B. Santhi and G. Suresh and Femilda Josephin J.S and Mohammed F. Albeshr and Farzad Kiani}, doi = {10.1016/j.engappai.2025.112252}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112252}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advanced machine learning techniques for predicting wear performance in graphene oxide particulate interpenetrating polymer network composites}, volume = {161}, year = {2025}, }
• (2025). Health state evaluation and analysis of equipment considering multi-scale data fusion. EAAI, 161, 112251. (www)

  @article{ , abstract = {Current researches of health state evaluation of equipment have increasingly emphasized the integration of data and knowledge to improve evaluation accuracy and interpretability, which leads to the emergence of hybrid information-based methods as a research hotspot. However, there are issues with redundant health indicators, multi-scale data with different features and measurement units, and interpretability measures in the evaluation. In this paper, an interpretable health state evaluation and analysis method of equipment is proposed based on trustworthy evidential reasoning rule (TERR). To deal with redundancy in health indicators, a multi-scale data analysis model is proposed based on the clustering analysis and Kruskal-Wallis test. It can select the optimal health indicators that effectively preserve the original evaluation information while reducing the model complexity. To integrate the selected indicators with uncertainty, a TERR-based multi-scale data fusion model is proposed, where the evidence weight, reliability, and trustworthiness are simultaneously considered. Also, an interpretable parameter optimization model is constructed to alleviate the uncertainty in initial evidential parameters. To study the interpretability of TERR, two sensitivity analysis methods are proposed, and the performance coefficient matrix of the model output to the perturbation is derived mathematically. This shows how the interpretability of TERR is enhanced and which external parameter has the greatest impact on outputs. Finally, a case study of health state evaluation of laser inertial measurement unit (LIMU) validates the effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Shuai-Wen Tang and Jiang Jiang and Jian-Bin Sun and Zhuo-Ting Yu}, doi = {10.1016/j.engappai.2025.112251}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112251}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Health state evaluation and analysis of equipment considering multi-scale data fusion}, volume = {161}, year = {2025}, }
• (2025). Unmanned surface vehicle autonomous racing and obstacle avoidance with robust adversarial deep reinforcement learning. EAAI, 161, 112250. (www)

  @article{ , abstract = {This paper presents an autonomous racing control method for Unmanned Surface Vehicles (USVs) based on robust adversarial deep reinforcement learning (ADRL) algorithm, which leverages the strengths of both deep reinforcement learning and adversarial training. Adversarial obstacles and various tracks are employed to train a policy, so the proposed method can enhance the robustness and generalization of autonomous USV racing while ensuring effective obstacle avoidance. A simulation environment for USV racing was developed to conduct the experiments with Unity3D. The performance of the proposed method in handling diverse track scenarios and obstacle avoidance is demonstrated through simulations. Quantitative results show that ADRL achieves dramatic improvements over baseline DRL methods: collision rates are reduced by 99.85%, task completion rates are improved by 73.75%, and navigation time efficiency is enhanced by approximately 3% while maintaining superior safety performance.}, archive = {J_EAAI}, author = {Jingyu Liu and Yuanda Wang and Changyin Sun}, doi = {10.1016/j.engappai.2025.112250}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112250}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unmanned surface vehicle autonomous racing and obstacle avoidance with robust adversarial deep reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). Multiscale constitutive modeling of anisotropic plasticity: Coupling the visco-plastic self-consistent model with the recurrent neural network and its implementation in finite element analysis. EAAI, 161, 112249. (www)

  @article{ , abstract = {The anisotropic and nonlinear strain-path-dependent nature of metal plasticity poses a major challenge for accurate constitutive modeling in finite element (FE) analysis. Traditional macroscale models are easily implemented but lack accuracy, while crystal plasticity (CP) models offer high fidelity at the cost of computational efficiency. To bridge this gap, we propose a deep neural network smart constitutive (DNNSC) framework that combines the visco-plastic self-consistent (VPSC) model with a gated recurrent unit (GRU) network. A VPSC model calibrated on pure aluminum generated 14,000 strain-paths for training GRU-based network. The optimized model has a prediction accuracy of up to 96 % on unknown strain-paths. Subsequently, the DNNSC model was implemented into the FE analysis through Fortran programming, and a benchmark simulation for thin sheet stamping was successfully performed. The simulation results demonstrated that the DNNSC model significantly improved prediction performance compared to conventional macroscale constitutive models. Especially, the ear height and plate thickness were accurately predicted with an accuracy of 91.85 % and 95.84 %, compared to only 68.85 % and 86.59 % achieved by the Yld model. Meanwhile, the simulation time was reduced to approximately one-tenth that of the fully coupled CP model, because the latter required calculating and homogenizing the mechanical responses of hundreds of grains at each integration point during the simulation. The DNNSC framework bridges the gap between CP models and FE simulations of plastic forming and breaks down the barrier between modeling and practical application. Furthermore, this framework can be extended to other materials by re-calibrating VPSC parameters and fine-tuning DNN parameters.}, archive = {J_EAAI}, author = {Ziwei Zhou and Liang Cheng and Huaidong Song and Haijing Guo and Ruolin Li and Lingyan Sun and Bin Tang}, doi = {10.1016/j.engappai.2025.112249}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112249}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multiscale constitutive modeling of anisotropic plasticity: Coupling the visco-plastic self-consistent model with the recurrent neural network and its implementation in finite element analysis}, volume = {161}, year = {2025}, }
• (2025). An online reinforcement learning method to improve control adaptability in robot-aided rehabilitation. EAAI, 161, 112248. (www)

  @article{ , abstract = {Rehabilitation robotics enables consistent and personalized therapy but still relies on complex, expert-driven tuning of control parameters. To address this, a reinforcement learning strategy based on Q-learning is proposed to autonomously adapt key parameters during upper-limb rehabilitation, without requiring prior task-specific knowledge. A systematic evaluation is conducted across combinations of control parameters (radial stiffness and execution time), performance-based reward functions (pointing accuracy and movement smoothness), and exploration strategies ( ɛ - greedy and Upper Confidence Bound). The Q-learning agent selects discrete actions (increase, decrease, or maintain the current value) for each control parameter, enabling real-time adaptation based on observed performance. The method is validated using a Kuka robotic arm in experiments involving 16 right-handed healthy subjects (13 males, 3 females) and 8 right-handed individuals simulating impaired motor behavior (5 males, 3 females). Motion signals are acquired through internal robot sensors, while a wearable physiological monitoring system define the Q-learning agent state. Reward improvement and exploration ratio are analyzed as key performance indicators and statistically compared across all tested conditions using the Mann–Whitney test. The results demonstrate that the proposed algorithm effectively adjusts control parameters online, with performance influenced by the reward function, exploration strategy, and selected control actions. Reward improvements of 0 . 11 ± 0 . 09 ( ɛ - greedy , reward based on pointing ability) and 0 . 13 ± 0 . 11 (reward based on smoothness, Upper Confidence Bound strategy) were observed in healthy subjects, indicating enhancements in pointing accuracy and movement smoothness. In simulated pathological cases, improvements of 0 . 08 ± 0 . 13 and 0 . 06 ± 0 . 16 were observed, respectively.}, archive = {J_EAAI}, author = {Rita Molle and Christian Tamantini and Clemente Lauretti and Emilio Maria Romano and Loredana Zollo}, doi = {10.1016/j.engappai.2025.112248}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112248}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An online reinforcement learning method to improve control adaptability in robot-aided rehabilitation}, volume = {161}, year = {2025}, }
• (2025). Fast prediction and compensation of curing deformation behaviours of composite parts with complex geometry based on neural operator on riemannian manifolds. EAAI, 161, 112247. (www)

  @article{ , abstract = {Controlling the curing deformation of composite parts is becoming increasingly challenging with the ever-increasing performance requirements of aerospace equipment. Mould surface compensation, which adjusts the mould surface to minimise the discrepancy between the cured part geometry and the nominal part geometry, has become a primary deformation control way in engineering. The existing mirror compensation methods focus on the deformation dominated by spring-in and are challenging for complex deformation modes. Surrogate model-based shape optimisation provides a feasible idea, but establishing a surrogate model to predict curing deformation fields on complex part geometries remains a challenge. Therefore, this study explores a novel neural operator-driven framework for fast curing deformation prediction and compensation. A clustering-based deformation field segmentation method is proposed to manipulate the mould surface morphing using limited design variables. The neural operator on Riemannian manifolds is introduced for the first time to establish the surrogate model between the mould surface and the curing deformation fields on complex part geometries. To control the global error distribution of the composite part, two error metrics are designed to optimise the mould surface by genetic algorithm. The verification results show that the proposed framework exhibits significant potential in predicting and compensating for the curing deformation field of composite parts with complex geometry.}, archive = {J_EAAI}, author = {Lu Chen and Yingguang Li and Jingyan Su and Weiwei Xu and Lin Hu and Gengxiang Chen and Xu Liu and Xiaozhong Hao}, doi = {10.1016/j.engappai.2025.112247}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112247}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast prediction and compensation of curing deformation behaviours of composite parts with complex geometry based on neural operator on riemannian manifolds}, volume = {161}, year = {2025}, }
• (2025). Enhanced detection of acute leukemia: A hybrid machine learning framework with adaptive weight-optimized level set evolution. EAAI, 161, 112244. (www)

  @article{ , abstract = {In this research, a novel, multistage framework is developed for the automated detection of Acute Lymphoblastic Leukemia (ALL) and Acute Myeloid Leukemia (AML), addressing challenges due to the inclusion of overlapping cells, noise, and unwanted cells. It integrates an Adaptive Weight-Optimized Level Set Evolution (AWOLSE) scheme to ensures precise and accurate segmentation, coupled with a marker-controlled watershed algorithm to improve performance in regions of cell overlap and contact. It ensures more reliable cell boundary delineation. For classification, a hybrid model combining the strengths of Random Forest (RF) and Support Vector Machine (SVM) is employed that delivering superior performance by leveraging the complementary advantages of these classifiers. Furthermore, feature extraction with the Gray Level Co-occurrence Matrix (GLCM), followed by feature selection with Principal Component Analysis (PCA), aids in the identification of relevant features. The suggested methodology beats its competitors, providing the best ALL identification results on the Acute Lymphoblastic Leukemia Image Database (ALLIDB), with 99.07% accuracy, 97.96% recall, 100.00% specificity, and 100.00% precision. Similarly, on the American Society of Hematology (ASH) database, the technique achieves superior AML identification results with 96.25% accuracy, 95.00% recall, 97.50% specificity, and 97.44% precision.}, archive = {J_EAAI}, author = {Pradeep Kumar Das and Adyasha Sahu and Sukadev Meher and Rutuparna Panda and Ajith Abraham}, doi = {10.1016/j.engappai.2025.112244}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112244}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced detection of acute leukemia: A hybrid machine learning framework with adaptive weight-optimized level set evolution}, volume = {161}, year = {2025}, }
• (2025). Reinforcement learning-based secure tracking control for nonlinear interconnected systems: An event-triggered solution approach. EAAI, 161, 112243. (www)

  @article{ , abstract = {This paper proposes a secure tracking control method for nonlinear interconnected systems based on reinforcement learning, addressing both security constraints and mismatched conditions in such systems. By utilizing system augmentation techniques, the tracking control problem is reformulated into a stabilization problem for the augmented system, simplifying the original control task. Drawing inspiration from the Chinese philosophical principle, we further transform the problem into a zero-sum game framework. Moreover, a control barrier function (CBF) is incorporated into a cost function to ensure that system trajectories remain within a predefined safe region. An event-triggered mechanism is introduced, and an event-based safety Hamilton–Jacobi–Isaacs (HJI) equation is established. An adaptive single evaluation network is designed, leveraging experience replay techniques to solve the HJI equation. Finally, the Lyapunov method is employed to prove the uniform ultimate boundedness of both the tracking error and the weight error. The effectiveness of the proposed method is validated through numerical examples.}, archive = {J_EAAI}, author = {Chunbin Qin and Suyang Hou and Mingyu Pang and Zhongwei Wang and Dehua Zhang}, doi = {10.1016/j.engappai.2025.112243}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112243}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reinforcement learning-based secure tracking control for nonlinear interconnected systems: An event-triggered solution approach}, volume = {161}, year = {2025}, }
• (2025). A quantized subtraction-convolution network for industrial lightweight edge interpretable diagnosis. EAAI, 161, 112241. (www)

  @article{ , abstract = {In addressing the prevalent challenges of delayed fault diagnosis, deployment complexity of advanced deep learning models on low-cost edge devices, and limited model interpretability, this study proposed an edge-cloud collaborative interpretable diagnosis based on quantized subtraction-convolution network. Firstly, an interpretable quantized subtraction-convolution network with lightweight three-layer structure is designed. Inspired by the adaptive spectral subtraction, a learnable sparse pulse kernel is designed to extract the fault feature as the subtraction layer. Subsequently, the convolution and classification layers are then integrated to produce interpretable results for efficient identification. To facilitate deployment and updates on edge devices, the quantized subtraction-convolution network is decomposed into a lightweight edge architecture and corresponding parameters. It can be deployed on edge devices, and an edge-cloud collaborative framework addresses its training and compression. Considering the sparse characteristics of quantized subtraction-convolution network, a sparse pulse quantization strategy and quantization-aware training technique were developed to compress the model parameters. Finally, a low-cost edge fault diagnosis node prototype with quantized subtraction-convolution network is designed for real-time edge fault diagnosis. Experiments shown that the proposed method achieved average accuracy of 99.88 percent with compression ratio of 9.5. The memory usage, floating-point operations per second, and average power consumption are respectively only 54 kilo binary byte, 0.053 mega binary byte, and 6.67 mJ. Actual gear edge diagnosis experiments confirmed the effectiveness, which can implement model inference in 0.045 s at the edge fault diagnosis node. It is anticipated that the proposed method will find extensive application in the industrial edge interpretable diagnosis.}, archive = {J_EAAI}, author = {Qihang Wu and Jun Luo and Wenbin Huang and Xiaoxi Ding}, doi = {10.1016/j.engappai.2025.112241}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112241}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A quantized subtraction-convolution network for industrial lightweight edge interpretable diagnosis}, volume = {161}, year = {2025}, }
• (2025). Lightweight multi-level feature integration transformer for image super-resolution. EAAI, 161, 112240. (www)

  @article{ , abstract = {Transformer-based methods have attracted significant attention in the field of image super-resolution. However, existing approaches typically concentrate on a single level of information for image reconstruction, and neglect the critical role of multi-level information in feature representation, resulting in the low utilization of the potential capabilities of Transformer. To address this limitation, we propose a novel Transformer model, named as Multi-Level Differential Window Attention Transformer (MLDAT), designed for multi-level information fusion. Specifically, our approach introduces a self-attention module based on differential windows to comprehensively extract and integrate feature information across varying window sizes. Additionally, we introduce a High-order Global Attention Module (HGAB) to combine the second-order attention with global self-attention, which facilitates the establishment of relationships between local features and the overall global feature context within an image while complementing local window information. Extensive experimental results demonstrate that our model can significantly improve the performance of image super-resolution , and achieves the better results compared with existing methods.}, archive = {J_EAAI}, author = {Shuheng Wang and Ziao Gong and Mengda Li and Shen Wu and Yilin He}, doi = {10.1016/j.engappai.2025.112240}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112240}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight multi-level feature integration transformer for image super-resolution}, volume = {161}, year = {2025}, }
• (2025). Three-dimensional object detection for autonomous driving via deep learning: A review. EAAI, 161, 112238. (www)

  @article{ , abstract = {With the rapid advancement of autonomous driving technology, there is an increasing demand for highly accurate and real-time vehicle perception systems. From an artificial intelligence (AI) perspective, three-dimensional (3D) object detection benefits from recent progress in deep neural networks, convolutional neural networks (CNNs), and transformer architectures, which provide powerful tools for feature extraction, spatial reasoning, and multi-modal data fusion. These AI techniques enable robust, uncertainty-aware predictions by effectively modeling complex sensor data. From an engineering standpoint, 3D object detectors serve as critical components in autonomous driving systems by translating AI-derived insights into real-time, high-accuracy vehicle perception. This paper reviews the research progress of deep learning-based 3D object detection algorithms in autonomous driving. First, commonly used data acquisition sensors are systematically categorized, and the most widely adopted 3D detection datasets and evaluation metrics are introduced; the standard 3D bounding-box representation and core network architectures are also explained. Second, algorithms are classified according to input data type: 1) single-modal methods, including vision-based, 3D data dimensionality reduction, point cloud-based, transformer-based, mamba-based, and hybrid point cloud approaches; 2) multi-modal methods, subdivided into serial fusion and parallel fusion strategies within network pipelines. The characteristics, contributions, and limitations of each category are summarized, and representative algorithms are compared across datasets to identify research trends. Finally, current challenges, such as sparse data handling, domain shifts, and real-time constraints, are examined, and prospective directions for future development are proposed.}, archive = {J_EAAI}, author = {Jiaqi Cai and Yiquan Wu}, doi = {10.1016/j.engappai.2025.112238}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112238}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-dimensional object detection for autonomous driving via deep learning: A review}, volume = {161}, year = {2025}, }
• (2025). Unscented kalman filter neural network with double-layer decomposition algorithm applied to the prediction of current efficiency in aluminum electrolysis processes. EAAI, 161, 112237. (www)

  @article{ , abstract = {Predictive modeling in predictive optimization control technology can effectively reduce energy consumption and improve production efficiency in the electrolytic aluminum process (EAP). Among the various modeling approaches, artificial neural networks (ANN) have been widely adopted in the EAP due to their strong capability to capture nonlinear relationships inherent in complex industrial systems. However, conventional ANN heavily rely on historical data to achieve optimal models, which limits their accuracy and generalization performance under strong disturbances and time-varying conditions. To address these problems, this article proposes a novel dynamic prediction model of unscented Kalman filter neural network with double-layer decomposition (UKFNN-DD), building upon the foundation of Kalman filter neural network (KFNN). First, singular value decomposition (SVD) is adopted to compute the square root of covariance matrices, enhancing the numerical robustness of the prediction algorithm and overcoming the shortcomings of traditional Cholesky decomposition. Furthermore, a dual-layer KFNN strategy is introduced to overcome the absence of one-step-ahead prediction in conventional state-space formulations. By applying a two-stage correction to the state variables using measurement data, the proposed method improves the adaptability of the model to external environmental variations. Finally, the prediction error of the state variables is optimized using a gradient descent algorithm, which improves the stability and reliability of the model’s prediction performance. Experimental results demonstrate that the proposed method significantly outperforms baseline methods, achieving a 4.38-fold reduction in mean absolute error (MAE) and a 77.29-fold reduction in the sum of squared errors (SSE), thereby verifying its superiority in dynamic prediction for aluminum electrolysis.}, archive = {J_EAAI}, author = {Xiaoyan Fang and Xihong Fei and Zhenyi Xu and Lei Su and Jing Wang}, doi = {10.1016/j.engappai.2025.112237}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112237}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unscented kalman filter neural network with double-layer decomposition algorithm applied to the prediction of current efficiency in aluminum electrolysis processes}, volume = {161}, year = {2025}, }
• (2025). Feature super-resolution-based method for small-scale target detection and segmentation in side-scan sonar. EAAI, 161, 112235. (www)

  @article{ , abstract = {The application of side-scan sonar in underwater target detection plays a significant role in marine engineering construction and ocean resource exploration. However, existing methods for small object detection often suffer from limited accuracy and robustness. To address this issue, we propose a feature super-resolution-based approach for detecting and segmenting small targets in side-scan sonar images. During network training, a feature super-resolution branch is first constructed. Both low-level and high-level features from the backbone of the You Only Look Once (YOLO) network are fed into this branch. After processing through an encoder-decoder architecture, a super-resolved feature map is reconstructed, and the network is optimized via backpropagation to enhance the ability to extract features of small targets. Furthermore, an exponential decay strategy is adopted to define the loss weights of different branches, establishing a branch-aware training mechanism to improve training effectiveness. During inference, the super-resolution branch is discarded to balance detection accuracy and inference efficiency. Experimental results demonstrate that the proposed method achieves superior performance in detecting and segmenting small-scale targets in side-scan sonar imagery, achieving state-of-the-art results on two public side-scan sonar small object datasets. Additionally, this approach can be extended as a training strategy for side-scan sonar target detection and segmentation networks. The source code is available at https://github.com/Yang-Code984/FSR_Sonar .}, archive = {J_EAAI}, author = {Zhiwei Yang and Jianhu Zhao and Xi Zhao and Chao Huang}, doi = {10.1016/j.engappai.2025.112235}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112235}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature super-resolution-based method for small-scale target detection and segmentation in side-scan sonar}, volume = {161}, year = {2025}, }
• (2025). Data-driven prediction of thermal and flow fields in magnetized boger-micropolar tri-hybrid nanofluids via deep artificial neural networks. EAAI, 161, 112232. (www)

  @article{ , abstract = {This study presents a deep learning framework for predicting the thermal and flow behavior of Boger micropolar tri-hybrid nanofluids under magnetized axial squeezing flow between two parallel disks. The fluid comprises gold, silver, and diamond nanoparticles dispersed in a water-based solution, forming a high-performance ternary hybrid heat transfer medium. A fully connected artificial neural network with three hidden layers (32–16–8 neurons) and hyperbolic tangent activation functions was trained on synthetic data generated using a boundary value problem solver. The model predicts four key physical quantities: axial velocity, streamwise velocity, microrotation, and temperature profile. Quantitative evaluation reveals excellent agreement between machine learning predictions and numerical benchmarks, with mean absolute errors consistently below 0.02 and coefficients of determination exceeding 0.99 across all outputs. Sensitivity analysis reveals the impact of the solvent fraction and vortex viscosity parameters on flow penetration and thermal stratification near the boundaries. This work introduces a novel combination of Boger–micropolar fluid dynamics, magnetohydrodynamic squeezing flow, and tri-hybrid nanoparticles into a unified ANN (Artificial Neural Networks) based surrogate, enabling accurate multi-output prediction of strongly coupled nonlinear fields. The proposed approach offers a scalable machine learning model for real-time optimization of nanofluid-based thermal management systems in engineering applications.}, archive = {J_EAAI}, author = {Mohammad Jalili and Hesam Ehsani and Ali Mirzagoli Ganji and Amirali Shateri and Mehdi mahboobtosi and Yuping Wu and Payam Jalili and Bahram Jalili and Davood Domiri Ganji}, doi = {10.1016/j.engappai.2025.112232}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112232}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven prediction of thermal and flow fields in magnetized boger-micropolar tri-hybrid nanofluids via deep artificial neural networks}, volume = {161}, year = {2025}, }
• (2025). Explainable knowledge recommendation for product innovation concept design based on knowledge graph and multi-task learning. EAAI, 161, 112231. (www)

  @article{ , abstract = {In the process of product innovation concept design, by enhancing the transparency of the knowledge recommendation process and providing designers with explainable recommendation results, decision-making time can be reduced, and design efficiency can be improved. Existing explainable knowledge recommendation methods can generate textual explanation information, but they often overlook the dynamic nature of the design process, which negatively affects the accuracy of the recommendations. To address this issue, the current study proposes a knowledge recommendation method for product innovation concept design based on knowledge graph (KG) and multi-task learning. Specifically, a TransD-based model is first constructed to perform the knowledge graph embedding (KGE) task. Then, KGE and Gated Recurrent Unit (GRU) are used to obtain dynamic knowledge demands across different temporal scales from the historical interactions of designers to enhance recommendation accuracy. A multi-task learning framework is subsequently introduced to enable feature sharing between the two tasks, improving the generalization and effectiveness of the model. Finally, an explanation strategy is designed by combining embedded knowledge and paths to provide optimal explainability information. Experimental results demonstrate that, compared with other state-of-the-art knowledge recommendation algorithms, the proposed method not only offers explainable recommendation results but also achieves higher accuracy, making it more suitable for real-world recommendation scenarios.}, archive = {J_EAAI}, author = {Yida Hong and Wenqiang Li and Hai Xiang and Chuanxiao Li and Changfu Wan}, doi = {10.1016/j.engappai.2025.112231}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112231}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable knowledge recommendation for product innovation concept design based on knowledge graph and multi-task learning}, volume = {161}, year = {2025}, }
• (2025). Towards reproducible machine learning-based process monitoring and quality prediction research for additive manufacturing. EAAI, 161, 112223. (www)

  @article{ , abstract = {Additive manufacturing (AM) is increasingly adopted across industries for its ability to support design flexibility, rapid prototyping, and mass customization. Machine learning (ML)-based cyber-physical systems (CPSs) have been extensively developed to improve the print quality of AM. However, the reproducibility of these systems has not been thoroughly investigated due to a lack of formal evaluation methods. Reproducibility, a critical component of trustworthy artificial intelligence, is achieved when an independent team can replicate the findings or artifacts of a study using a different experimental setup and achieve comparable performance. In many publications, critical information necessary for reproduction is often missing due to a lack of comprehensive AM and ML domain knowledge, resulting in systems that fail to replicate the reported performance. Integrating AM and ML domain knowledge, this paper proposes a reproducibility investigation pipeline and a reproducibility checklist for ML-based AM process monitoring and quality prediction systems. Based on the CRoss Industry Standard Process (CRISP) methodology, the pipeline guides researchers through the key steps required to reproduce a study, while the checklist systematically extracts information relevant to reproducibility from the publication. We validated the proposed approach through two case studies: reproducing a fused filament fabrication warping detection system and a laser powder bed fusion melt pool area prediction model. Both case studies confirmed that the pipeline and checklist successfully identified missing information, improved reproducibility, and enhanced the performance of reproduced systems. Based on the proposed checklist and leveraging large language models, a reproducibility survey was conducted to assess the current reproducibility status within this research domain.}, archive = {J_EAAI}, author = {Jiarui Xie and Mutahar Safdar and Andrei Mircea and Bi Cheng Zhao and Yan Lu and Hyunwoong Ko and Zhuo Yang and Yaoyao Fiona Zhao}, doi = {10.1016/j.engappai.2025.112223}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112223}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards reproducible machine learning-based process monitoring and quality prediction research for additive manufacturing}, volume = {161}, year = {2025}, }
• (2025). Degradation-based predictive energy management for intelligent fuel cell hybrid electric vehicles with a novel deep reinforcement learning architecture. EAAI, 161, 112222. (www)

  @article{ , abstract = {The degradation of fuel cell systems (FCS) affects the energy management performance of fuel cell hybrid electric vehicles (FCHEVs), especially in the case of severe degradation. This study develops a novel predictive energy management architecture, which consists of the Extended Long Short-Term Memory (xLSTM) and Soft Actor-Critic (SAC). Specifically, the speed predictor is built using xLSTM network and leverages the speed and position information of the preceding and following vehicles. In order to provide a reliable basis for energy management in degradation scenarios, an FCS degradation model is developed, which enables the dynamic mapping of output efficiency curves under varying state-of-health (SOH) conditions. Sequentially, a SAC-based agent is employed as the energy management strategy (EMS), which innovatively takes full account of the impact of the FCS SOH on energy allocation, achieving proactive degradation-aware energy allocation. Simulation results demonstrate that the developed xLSTM architecture achieves a 71 % improvement in speed prediction accuracy compared to Transformer-based models within the Next Generation Simulation validated scenario. Moreover, at SOH = 90 %, the newly designed EMS reduces operational costs by 8.7 % and 10.7 % compared to conventional methods under the New European Driving Cycle and Worldwide Harmonized Light Vehicles Test Procedure, while decreasing FCS degradation costs by 17 % and 51 %, respectively. The innovative approach not only elevates energy efficiency but also prolongs FCS operational longevity via intelligent SOH-aware, which establishes a new paradigm for lifecycle-optimized energy management in FCHEVs.}, archive = {J_EAAI}, author = {Zhigen Nie and Zhuangfeng Shi and Yufeng Lian and Hao Song}, doi = {10.1016/j.engappai.2025.112222}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112222}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Degradation-based predictive energy management for intelligent fuel cell hybrid electric vehicles with a novel deep reinforcement learning architecture}, volume = {161}, year = {2025}, }
• (2025). Observer based fault tolerant control design for saturated nonlinear systems with full state constraints via a novel event-triggered mechanism. EAAI, 161, 112221. (www)

  @article{ , abstract = {Safety plays a crucial role in many promising applications as they must comply with stringent safety regulations while staying within physical limits. Therefore, safety fault-tolerant systems with saturated nonlinearities and input constraints are investigated in this paper. An event-triggered safety fault tolerant control (FTC) method based on adaptive dynamic programming (ADP) is proposed. The constrained state is modeled using a smooth mapping function, thus transforming the original system into an unconstrained framework. A fault observer is designed for unknown actuator faults occurring in the system. When an unknown fault occurs in the system, the actuator is compensated in real time according to the fault. The optimal safety value function is approximated by constructing a single critic neural network (NN). Then, a novel event-triggered mechanism is proposed, which allows the algorithm to obtain the optimal control law without constructing the event-triggered Hamilton–Jacobi–Bellman (HJB) equation. In addition, by adjusting the size of the parameter τ under the event triggered condition, different demands on the number of event triggers are realized, which in turn ensures the efficient use of resources while leading to the trade-off of optimal control. This paper also incorporates an empirical playback technique in the design of the critic’s update law to address the challenges associated with continuous incentive requirements. It is theoretically proven that all signals of the system are consistent with the limit bounds, thus ensuring the stability of the system.}, archive = {J_EAAI}, author = {Chunbin Qin and Mingyu Pang and Zhongwei Wang and Suyang Hou and Dehua Zhang}, doi = {10.1016/j.engappai.2025.112221}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112221}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Observer based fault tolerant control design for saturated nonlinear systems with full state constraints via a novel event-triggered mechanism}, volume = {161}, year = {2025}, }
• (2025). Cybersecurity enhancement using conditional generative adversarial network with transformer-based conditional variational autoencoder. EAAI, 161, 112220. (www)

  @article{ , abstract = {Since, Artificial Intelligence is highly developing and concatenating into several domains, cybersecurity is an important field of delivering both the advantages and disadvantages. In addition to this, Artificial Intelligence is applied in a wide variety of applications like healthcare sector, content creation and entertainment and financial industries. Therefore, this work finds the efficiency of Artificial Intelligence -oriented cybersecurity metrics in succeeding the digital environment over elevating cyber threats. Here, the developed models consist of two different stages while implementing the model. In the first stage, the essential dataset is assembled from the benchmark data source. These datasets are assembled by using Generative Artificial Intelligence (Gen Artificial Intelligence networks). Consequently, the raw data is given as an input to Conditional Hybrid Network for cybersecurity enhancement. Further, the Transformer-based Conditional Variational Autoencoders with Spatial-temporal Attention are designed for feature extraction that is subjected to the Conditional Generative Adversarial Network for classifying the cyber attacks. Henceforth, the developed network is evaluated and designed with multiple measures. Comparing baseline models, the suggested network obtains higher performance for developing security over cyber networks.}, archive = {J_EAAI}, author = {Prithvipal Singh and Sandeep Singh and Gurupdesh Singh and Amritpal Singh}, doi = {10.1016/j.engappai.2025.112220}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112220}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cybersecurity enhancement using conditional generative adversarial network with transformer-based conditional variational autoencoder}, volume = {161}, year = {2025}, }
• (2025). A modal transfer enhanced deep learning for structural dynamic response with sparse spatial data. EAAI, 161, 112218. (www)

  @article{ , abstract = {Sparse spatial data in structural health monitoring (SHM) presents significant challenges due to large spatial gaps and high uncertainty, which hinder the generalization ability and performance of traditional data-driven methods that rely on dense sensor networks. To address this issue, this paper proposes a novel modal transfer-enhanced deep learning (MT-DL) model for the reconstruction and prediction of structural dynamic responses. The core idea is to integrate physical modal information (mode shapes) into the deep learning framework via transfer learning. This integration enables the model to capture the spatial correlation among structural nodes, even under extremely limited data conditions. The MT-DL model is initially validated on a simply supported beam with varying stiffness, where it demonstrates significantly higher reconstruction accuracy compared to both conventional deep learning (DL) approaches and the graph convolutional network (GCN) model under sparse sensor conditions. To demonstrate the model's robustness and generalizability, it is further applied to the dynamic response reconstruction of a continuous beam subjected to moving loads, a plate with complex boundary conditions under impact loading, and a slender flexible riser undergoing vortex-induced vibration (VIV). The results show that the proposed MT-DL model, by incorporating physical principles into data-driven learning, not only enhances prediction accuracy with minimal data but also offers improved physical interpretability. This approach provides a promising solution for structural monitoring in scenarios where dense sensor instrumentation is impractical or cost-prohibitive.}, archive = {J_EAAI}, author = {Yangyang Liao and Yajuan Xie and Hesheng Tang and Zihan Xia and Songtao Xue}, doi = {10.1016/j.engappai.2025.112218}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112218}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A modal transfer enhanced deep learning for structural dynamic response with sparse spatial data}, volume = {161}, year = {2025}, }
• (2025). A novel deep learning pipeline for coronary artery analysis in X-ray angiography. EAAI, 161, 112217. (www)

  @article{ , abstract = {X- ray angiography is a primary and essential assistive method for the accurate diagnosis of coronary blockage. Segmentation of coronary arteries and analysis of coronary artery blockage are the two primary tasks performed by cardiologists. We have proposed Coronary Artery Segmentation, Blockage Detection, and Measurement (CASBloDaM) a state-of-the-art pipeline integrated deep learning framework with conventional image processing techniques to achieve accurate coronary artery analysis in X-ray angiographic images. A new private dataset of 214 X-ray angiographic images is prepared for training, validation, and testing of the model by manual pixel annotation. The UNet3+ deep learning model is trained on X-ray angiographic images for accurate artery segmentation while multiple image processing techniques are developed to perform blockage detection and measurement. The model shows excellent performance in artery segmentation and the dice score of 0.989, accuracy of 0.985, specificity of 0.913 and sensitivity of 0.847 is achieved during the testing which is superior when compared against the previous reported works. An accuracy of 0.853, 0.882 and 0.725 are achieved by the proposed catheter detection, sub-artery removal and blockage detection algorithms respectively when evaluated for 150 test images. The proposed methodology achieved a Mean Square Error (MSE) of 28.66 and an R-square of 0.81 for blockage percentage estimation, and an MSE of 69.91 with an R-square of 0.99 for blockage location, compared to manual calculations. The novelty of the present work is the end-to-end integrated framework which can accurately perform coronary artery analysis in X-ray angiographic images and propose its clinical usage.}, archive = {J_EAAI}, author = {Karan V. Padariya and Abhishek Raval and Pranay Soni and Harsh Kapadia}, doi = {10.1016/j.engappai.2025.112217}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112217}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel deep learning pipeline for coronary artery analysis in X-ray angiography}, volume = {161}, year = {2025}, }
• (2025). A multi-teacher knowledge distillation-based framework for long-term respiratory monitoring and prediction with a novel flexible wearable sensor in healthcare engineering. EAAI, 161, 112216. (www)

  @article{ , abstract = {Respiratory monitoring plays a critical role in early health warnings and preventive care, offering significant potential for advancements in healthcare engineering. Wearable respiratory monitoring devices, known for their compact design, portability, and real-time capabilities, face challenges such as limited long-term comfort, environmental interference, and signal inaccuracies. In this study, we propose a novel wearable respiratory monitoring framework, RAMP, which integrates an innovative artificial muscle-based flexible sensor system with advanced deep learning modules. The system is designed to reconstruct and analyze respiratory data across various human activities and predict long-term respiratory function. Utilizing a multi-teacher knowledge distillation mechanism, the framework optimizes a student model for enhanced prediction accuracy. Experimental results demonstrate the mean absolute percentage error (MAPE) of 7.28 and mean absolute error (MAE) of 11.92, highlighting the feasibility and effectiveness of system. This work advances the development of portable health monitoring devices and provides a robust foundation for long-term respiratory activity assessment and forecasting, contributing to the broader field of healthcare engineering and personalized medicine.}, archive = {J_EAAI}, author = {Ke Li and Qing Wang and Haoke Liu and Mingke Wang and Suiyuan Zhu and Xiang Wang and Jing Qin}, doi = {10.1016/j.engappai.2025.112216}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112216}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-teacher knowledge distillation-based framework for long-term respiratory monitoring and prediction with a novel flexible wearable sensor in healthcare engineering}, volume = {161}, year = {2025}, }
• (2025). Dynamic mask network based on spiking neural convolutional model for missing modality brain tumor segmentation. EAAI, 161, 112215. (www)

  @article{ , abstract = {Brain tumor segmentation is a medical image processing task aimed at accurately locating and isolating tumor regions from brain scan images (e.g., Magnetic Resonance Imaging, MRI) in order to help doctors in diagnosis, treatment planning and surgical navigation. Automatic brain tumor segmentation is extremely challenging due to incomplete feature representation in the case of missing modalities and insufficient inter-modal information interaction. To this end, this paper proposes a novel dynamic threshold mask Transformer network for the missing modality brain tumor segmentation task, which is designed based on the nonlinear spiking neural convolutional model. The network consists of four independent encoders and a shared decoder to extract the features of each modality and perform shared representation learning. Among them, the dynamic threshold mask Transformer introduces learnable embedding vectors, generates dynamic masks on top of static masks to achieve fine-grained feature filtering, and enhances the ability of inter-modal information interaction. The adaptive gating weighting module and the channel cross spiking neural P attention module fuse modal features layer by layer in both spatial and channel dimensions to strengthen the modeling capability of local and global features. We conducted extensive comparative experiments on different missing modal cases in the BraTS2020 and BraTS2018 datasets. The experimental results show that the method effectively improves the robustness of missing modalities and the performance of brain tumor segmentation while maintaining the computational efficiency, and has good generalization ability and practicality.}, archive = {J_EAAI}, author = {Junjie Li and Rui Cai and Bing Li and Hong Peng}, doi = {10.1016/j.engappai.2025.112215}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112215}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic mask network based on spiking neural convolutional model for missing modality brain tumor segmentation}, volume = {161}, year = {2025}, }
• (2025). Enhancing accuracy in fall detection and prediction for elderly individuals using ensemble wavelet neural network and maximal overlap discrete wavelet transform. EAAI, 161, 112213. (www)

  @article{ , abstract = {Falls are the second-largest risk factor for the health of the elderly. Various researchers have proposed approaches for monitoring health and falls using only the Internet of Things (IoT) and sensors. However, relying on a single sensor limits accuracy in fall risk prediction. Data from multiple sensors can be used to classify human posture through learning-based algorithms stemming from machine learning. Recently, posture-detecting sensors have become increasingly popular, with InvenSense's Inertial Measurement Unit 9250 (IMU 9250) being a notable example for detecting elderly motion and transmitting data to a computer. This paper proposes the Ensemble Wavelet Network Framework (EWNNET) for fall prediction and detection using the Maximal Overlap Discrete Wavelet Transform (MODWT). The inertial measurement unit (IMU) dataset, containing real-time posture data from seven sensors, is split into 80 % for training and 20 % for testing. The EWNNET approach is compared with other machine-learning methods for fall detection, showing improved accuracy, sensitivity, precision, F-score, and specificity, with a 96.7 % classification success rate.}, archive = {J_EAAI}, author = {Safa Hussein Mohammed and Yangyu Fan and Guoyun Lv and Shiya Liu}, doi = {10.1016/j.engappai.2025.112213}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112213}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing accuracy in fall detection and prediction for elderly individuals using ensemble wavelet neural network and maximal overlap discrete wavelet transform}, volume = {161}, year = {2025}, }
• (2025). Leakage localization methodology based on dynamic pressure signal for subsea pipeline. EAAI, 161, 112212. (www)

  @article{ , abstract = {Leakage is one of the most critical failure forms of subsea pipeline. The accurate leakage localization is of great significance to ensure the safe and reliable transportation of subsea pipeline. Leakage locations are considered to be discretely distributed along the subsea pipeline. However, an overabundance of nodes in leakage localization model is caused by excessive discretization. The performance of a leakage localization model with much localization points is poor. Furthermore, the data collected in site usually contains a lot of noise which reduce the effectiveness of leakage localization. A leakage localization methodology based on dynamic pressure signal for subsea pipeline is proposed in this paper. A noise reduction model based on variational mode decomposition (VMD) algorithm combined with power spectral density (PSD) is established to reduce noise of leakage signal. An improved K-means grouping model is developed to mining data for inherent similarity and cluster leakage characteristic. It improves robustness of the leakage localization model. A radial basis function (RBF) neural network leakage localization model optimized by the pelican optimization algorithm (POA) is used to identify the location of leakage. A leakage experiment is used to study performance of this methodology. The localization accuracy of the proposed leakage localization methodology is more than 90 %, the localization error is less than 16 cm. After neural network combined with improved grouping model, average leakage localization accuracy increased by 15.65 %, average absolute error decreased 8.64 cm. The proposed methodology provides an effective tool for leakage localization of critical equipment in subsea production system.}, archive = {J_EAAI}, author = {Guowei Ji and Baoping Cai and Xuelin Liu and Yi Jiang and Yixin Zhao and Qingping Li and Lei Gao and Kaizheng Wu}, doi = {10.1016/j.engappai.2025.112212}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112212}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Leakage localization methodology based on dynamic pressure signal for subsea pipeline}, volume = {161}, year = {2025}, }
• (2025). Enhancing entity and relation extraction with dynamic hard negative augmentation framework. EAAI, 161, 112211. (www)

  @article{ , abstract = {Entity and relation extraction, a fundamental task in information extraction, plays a crucial role in modeling unstructured text by identifying meaningful entities and their semantic relationships. While existing methods have shown effectiveness, they still face challenges in accurately identifying entity boundaries and extracting complex relationships. These challenges primarily arise from current contrastive learning approaches, which uniformly handle all negative samples in boundary detection and relation extraction without emphasizing the learning of hard negative samples. Additionally, the scarcity of hard negative samples limits the exploration of the state space near the anchors. To tackle these challenges, we introduce Dynamic Hard Negative Augmentation, an innovative framework designed to strategically explore and generate hard negative samples, thereby enhancing the learning of challenging cases through adaptive contrastive learning. During the negative sample augmentation process, we employ adversarial training to explore underrepresented areas of hard negative samples, generating a comprehensive coverage of the hard negative sample space to effectively explore the state space. We further introduce a dynamic enhancement mechanism that continuously optimizes the proportion of hard negative samples during training, ensuring targeted learning of these challenging cases.}, archive = {J_EAAI}, author = {Qibin Li and Shengyuan Bai and Nai Zhou and Nianmin Yao}, doi = {10.1016/j.engappai.2025.112211}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112211}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing entity and relation extraction with dynamic hard negative augmentation framework}, volume = {161}, year = {2025}, }
• (2025). NashDQNSleep: Nash-based deep Q-network adaptive sleep scheduling for energy efficiency and age of information optimization in industrial internet of things. EAAI, 161, 112210. (www)

  @article{ , abstract = {Fresh and timely data is essential for the sustainable operation of Industrial Internet of Things (IIoT) systems, which support real-time monitoring and decision-making tasks. Sensor nodes typically need to remain active for long durations to maintain data freshness, resulting in high energy consumption. Traditional sleep scheduling methods often trade off energy savings with data freshness, and many rely on centralized sink-node coordination, which can create communication bottlenecks and increase energy use. This paper introduces NashDQNSleep , a novel decentralized sleep scheduling algorithm that uniquely combines game-theoretic modeling with deep reinforcement learning to address these challenges. Unlike conventional DQN approaches that optimize node behavior independently or centrally, NashDQNSleep formulates the scheduling problem as a general-sum stochastic game and uses Deep Q-Networks (DQNs) to compute Nash equilibria. This enables sensor nodes to make autonomous, yet strategically coordinated decisions based on peer-to-peer information exchange, achieving a stable and efficient balance between energy consumption and data freshness without centralized control. Extensive simulations on diverse IIoT network topologies demonstrate that NashDQNSleep improves energy efficiency by up to 35%, reduces Age of Information (AoI) by 40%, and increases successful data transmissions by 15%–20% compared to state-of-the-art decentralized and centralized methods. These results establish NashDQNSleep as an effective, scalable, and practical solution for reliable and sustainable IIoT operations.}, archive = {J_EAAI}, author = {Partha Sarathi Banerjee and Soumyapriya Goswami and Debashis De}, doi = {10.1016/j.engappai.2025.112210}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112210}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {NashDQNSleep: Nash-based deep Q-network adaptive sleep scheduling for energy efficiency and age of information optimization in industrial internet of things}, volume = {161}, year = {2025}, }
• (2025). A novel contrastive learning framework for multi-parameter optimization in 3D printing. EAAI, 161, 112209. (www)

  @article{ , abstract = {Additive manufacturing (3D printing) revolutionizes prototyping and production through unparalleled material efficiency. However, part quality remains highly sensitive to parameter variations, where subtle deviations induce defects, material waste, and process instability. Traditional quality control methods – reliant on rule-based heuristics or manual inspections – fail to address complex multi-parameter interactions and fine-grained anomalies, limiting industrial scalability. This article focuses on an important issue in 3D printing: How can we develop a robust, automated framework to simultaneously detect and classify subtle multi-parameter anomalies in 3D printing, overcoming the limitations of manual and single-defect-focused approaches? We propose a supervised contrastive learning framework integrating Vision Transformers (ViT) to learn discriminative feature representations for multi-parameter optimization. By maximizing intra-class similarity and inter-class separation, our model captures nuanced variations across printing scenarios. The ViT architecture processes real-time printing images, while contrastive loss ensures compact feature clusters for “Low”, “Optimal”, and “High” parameter classes. Experimental evaluations on open-source datasets demonstrate our framework achieves 8.45% accuracy, outperforming conventional CNNs by 10.11%. Real-world validation shows robust performance across critical parameters: flow rate (86.5% accuracy in nominal ranges), feed rate (87% accuracy), and extrusion temperature (90% accuracy at optimal settings). The ViT’s self-attention mechanism enables precise detection of localized anomalies, such as under-extrusion and layer misalignment.}, archive = {J_EAAI}, author = {Jieyang Peng and Simon Kreuzwieser and Dongkun Wang and Andreas Kimmig and Zhi Fan and Jianing Li and Jivka Ovtcharova}, doi = {10.1016/j.engappai.2025.112209}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112209}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel contrastive learning framework for multi-parameter optimization in 3D printing}, volume = {161}, year = {2025}, }
• (2025). A lightweight model based on multi-scale feature fusion for ultrasonic welding surface defect detection. EAAI, 161, 112208. (www)

  @article{ , abstract = {Ultrasonic welding technology is crucial in industrial and medical fields, relying on precise surface defect detection for quality assurance. Traditional methods suffer from low accuracy, efficiency, high costs, and complex implementation. Additionally, current neural networks for ultrasonic surface defect detection struggle to balance parameter optimization with detection accuracy. To solve this problem, we proposed a lightweight model based on multi-scale feature fusion for the Ultrasonic Weld Surface Defect Detection Network (UWSDNet). First, the feature extraction module with reparameterization technology (FRT) and application of efficient multi-scale attention (EMA) are proposed to alleviate network redundant parameters and computational overhead brought by welding background. Secondly, the multi-core feature enhancement module (MCM) is introduced. It enhances multi-scale object detection with fewer parameters to cope with the actual edge deployment of ultrasonic welding. Finally, the lightweight asymmetric detection head (LADH) and contextual and spatial feature calibration network (CSFCN) are introduced into the network. To improve the multi-core dimensional feature capture capability, to solve the problem of large size span of ultrasonic welding surface defects. Experimental evaluations on a self-built ultrasonic welding wire harness defect dataset show that UWSDNet achieves the mean average precision (mAP) of 88.9%, the precision of 95.6% with parameters of 12.7M. In addition, UWSDNet achieves excellent performance on the publicly available NEU-DET dataset, demonstrating strong generalization and application potential in industrial defect detection.}, archive = {J_EAAI}, author = {Rui Liu and Lun Zhao and Yu Ren and Zhonghua Shen and Liya Li and Jianfeng Luo and Zeshan Abbas}, doi = {10.1016/j.engappai.2025.112208}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112208}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight model based on multi-scale feature fusion for ultrasonic welding surface defect detection}, volume = {161}, year = {2025}, }
• (2025). Artificial neural network solution for magnetohydrodynamic ternary hybrid second-grade nanofluid in rotating jeffery-hamel flow under heat generation: Entropy generation analyses. EAAI, 161, 112207. (www)

  @article{ , abstract = {This research-work examines entropy generation in Magnetohydrodynamic (MHD) ternary hybrid nanofluid under rotating Jeffery-Hamel flow with heat generation alongside thermal radiation. Using appropriate similar transformation equations, the original partial differential equations (PDEs) are converted into nonlinear ordinary differential equations (ODEs), while Artificial Neural Networks (ANNs), which highly extensively used as universal function approximators, is considered in this investigation. In fact, to provide efficient solutions for the nonlinear transformed problem of the magnetohydrodynamic ternary hybrid second-grade nanofluid in rotating Jeffery-Hamel flow, solvers consider log-sigmoid, radial basis and tan-sigmoid activation functions, optimized with an interior point method to generate optimal weights of each considered ANN model. The model relies on assumptions of incompressible steady flow together with negligible viscous energy loss and usage of Rosseland radiation approximation. The predictions made through the Artificial Neural Network (ANN) match numerical benchmarks successfully thus demonstrating better accuracy when modeling velocity, temperature and entropy profiles. The research data indicates that systems attaining better thermal qualities operate effectively under elevated nanoparticle levels and radiation conditions which provides fundamental knowledge to enhance thermal system design. The results from the proposed schemes match numerical solutions precisely thus showing their high accuracy for evaluating ternary hybrid nanofluid flow and heat behavior. Results show that the velocity and temperature profiles of the Magnetohydrodynamic (MHD) ternary hybrid nanofluid reveal significant improvements at higher thermal volume fractions of nanoparticles. The rate of flow increases, perhaps because inertial forces prevail over viscous forces, thereby improving flow dynamics. With higher Radiation numbers associated with higher permeability, the local temperature gradients decrease promoting heat transfer rates within the fluid.}, archive = {J_EAAI}, author = {Nouar Ahcene and Amar Dib and Farhan Lafta Rashid and Kezzar Mohamed and Mohamed Rafik Sari and Manal Elzain and Hamiden Abd El-Wahed Khalifa}, doi = {10.1016/j.engappai.2025.112207}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112207}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial neural network solution for magnetohydrodynamic ternary hybrid second-grade nanofluid in rotating jeffery-hamel flow under heat generation: Entropy generation analyses}, volume = {161}, year = {2025}, }
• (2025). Intelligent compound fault decoupling of rolling bearing based on parallel capsule network. EAAI, 161, 112206. (www)

  @article{ , abstract = {Due to the complex working environment of rolling bearings, various components such as inner ring, outer ring, rolling element and cage in bearing may interact with each other, leading to a compound fault formed by a variety of single fault coupling. Most methods generally regard compound faults as single faults, ignoring the interaction between single faults, which is not conducive to decoupling compound faults into multiple single faults and formulating maintenance plans. Moreover, the capsule network requires stacking multiple capsule layers to enhance performance, which significantly increases model parameters and consumes substantial memory resources. Therefore, a compound fault intelligent decoupling method based on parallel capsule network combining dynamic routing and attention routing is proposed in this study. Firstly, the Omni-Scale block is added to the feature extraction part, which can cover different sizes of receptive fields to enhance the feature extraction ability of the network. Secondly, an attention routing module is proposed, which realize the transmission of information from low-level capsules to high-level capsules by calculating the correlation between the same layers. Finally, the parallel capsule decoupling layer is constructed by using dynamic routing and attention routing. This method is especially suitable for practical engineering scenarios where compound bearing fault samples are limited and computational resources are constrained, providing a lightweight and effective solution for intelligent fault diagnosis. Experimental results show that the proposed method significantly reduces model complexity while maintaining high diagnostic performance under small-sample conditions, with the ablation study further confirming the meaningful contribution of each core module.}, archive = {J_EAAI}, author = {Renwang Song and Chenyu Jiao and Hui Shi and Linying Chen}, doi = {10.1016/j.engappai.2025.112206}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112206}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent compound fault decoupling of rolling bearing based on parallel capsule network}, volume = {161}, year = {2025}, }
• (2025). Deep reinforcement learning-based strategic bidding in electricity markets via variational autoencoder-assisted competitor behavior learning. EAAI, 161, 112205. (www)

  @article{ , abstract = {In a deregulated electricity market, self-interested producers have incentives to offer strategically for maximizing their own profits. While deep reinforcement learning (DRL) has shown great potential for solving such strategic bidding problems, existing methods typically oversimplify strategic action spaces and neglect the influence of competitors' offering behaviors. To bridge these gaps, this paper proposes a novel DRL-based framework to model and solve the strategic bidding problem of an individual producer by jointly considering price-quantity offering actions and the dynamic behaviors of market competitors. First, a bilevel optimization model is formulated to incorporate offering actions on price-quantity pairs. Then, a data-driven framework that combines a variational autoencoder with a density-based clustering method is proposed to learn and capture competitors' offering behaviors. Finally, an imitation learning-integrated DRL algorithm is developed to improve learning stability and solution quality for strategic bidding with price-quantity actions and competitors' offering behaviors. Case studies on the IEEE-30 bus system show that the proposed framework obtains a 28.12 k$ (24.25 %) increase in average profit compared to the existing approach, demonstrating its effectiveness and adaptability under dynamic market conditions.}, archive = {J_EAAI}, author = {Fei Hu and Yong Zhao and Yaowen Yu and Yuanzheng Li}, doi = {10.1016/j.engappai.2025.112205}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112205}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep reinforcement learning-based strategic bidding in electricity markets via variational autoencoder-assisted competitor behavior learning}, volume = {161}, year = {2025}, }
• (2025). A deep-supervised framework utilizing edge features and global texture features for distortion rectification in fisheye images. EAAI, 161, 112204. (www)

  @article{ , abstract = {Fisheye cameras, with their ultra-wide field of view, are increasingly deployed in urban environments. However, the severe radial distortion inherent in fisheye imagery presents formidable challenges for downstream computer vision tasks, including object recognition, motion estimation, and semantic segmentation. To address this issue, a deeply supervised framework for distortion rectification that leverages edge features and global texture information (DR-EGT) is proposed, which uniquely integrates edge-aware structural features with global texture information under a unified deep supervision strategy. Unlike conventional approaches that rely solely on texture priors or geometric assumptions, DR-EGT introduces a hierarchical supervision mechanism that simultaneously leverages low-level edge contours and high-level texture semantics to guide the distortion correction process. This joint supervision enables the network to learn the distortion flow field of the entire image, which facilitates the reconstruction of geometrically accurate and perceptually sharp undistorted images. On the Places2 dataset, DR-EGT outperforms existing advanced methods, achieving a 12.12 % increase in PSNR (Peak Signal-to-Noise Ratio), 7.44 % improvement in SSIM (Structural Similarity Index Measure), 16.42 % reduction in MAE (Mean Absolute Error), and a 78.17 % gain in FID (Fréchet Inception Distance), demonstrating its superior reconstruction fidelity and perceptual quality. The results demonstrate the ability of DR-EGT not only correct complex fisheye distortion but also suppress post-correction artifacts and visual degradation.}, archive = {J_EAAI}, author = {Yicheng Chen and Shuang Li and Yuhuan Gu and Chang Feng and Changhai Zhai}, doi = {10.1016/j.engappai.2025.112204}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112204}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep-supervised framework utilizing edge features and global texture features for distortion rectification in fisheye images}, volume = {161}, year = {2025}, }
• (2025). Spectrum prior-based and visibility fusion method for underwater image enhancement. EAAI, 161, 112203. (www)

  @article{ , abstract = {When light propagates in water, it undergoes scattering and absorption phenomena, which typically result in haze, high blur, low contrast and color distortion, making it extremely challenging to obtain high-quality images. To address these issues, many existing methods target image enhancement by correcting specific aspects such as color shift or contrast. However, challenges like poor visibility and low-light conditions are often overlooked. In this paper, we proposed a spectrum prior-based and visibility fusion method (SPV) to enhance underwater images in terms of color, contrast, and visibility. Unlike existing methods, SPV complements the advantages of both physical and non-physical models, comprehensively addressing the problems of reduced visual visibility, color distortion, and low contrast caused by low-light environments, thereby significantly improving the overall image quality. We proposed a dehazing module based on spectral information priors, which reliably restores image quality under complex water conditions. Additionally, we introduced a color correction module based on human color perception and employed morphological operations, effectively solving the issues of color shift and unclear contours in underwater images. Furthermore, we proposed a visibility enhancement module based on the fuzzy c-means clustering method to improve image contrast and visibility, particularly under low-light conditions. Finally, through a detail enhancement fusion module, we simultaneously addressed problems related to color shift, low contrast, and low visibility. SPV showed excellent performance in application tests including feature point matching, geometric rotation estimation, and edge detection. Comparative experiments on four real underwater datasets against 14 advanced enhancement methods demonstrated promising results.}, archive = {J_EAAI}, author = {Qifeng Liu and Xin Yan and Lu Shen and Qiang Li}, doi = {10.1016/j.engappai.2025.112203}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112203}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spectrum prior-based and visibility fusion method for underwater image enhancement}, volume = {161}, year = {2025}, }
• (2025). An integrated multi-criteria decision-making approach with unknown weight information under probabilistic dual-hesitant pythagorean fuzzy environment and its application to supplier selection. EAAI, 161, 112202. (www)

  @article{ , abstract = {The preference ranking organization method for enrichment evaluation (PROMETHEE) has been proved to be one of the most effective techniques to rank alternatives of multi-criteria decision-making (MCDM) problems. However, the existing PROMETHEE cannot accurately adjust the representation range of uncertain information. Besides, the weight determination in PROMETHEE heavily relies on the decision matrix of alternatives on the criteria. Moreover, the information aggregation in PROMETHEE models lacks consideration of the interrelationship among criteria. To address the aforementioned shortcomings, this paper introduces a novel MCDM method that integrates the best-worst method (BWM) and PROMETHEE to help the decision-maker (DM) select the optimum alternative under the probabilistic dual-hesitant Pythagorean fuzzy (PDHPF) environment. Firstly, we extend PROMETHEE method to PDHPF scenario, which not only helps the DM depict subjective evaluations, but also provides the DM with a laxer constraint to present decision information. Secondly, the PDHPF power weighted Hamy mean operator (PDHPFPWHM) is utilized to aggregate the preference information of PROMETHEE. Additionally, the BWM method is utilized and extended to the PDHPF environment for acquiring optimal weights of criteria. The high efficiency and consistency of BWM are well-suited for the complex PDHPF environment. Finally, the method is applied to a semiconductor supplier selection case to demonstrate the validity, superiority, and feasibility of the proposed approach.}, archive = {J_EAAI}, author = {Huzhi Xue and Haihua Xie and Butian Zhao and Jun Wang}, doi = {10.1016/j.engappai.2025.112202}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112202}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated multi-criteria decision-making approach with unknown weight information under probabilistic dual-hesitant pythagorean fuzzy environment and its application to supplier selection}, volume = {161}, year = {2025}, }
• (2025). A novel approach for understanding the parking demand for internal access roads in hub car parks. EAAI, 161, 112201. (www)

  @article{ , abstract = {The hubs serve as pivotal nodes within urban transport networks, and the car parks associated with these hubs constitute an integral component. The prevalent practice in China of utilizing internal access roads within car parks for passenger pick-up by online car-hailing vehicles has engendered a novel category of parking demand. However, this development presents a significant challenge to the efficient operation of hub car parks. A thorough analysis and comprehension of the parking demand for internal access roads is essential in devising suitable management strategies to enhance parking efficiency. Initially, this study categorizes the parking demand for internal access roads into three distinct groups, applying parking duration and the number of entries as classification criteria. Subsequently, considering the number of train frequencies, the categorized parking demands are forecasted independently utilizing the Long Short-Term Memory (LSTM) model, complemented by the Shapley Additive exPlanations (SHAP) method for model interpretation. Ultimately, Vector Autoregression (VAR) is employed to investigate the interaction mechanism between the parking demand for parking spaces and the parking demand for internal access roads. The findings indicate that optimal predictive performance is attained when employing a time interval of 15 min and an input step size of 4. Competition between the parking demand for internal access roads and the parking demand for parking spaces primarily occurs within respective categories, underscoring the need for external interventions when these demands are imbalanced.}, archive = {J_EAAI}, author = {Qianyi Hu and Weidong Liu and Chenyu Yan and Chu Zhang and Jun Chen and Changyin Dong}, doi = {10.1016/j.engappai.2025.112201}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112201}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel approach for understanding the parking demand for internal access roads in hub car parks}, volume = {161}, year = {2025}, }
• (2025). Machine learning for disease-specific prediction of high-cost patients. EAAI, 161, 112200. (www)

  @article{ , abstract = {High-cost patients incur disproportionately high medical expenses, and identifying them proactively is crucial for effective healthcare management. While previous research has focused on identifying high-cost patients based on overall expenditure, there has been a lack of studies analyzing them in the context of specific diseases. This study addressed this gap by leveraging data from the National Health Insurance Service (NHIS) of South Korea, spanning 2015 to 2019, to develop predictive models for identifying these patients. We trained models using data from 880,000 individuals to predict high-cost patients in 2019 using resource-efficient machine learning algorithms such as Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Neural Networks (NN) that minimize computational overhead, with undersampling techniques applied to handle data imbalance. We focused on the six major disease categories that account for the highest medical expenditures in South Korea: diseases of the musculoskeletal system (DMS), circulatory system (DCS), eye and ear (DEA-DEM), digestive system (DDS), genitourinary system (DGS), and respiratory system (DRS). We discovered that disease-specific analyses revealed important predictive factors that were not apparent in aggregate analyses. For example, hemoglobin levels emerged as crucial predictors for DCS, while body mass index (BMI) proved essential for DMS prediction. These findings enhance our understanding of the factors contributing to high medical costs and provide a foundational framework for healthcare providers and policymakers to develop more targeted and effective health management strategies.}, archive = {J_EAAI}, author = {Inwoo Tae and Hyeongwoo Kong and Junghye Lee and Yongjae Lee}, doi = {10.1016/j.engappai.2025.112200}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112200}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning for disease-specific prediction of high-cost patients}, volume = {161}, year = {2025}, }
• (2025). Optimizing mean conditional value-at-risk portfolios through deep neural network stock prediction. EAAI, 161, 112198. (www)

  @article{ , abstract = {Portfolio optimization is essential in financial decision-making, requiring a balance between risk minimization and return maximization. Effective stock selection significantly influences portfolio performance. Traditional methods often struggle to effectively integrate advanced risk assessment techniques with stock selection. To enhance portfolio diversification and improve risk-adjusted returns, this study integrates deep learning-based stock selection with the mean conditional Value-at-Risk (MCVaR) model and entropy constraints to enhance portfolio diversification and risk-adjusted returns. Various deep neural networks, including Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Multi-Layer Perceptrons (MLP), and Radial Basis Function Neural Networks (RBFN), are employed to rank stocks based on risk and return characteristics. The top-ranked stocks with the lowest risk are selected for portfolio construction. The entropy constraint is introduced to prevent excessive weight concentration, ensuring a well-diversified portfolio. Historical datasets from the Bombay Stock Exchange (BSE), India, B3 Stock Exchange, Brazil, and Shanghai Stock Exchange, China, are used for validation, with performance assessed on an out-of-sample dataset. Additionally, the efficacy of the suggested approach is evaluated by contrasting it with other machine learning and conventional portfolio optimization techniques. Experimental results demonstrate that the LSTM+MCVaR model with entropy constraint consistently outperforms other deep learning and conventional optimization methods, achieving superior cumulative returns and Sharpe ratios. The findings highlight the potential of combining LSTM forecasting with MCVaR optimization and entropy regularization for robust, diversified portfolio construction.}, archive = {J_EAAI}, author = {Jyotirmayee Behera and Pankaj Kumar}, doi = {10.1016/j.engappai.2025.112198}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112198}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing mean conditional value-at-risk portfolios through deep neural network stock prediction}, volume = {161}, year = {2025}, }
• (2025). Design and analysis of an unbiased intelligent recommendation system for all-rounders in cricket based on multiple criteria decision making. EAAI, 161, 112197. (www)

  @article{ , abstract = {Designing and analyzing an unbiased intelligent all-rounder recommendation system in cricket is a critical and complex decision-making task, where performance prediction itself is a crucial issue. The majority of existing works on cricket focus on batsmen, bowlers, and teams. However, performance analyses of all-rounders are hardly found. Hence, the motivation of this paper is to propose an artificial intelligence (AI)-based method for assessing the performance of all-rounders utilizing various multiple-criteria decision-making (MCDM) techniques. With this goal in mind, effective attributes are considered for evaluating all-rounder bowling and batting performances. Case studies using a set of 20 all-rounders have been taken from the recent International Cricket Council (ICC) all-rounders list to determine their ranking. The performance of various MCDM techniques is investigated using ICC rankings, with the Spearman Rank Correlation Coefficient applied to the One Day International (ODI) format. The results indicate that the Criteria Importance Through Intercriteria Correlation (CRITIC)-VlseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR) method yields quite promising insights, achieving a higher correlation with ICC rankings (Spearman Rank Correlation: 0.794) and hence can be used as an intelligent AI-based all-rounder recommendation system. Finally, we have also conducted a sensitivity analysis on the ranking outcomes of all-rounders to examine the utility and robustness of our proposed MCDM approach. The proposed approach is beneficial for team selectors, analysts, and fantasy sports platforms, offering a fairer and more reliable ranking of all-rounders.}, archive = {J_EAAI}, author = {Nayan Ranjan Das and Imon Mukherjee and Goutam Paul}, doi = {10.1016/j.engappai.2025.112197}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112197}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Design and analysis of an unbiased intelligent recommendation system for all-rounders in cricket based on multiple criteria decision making}, volume = {161}, year = {2025}, }
• (2025). Semantic analysis-based recommender system using sequential clustering and convolutional neural network. EAAI, 161, 112196. (www)

  @article{ , abstract = {Accurate prediction of user preferences and generation of personalized recommendations remain as critical challenges in intelligent recommendation systems. In this study, we propose a novel recommendation model that transforms the rating prediction problem into a single-label multiclass classification task. The model integrates three key components: (1) ordered clustering information derived from user review text similarity, (2) rating rank similarity reflecting users’ behavioral tendencies, and (3) a convolutional neural network (CNN) to extract semantic representations from user textual data. First, user review embeddings are clustered to capture high-level semantic preferences, where cluster indices are utilized as ordered categorical features. Second, rating rank similarity features are constructed by comparing the relative ranking of items rated by similar users. These features are fused and fed into a CNN model, which outputs a predicted rating class (e.g., 1–5 stars) for each unobserved item, treated as a single-label classification target. To generate final Top-N recommendations, we further incorporate user-specific rating habits and item popularity to re-rank the classification outputs. The experimental results on public benchmark datasets indicate that our model substantially improves the prediction accuracy and recommendation quality compared with existing baselines. The proposed method offers a robust and interpretable approach to bridging textual review semantics, user behavior, and deep learning for rating-aware personalized recommendation.}, archive = {J_EAAI}, author = {Yanjun Xu and Chunqi Tian and Wei Wang and Lizhi Bai}, doi = {10.1016/j.engappai.2025.112196}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112196}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semantic analysis-based recommender system using sequential clustering and convolutional neural network}, volume = {161}, year = {2025}, }
• (2025). Ranking-oriented cross-modal hashing. EAAI, 161, 112195. (www)

  @article{ , abstract = {Deep cross-modal hashing has become a mainstream solution for multimedia retrieval. Most methods utilize pair-wise or multi-wise loss as proxies for ranking consistency, focusing primarily on local relational patterns. However, these schemes are not well aligned with global ranking metrics, which consider the relative order among all candidate items. This misalignment may push semantically similar items down the list. Additionally, such methods heavily rely on the selection of negative samples, which can cause training instability and require careful tuning. To address this problem, we propose a novel Ranking-Oriented Cross-Modal Hashing (ROCMH) framework, which performs global ranking optimization based on a vision-language model. Specifically, we design a differentiable ranking-oriented objective, called cross-modal ranking alignment loss. It smoothly simulates discrete ranking and naturally models ranking relations in candidate lists, thereby promoting more consistent cross-modal ranking. Meanwhile, considering the perceptual gap between visual and textual, we suggest a nanoparam training strategy with modality-aware prompts. This strategy greatly reduces the number of trainable parameters while ensuring that each modality provides unique signals for semantic information learning. Extensive experiments on three public datasets demonstrate that the method achieves superior retrieval performance and reduces the number of trainable parameters to less than 0.2% of the vision-language model. The source code is available: https://github.com/QinLab-WFU/ROCMH .}, archive = {J_EAAI}, author = {Yadong Huo and Qibing Qin and Wenfeng Zhang and Lei Huang}, doi = {10.1016/j.engappai.2025.112195}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112195}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ranking-oriented cross-modal hashing}, volume = {161}, year = {2025}, }
• (2025). Matching quality-guided model-free satellite pose estimation. EAAI, 161, 112194. (www)

  @article{ , abstract = {With the development of learning-based techniques and large-scale public datasets, satellite pose estimation has seen significant progress for the past several years. However, most of current methods still rely on a known three-dimensional (3D) model of the object for the pose estimation, limiting its generalization and wide application. To this end, we propose a model-free pose estimation method, which takes as input only a set of images. The proposed method consists of two stages, i.e. the reconstruction stage and pose estimation stage, of which the former reconstructs a 3D model from the input images and the pose is estimated via two-dimensional (2D)-3D feature matching by the latter. More importantly, a matching quality guidance strategy is introduced to further improve the robustness to in-plane rotation during feature matching. Additionally, since no known 3D model is assumed, our method generalizes well to novel objects without retraining. We provide evaluation results on datasets of several satellites with different structures, which demonstrate impressive performances without known models against current methods.}, archive = {J_EAAI}, author = {Zhaoshuai Qi and Yating Liu and Yanning Zhang}, doi = {10.1016/j.engappai.2025.112194}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112194}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Matching quality-guided model-free satellite pose estimation}, volume = {161}, year = {2025}, }
• (2025). Prebuilt spatiotemporal index: An exploration of efficient real-time data storage in intelligent transportation systems. EAAI, 161, 112193. (www)

  @article{ , abstract = {With the widespread deployment of Internet of Things (IoT) devices in Intelligent Transportation Systems (ITS), real-time spatiotemporal data has grown rapidly. Hybrid data-bases have emerged as the mainstream solution for managing such data. The LSM R*-tree, which integrates Log-Structured Merge-trees (LSM-trees) for time-series ingestion and R*-trees for spatial queries, is now a widely used spatiotemporal index. However, storage schemes based on LSM R*-tree structures involve real-time index construction, leading to significant overhead and write latency. To address these challenges, this paper proposes a prebuilt index-based data storage workflow that shifts index construction ahead of data writing. This approach allows the database to directly apply the prebuilt index at write time, thereby minimizing real-time construction costs and enhancing write performance. To support index prebuild, we propose a lightweight embedding scheme, Index2Vec, specifically designed for R*-tree structures. Based on this, we extend the Transformer architecture and develop the R*-tree Prediction Network for ITS (RTPN4ITS), which achieves efficient inference on resource-constrained edge devices. Experimental results show that the prebuilt R*-tree index improves query efficiency by up to 90% over time-index-only schemes and enhances write performance by nearly 50% compared to real-time indexing. The proposed RTPN4ITS model achieves robust accuracy across varying traffic densities, reaching 85% accuracy in dense conditions. Moreover, the Index2Vec embedding enhances the Transformer’s structural awareness. In summary, this paper proposes an efficient prebuilt indexing strategy and lightweight embedding-based model for real-time spatiotemporal data management in ITS.}, archive = {J_EAAI}, author = {Yiran Shao and Kangshuai Zhang and Yong Zhou and Zhenwu Chen and Yang Yang and Lei Peng}, doi = {10.1016/j.engappai.2025.112193}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112193}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prebuilt spatiotemporal index: An exploration of efficient real-time data storage in intelligent transportation systems}, volume = {161}, year = {2025}, }
• (2025). An error complementarity-based iterative learning approach via categorical boosting for student performance prediction. EAAI, 161, 112192. (www)

  @article{ , abstract = {Predicting student performance is important in achieving academic success and student development in many educational applications (e.g., academic early warning and early interventions). To accurately predict student performance, we propose an error complementarity-based iterative learning approach for student performance prediction. Our goal is to improve the prediction performance by iteratively reducing the prediction error. Specifically, in model construction, we first select the most important features to train the target estimator. Then, we obtain the errors between the target and predicted values. These errors are used to train the error estimator using the remaining features. Similarly, we iteratively train the model to learn from the errors until the desired conditions are met. In model prediction, we predict the testing sample using the target estimator to obtain the predicted value. Next, we predict the error for the next iteration using the corresponding error estimator. This process is repeated, and the final prediction is obtained by adding the predicted value and the error values. The extensive experiments from different educational datasets show that by using our error complementarity-based iterative learning approach, the proposed model outperforms the competing models for prediction accuracy. Furthermore, statistical testing conducted over 20-run experiments confirms the significant advantage of our proposed model. This suggests the iterative learning is effective for predicting student performance.}, archive = {J_EAAI}, author = {Zongwen Fan and Jin Gou and Cheng Wang}, doi = {10.1016/j.engappai.2025.112192}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112192}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An error complementarity-based iterative learning approach via categorical boosting for student performance prediction}, volume = {161}, year = {2025}, }
• (2025). Federated cross-device cluster dynamic time planning warping algorithm in the industrial internet of things. EAAI, 161, 112191. (www)

  @article{ , abstract = {Production lines and heterogeneous devices generate diverse time-series data in the Industrial Internet of Things (IIoT). This diversity necessitates frequent updates of local models when using federated learning (FL). FL has become widespread in the IIoT, these continual model updates can hinder timely information exchange among cross-industry devices. They constrain the system's ability to handle high concurrency in read and write operations. Therefore, we propose a federated cross-device cluster dynamic time planning warping (Fed-cDTW) tailored for the IIoT. We incorporate each device's historical behavior data by constructing a functional characteristic distance matrix. We design a symmetric time warping lower bound function to measure the similarity among multi-source time-series data. This approach enables an adaptive, dynamically organized industrial device cluster. Meanwhile, we employ the federated dynamic (FedD) dual-weight optimization within the device cluster to enhance the models' generalization and robustness. The experimental results evaluated the performance of the proposed method. We compare our method with baselines, Fed-cDTW improves classification accuracy by 0.8 percentage points (pp), 1.7 pp, 9.6 pp, 14.4 pp, and 49.5 pp. Convergence is accelerated by 0.04, 0.05, 0.35, 0.42, and 0.46. Total communication cost is reduced by 40 megabytes (MB), 56 MB, 84 MB, 120 MB, and 168 MB.}, archive = {J_EAAI}, author = {Yifan Zhao and Shibao Sun and Pengcheng Zhao and Yatong Wang and Jianfeng Liu}, doi = {10.1016/j.engappai.2025.112191}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112191}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Federated cross-device cluster dynamic time planning warping algorithm in the industrial internet of things}, volume = {161}, year = {2025}, }
• (2025). Machine learning-based prediction of compressive strength in sustainable self-compacting concrete. EAAI, 161, 112190. (www)

  @article{ , abstract = {The performance and durability of conventional concrete (CC) is significantly influenced by supplementary cementitious materials (SCMs) and recycled coarse aggregate (RCA). Thus, the intrusion of enrich SCMs with RCA in the cementitious matrix delivers utmost properties. This research focuses on the use SCMs and RCA in the self-compacting concrete (SCC) system and their sustainability in the development of concrete resources. Standard experimental methods for forecasting the compressive strength (CS) of SCC have constraints in terms of efficiency, time consumption, and cost. Thus, its prediction is crucial without the need for laborious experimental procedures. This work employs and evaluates a multiplicity of machine learning (ML) models to predict the CS of the cementitious matrix to tackle these issues. Thus, considered a more effective and cost-saving solution in comparison with traditional approaches. Therefore, ML approaches like, gene expression programming (GEP), decision trees (DT), and support vector regression (SVR) were employed. The performance of the model is evaluated by employing the coefficient of determination (R 2 ), statistics, and uncertainty analysis. Individual Conditional Expectation (ICE), and Partial Dependence Plot (PDP) are used to analyze the effect of parameters on strength. The findings suggest that GEP performs best achieving superior R 2 > 0.90 for the training, validation, and test data sets with a lower error. While, the uncertainty analysis shows that all modeled values lie below the threshold value. The ICE and PDP graphs confirms that cement, age, and water-cement ratio have highly relation to outcomes. The RCA replacement ratio is more important than the CA one, and SCMs play an essential role in the development of concrete compressive strength, although not as much as cement. In addition, SP depicts major contribution to SCC. Moreover, graphical user interface (GUI) is also developed to help users/researcher that will facilitate them to estimate the strength of SCC in practical applications.}, archive = {J_EAAI}, author = {Jingguo Gou and Athar Zaman and Furqan Farooq}, doi = {10.1016/j.engappai.2025.112190}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112190}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-based prediction of compressive strength in sustainable self-compacting concrete}, volume = {161}, year = {2025}, }
• (2025). A robust fuzzy twin support vector machine with kernel-target alignment for binary classification. EAAI, 161, 112189. (www)

  @article{ , abstract = {Many algorithms similar to the twin version of the support vector machine and their variants have shown better results in the binary classification of nonlinear data points. But in the presence of outlier and noise, these algorithms exhibit low generalization efficiency. To alleviate this challenge, recently proposed, kernel-target alignment based fuzzy least square twin bounded support vector machine (KTA-FLSTBSVM) used fuzzy membership values and is solved using least squares. Inspired by this strategy, for further improvement, we propose a novel approach called decision support kernel-target alignment based fuzzy least square twin bounded support vector machine (DS-KFIFTBSVM). DS-KFIFTBSVM considers the kernelized fuzzy membership values with the regularized twin support vector machine and solves for linear and nonlinear data points using a functional iterative approach. In DS-KFIFTBSVM, the solution is obtained by solving a linearly convergent iterative scheme rather than solving quadratic programming problems. The proposed DS-KFIFTBSVM offers better generalization efficiency, which has been evaluated using both linear and Gaussian kernels, mostly on artificially developed and publicly accessible datasets with diverse dimensionalities. In terms of various performance evaluation metrics, including specificity, precision, false positive rate, rate of misclassification error, F_score, and geometric mean in linear and non-linear cases, DS-KFIFTBSVM outperforms various baseline approaches. It shows the highest accuracy in various datasets, including 98.1884 % for musk dataset (linear kernel) and 100 % for the glass dataset (Gaussian kernel). Further statistical analysis confirms its classification efficiency.}, archive = {J_EAAI}, author = {Deepak Gupta and Barenya Bikash Hazarika and Umesh Gupta and Witold Pedrycz}, doi = {10.1016/j.engappai.2025.112189}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112189}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust fuzzy twin support vector machine with kernel-target alignment for binary classification}, volume = {161}, year = {2025}, }
• (2025). Comprehensive assessment of pore-scale segmentation techniques for image-based multiphase flow characterization in porous media. EAAI, 161, 112188. (www)

  @article{ , abstract = {Evaluation of multiphase fluid flow behavior based on pore-scale imaging is significantly influenced by the accuracy of the segmentation techniques. This study sheds light on the influence of various traditional and Artificial Intelligence (AI) algorithms on the major static, dynamic, volumetric, and topological properties of the porous media. The traditional methods include the Gaussian Mixture Model (GMM), Multi-Otsu’s thresholding (MOT), and Random Walk (RW), and the AI-based techniques include Trainable Weka (TW), as well as seven deep convolutional autoencoder architectures. The assessment was carried out by comparing the calculated phase fraction, contact angle, capillary pressure, effective permeability, and topology for multiphase flow images of Bentheimer sandstone. The results revealed that combining Residual Networks (ResNet) and UNet structures (UResNet) marginally performs better based on the statistical metrics. Compared to ground-truth watershed images, applying other approaches except TW yielded acceptable static and volumetric parameters. The main discrepancies were in contact angles, capillary pressure, and phase topology. These discrepancies also influenced calculated effective permeabilities, yielding unrealistic trends. UResNet provided the best accuracy, considering the watershed the ground truth. MOT and RW showed similar performance, while GMM performed inconsistently throughout the analysis and TW produced acceptable results for certain fractional flows. The segmentation approaches affected topological features most, whereas properties such as phase fractions were less affected. The inconsistent behavior and effectiveness of used algorithms indicate that the analysis of multiphase images necessitates a meticulous choice of segmentation techniques, emphasizing AI-based algorithms that ensure consistency across all evaluations.}, archive = {J_EAAI}, author = {Javad Siavashi and Mohammad Sharifi}, doi = {10.1016/j.engappai.2025.112188}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112188}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comprehensive assessment of pore-scale segmentation techniques for image-based multiphase flow characterization in porous media}, volume = {161}, year = {2025}, }
• (2025). Land use and land cover change detection from multisource satellite imagery – A hybrid convolutional neural network approach. EAAI, 161, 112187. (www)

  @article{ , abstract = {Remote sensing and satellite imagery analysis enable us to observe and understand environmental changes over time. However, such an endeavor can be complex and time-consuming. To overcome this challenge, we propose a framework for leveraging Google Earth Engine (GEE) to analyze Earth's surface changes over time. Our framework starts by assessing land cover changes using satellite imagery, evaluating the effectiveness of vegetation indices in classification, and generating accurate land cover maps for a specific region. We examine the validity of this framework on the area near Rio Do Sul, located in the Santa Catarina district of Brazil, to classify and recognize the alteration in land use and land cover (LULC) from 2019 to 2023. In this effort, we develop a pre-trained Convolutional Neural Network (CNN) for feature extraction from the Landsat 8, Sentinel-1, and Sentinel-2 images and then link those into the random forest (RF) algorithm for spatial morphology and temporal change logic to map the long-term annual time series and detect changes in the region at hand. The novelty of this study arises from developing a CNN-RF hybrid model and implementing a data preprocessing pipeline, which includes progressive techniques for cloud masking and radiometric calibration, to ensure higher accuracy and reliability in land use and land cover classification. This CNN-RF hybrid model successfully traced the increasing and decreasing rate in the built-up area, water bodies, forest, fallow land, plantations, and grassland with an overall accuracy of 96.39 % and 94.15 % for 2019 and 2023, respectively.}, archive = {J_EAAI}, author = {M.S. Babitha and A. Diana Andrushia and N. Anand and M.Z. Naser and Y. Pari}, doi = {10.1016/j.engappai.2025.112187}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112187}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Land use and land cover change detection from multisource satellite imagery – A hybrid convolutional neural network approach}, volume = {161}, year = {2025}, }
• (2025). Safety monitoring digital twin-based centralized model consolidation mechanism using dynamic node selection for multi-worker safety prediction. EAAI, 161, 112186. (www)

  @article{ , abstract = {The construction industry remains one of the most hazardous sectors, requiring innovative solutions to safeguard workers, especially in complex, dynamic outdoor environments. Digital Twin (DT) technology offers promising capabilities for real-time safety monitoring through virtual replicas of physical systems. However, existing DT frameworks rarely integrate comprehensive safety monitoring via a centralized model consolidation mechanism, such as Federated Learning (FL), which is explicitly tailored for multi-worker scenarios. Addressing this challenge, this paper proposes a FL-based Safety Monitoring Digital Twin (SMDT) framework designed to enhance multi-worker safety in resource-constrained settings. This enables real-time safety monitoring and control by representing on-site workers as virtual objects within a synchronized DT environment. A dynamic node selection mechanism based on client performance is employed to optimize global model convergence in FL. To validate the proposed approach, an edge computing-based experimental testbed using actual Raspberry Pi devices was implemented, using real-world construction safety data including worker status, weather conditions, and building structural parameters. Experimental results demonstrate the effectiveness of the proposed framework in significantly improving safety predictions and real-time monitoring efficiency. This research establishes a foundational work towards safer construction sites through intelligent, synchronized safety monitoring systems.}, archive = {J_EAAI}, author = {Sa Jim Soe Moe and Atif Rizwan and Anam Nawaz Khan and Rongxu Xu and Do Hyeun Kim}, doi = {10.1016/j.engappai.2025.112186}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112186}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Safety monitoring digital twin-based centralized model consolidation mechanism using dynamic node selection for multi-worker safety prediction}, volume = {161}, year = {2025}, }
• (2025). Universal entity linking. EAAI, 161, 112185. (www)

  @article{ , abstract = {Entity linking is a process of connecting mentions of entities in a document to corresponding entries in a knowledge base. Traditional models for entity linking often require specific fine-tuning to work with knowledge bases other than those they were originally pretrained on, which limits their flexibility and scalability. Building on the concept of entity profile generation, we propose a novel approach that enables entity linking across various knowledge bases without the need for such fine-tuning. Our pipeline leverages a fine-tuned Large Language Model, a generic embedding model, and a vector store to achieve high precision on the TweekiGold and Reuters-128 datasets. Additionally, it demonstrates strong retrieval rates across the TweekiGold , Reuters-128 , and ISTEX-1000 Wikidata entity linking datasets. We also illustrate the applicability of our method to other knowledge bases, using the Agrovoc knowledge base as an example. This solution offers a more versatile and scalable approach to entity linking.}, archive = {J_EAAI}, author = {Adam Aron Rynkiewicz and Raul Palma and Piotr Formanowicz}, doi = {10.1016/j.engappai.2025.112185}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112185}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Universal entity linking}, volume = {161}, year = {2025}, }
• (2025). A search method for fractured-vuggy reservoir inter-well connectivity path based on multi-modal multi-agent. EAAI, 161, 112184. (www)

  @article{ , abstract = {The complex geological structure of carbonate reservoirs and the intricate fracture-vuggy configurations obscure inter-well connectivity, making its evaluation challenging. Conventional studies primarily rely on seismic static data to delineate fracture-vuggy reservoirs, but the limited recognition accuracy hampers the precise characterization of inter-well connectivity and the spatial configuration of fractures and vugs. To address this, this study constructs a 3D (Three-Dimensional) search environment and use multi-modal static and dynamic data and proposes a multi-agent connected channel search model based on deep reinforcement learning. The model treats multiphase fluid as an agent and incorporates Swin Transformer (Shift Window Transformer) to extract large-scale fracture features from seismic data, providing global prior information for path search. A Graph Attention Network is established based on dynamic response relationships to extract spatial geological features, while a multi-head self-attention mechanism captures real-time fluid interactions in various directions. The model fuses multi-modal features, including seismic attributes and production data, to generate decisions and automatically search for inter-well connectivity channels. Experiments were conducted using the WE1 and WE5 well groups from the fault-controlled karst reservoirs in the Tahe oilfield, with results compared against tracer tests. The findings demonstrate that the proposed model's automatic search paths closely align with seismic data and tracer test results, effectively capturing the spatial distribution of fractures and vugs across different scales. This validates the model's effectiveness in evaluating inter-well connectivity in complex carbonate reservoirs.}, archive = {J_EAAI}, author = {Wenbin Jiang and Dongmei Zhang and Hong Cao and Xiaofeng Wang}, doi = {10.1016/j.engappai.2025.112184}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112184}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A search method for fractured-vuggy reservoir inter-well connectivity path based on multi-modal multi-agent}, volume = {161}, year = {2025}, }
• (2025). Continuous spatio temporal prompts for visual tracking. EAAI, 161, 112183. (www)

  @article{ , abstract = {Currently, visual single-object tracking methods utilize online template updates to combine temporal information. However, these methods rely on confidence scores to evaluate the reliability of the current template, which may result in a template not being updated for an extended period. Moreover, advanced trackers select bounding boxes based solely on the similarity between the template and the search area, which can lead to tracking drift when encountering deformable or similar targets. To alleviate these limitations, we propose a Spatio Temporal Prompt Tracker (STPTrack), which utilizes the prior information about small changes of object state between successive frames. Different from previous tracking methods that mainly rely on templates and similarity scores, STPTrack transfers the object position and shape information of the previous frame to the current frame as continuous spatio temporal prompt for the first time, and realizes the efficient fusion of spatio temporal information through the prompt encoder and the fusion decoder module. Specifically, it encodes the bounding box coordinates or mask information of the previous frame and the response points of the current frame as prompt features, and then combines prompt tokens with search tokens through the fusion decoder to provide the potential location of the object for the search feature map. Our STPTrack sets a new state-of-the-art performance on six tracking benchmark datasets.}, archive = {J_EAAI}, author = {Meng Sun and Xiaotao Liu and Yifan Li and Hongyu Wang and Dian Yuan and Jing Liu}, doi = {10.1016/j.engappai.2025.112183}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112183}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Continuous spatio temporal prompts for visual tracking}, volume = {161}, year = {2025}, }
• (2025). Length-aware center loss for sequence to sequence thai scene text recognition. EAAI, 161, 112182. (www)

  @article{ , abstract = {Thai scene text recognition is a challenging task because Thai can be written in both horizontal and vertical directions, allowing characters to be stacked vertically. To address this issue, our previous work combined vertically stacked characters to create new characters. However, this strategy introduced many similar characters. In this paper, we further investigate this problem and propose the Length-aware Center Loss (LC) for Thai scene text recognition. The original center loss was designed for single object recognition tasks. When applied to multi-label tasks like text recognition, center loss is only effective when the lengths of the labels and prediction results are consistent. This can lead to an extreme case where all images receive incorrect predicted text lengths to minimize loss, severely interfering with the recognition process. Therefore, we propose the Length-aware Center Loss for text recognition. We also design the Length Supervision Module (LSM) and the Feature Clustering Module (FCM) to work alongside the LC loss. LSM predicts text length to provide additional supervision signals, while FCM aims to improve recognition performance by minimizing the distance between the features of corresponding class centers. Since there is no publicly available Thai scene text dataset, we have collected a new dataset containing more than 170,000 samples. Extensive experiments conducted on this dataset show that our method achieves superior performance in both string-level and character-level accuracy compared to other methods.}, archive = {J_EAAI}, author = {Hongjian Zhan and Chun Li and Bing Yin and Yue Lu}, doi = {10.1016/j.engappai.2025.112182}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112182}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Length-aware center loss for sequence to sequence thai scene text recognition}, volume = {161}, year = {2025}, }
• (2025). Diffusion-based vision-language model for zero-shot anomaly detection in medical images. EAAI, 161, 112181. (www)

  @article{ , abstract = {With the rapid advancement of diagnostic technology, the ability to detect pathological areas such as tumors and polyps has significantly improved. This progress provides medical imaging specialists with more precise visual information to support anomaly identification, diagnosis, treatment planning, and patient monitoring. However, existing unsupervised and semi-supervised anomaly detection methods struggle with data privacy constraints, limited annotated medical datasets, and challenges in generalization. Zero-Shot Anomaly Detection (ZSAD), which enables the detection of unseen categories without requiring class-specific training, has emerged as a promising solution by leveraging the vision-language alignment capabilities of Vision-Language Models (VLMs), such as Contrastive Language-Image Pretraining (CLIP). Despite recent progress, ZSAD remains hindered by high noise levels, sparse targets, and poor adaptability in complex medical imaging scenarios. To address these issues, we propose a novel framework: DiffusionCLIP, a diffusion-based VLM for zero-shot anomaly detection in two-dimensional medical images. Specifically, DiffusionCLIP integrates diffusion models into the VLM to progressively denoise multi-level features extracted from the CLIP visual encoder, enhancing feature robustness and discriminability. A multi-level feature fusion strategy is designed to aggregate multi-scale representations from different depths of the visual encoder, ensuring complementary semantic alignment across layers. In addition, a dynamically modulated weight loss function is introduced to adaptively balance the learning of hard and easy samples, further improving model generalization. Extensive experiments on multiple benchmark medical imaging datasets, demonstrate that the proposed method significantly outperforms existing zero-shot anomaly detection approaches in terms of accuracy, robustness, and generalization.}, archive = {J_EAAI}, author = {Yanhui Chen and Hongkang Tao and Zan Yang and Yunkang Cao and Chen Jiang and Longhua Hu and Pengwen Xiong and Haobo Qiu}, doi = {10.1016/j.engappai.2025.112181}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112181}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Diffusion-based vision-language model for zero-shot anomaly detection in medical images}, volume = {161}, year = {2025}, }
• (2025). Surface roughness prediction in turning processes for grey cast iron: A hybrid machine learning approach integrating infrared thermography. EAAI, 161, 112180. (www)

  @article{ , abstract = {Workpiece surface quality is a critical control parameter in machining processes, influencing functional performance, dimensional precision, and wear resistance. However, accurately predicting surface roughness is complex, often limited by the computational demands of traditional high-precision methods and the reliance of existing models solely on cutting parameters, hindering real-time monitoring. This research introduces a hybrid Artificial Neural Network (ANN) methodology specifically developed for predicting surface roughness of grey cast iron GG-25 workpieces machined in turning processes, a material previously unstudied in this context. The methodology integrates real-time infrared thermal measurements from multiple defined regions of interest (ROIs) within the tool-workpiece contact zone, along with cutting parameters. Experimental results demonstrated that feed rate (f) is the most significant cutting parameter (effect = 0.43) affecting surface quality, followed by its combination with cutting speed (V c ) (effect = −0.25) and cutting speed (effect = 0.18). Correlation and non-linear regression analyses revealed complex, often exponential relationships between temperature and surface roughness, showing temperature an upward trend as machining progressed. The developed ANN achieves a correlation coefficient (R) value of 0.99 both when predicting the roughness arithmetic mean deviation (Ra) parameter in training conditions and when using data from experiments not used in training (validations data). Moreover, the model reaches a correlation coefficient value of 0.85 (test data) under cutting conditions different from those used in experiments, demonstrating robustness, significantly outperforming Support Vector Regression (SVR). This model represents a highly potential tool for real-time online inspection.}, archive = {J_EAAI}, author = {Sergio Aguado and Marcos Pueo and Raquel Acero and Ana Cristina Majarena and Jorge Santolaria}, doi = {10.1016/j.engappai.2025.112180}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112180}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Surface roughness prediction in turning processes for grey cast iron: A hybrid machine learning approach integrating infrared thermography}, volume = {161}, year = {2025}, }
• (2025). Legendre multiwavelet-based feature attention guidance lightweight network for accurate steel surface defect classification. EAAI, 161, 112179. (www)

  @article{ , abstract = {Steel surface defects can significantly affect the quality and appearance of industrial products such as aerospace, construction application fields, and so on. Due to the multi-scale morphology, low contrast, and random positions of the defects, achieving a favorable balance between detection accuracy and speed remains challenging in practical applications. To address these challenges, this paper combines Legendre multiwavelet (LW) with feature attention guidance (FAG) mechanism to devise a novel high-accuracy lightweight network (LWFAG-LNet) for surface defect classification. More precisely, LW bases with rich regularities are first utilized to match complex geometric characteristics across multi-wavelet and multi-scale resolution levels. Subsequently, the FAG module effectively fuses shallow high-resolution detailed features with deep low-resolution contextual features, significantly reducing the depth of the convolutional neural network (CNN). To the third step, a lightweight CNN module with four convolutional blocks is designed to further extract deep features while preserving the most valuable defect information. The proposed model achieves the highest recognition accuracy with a simple structure and a compact parameter configuration. Extensive experiments are conducted on the Northeastern University-Classification (NEU-CLS), Xsteel surface defect dataset (X-SDD), and Kungliga Tekniska Högskolan Royal Institute of Technology Textures under varying Illumination, Pose and Scale (KTH-TIPS) dataset to verify the model's effectiveness and generalization capability. The results reach classification accuracies of 99.80 %, 98.49 %, and 99.06 %, outperforming existing models by about 1.45 %, 3.45 %, and 2.04 %, respectively. In summary, the proposed LWFAG module can be flexibly applied to various lightweight network frameworks, demonstrating great potential for real-time surface defect detection.}, archive = {J_EAAI}, author = {Xiaoyang Zheng and Weishuo Liu and Yan Huang}, doi = {10.1016/j.engappai.2025.112179}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112179}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Legendre multiwavelet-based feature attention guidance lightweight network for accurate steel surface defect classification}, volume = {161}, year = {2025}, }
• (2025). Detecting object hue bin in the CIECAM02 uniform color space using a multi-layer perceptron for optimal spectrum generation. EAAI, 161, 112178. (www)

  @article{ , abstract = {In this paper, we present an innovative enhancement to an existing multi-channel light-emitting diode luminaire by integrating artificial intelligence and image analysis capabilities. The system captures an image of the object to be illuminated using a custom-developed Android application. The dominant color in the image is first identified, and then a multi-layer perceptron neural network classifies it into one of 16 hue bins within a uniform color appearance model. This classification process enables the luminaire to select a predefined spectrum designed to enhance the visual appeal of the illuminated object. To ensure robust hue bin classification, we developed a rich dataset with 7920 samples, collected using eleven smartphones under ten distinct light sources. Using this diverse dataset, thirteen multi-layer perceptron neural networks with varying input features were trained and compared, achieving hue classification accuracies, precision, and recall all ranging from 95 % to 99 % on the test set. One of the trained multi-layer perceptron networks was integrated into the Android application, with an inference time of 90 μs on a mid-range smartphone, demonstrating the practical application of artificial intelligence for on-demand hue analysis and lighting optimization. This advancement not only enhances the luminaire's functionality but also enriches the user experience by providing intelligent, on-demand lighting solutions tailored to the object's dominant color.}, archive = {J_EAAI}, author = {Esmat Kishani Farahani and Kaveh Ahmadian Tazehmahaleh}, doi = {10.1016/j.engappai.2025.112178}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112178}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detecting object hue bin in the CIECAM02 uniform color space using a multi-layer perceptron for optimal spectrum generation}, volume = {161}, year = {2025}, }
• (2025). Generalized zero-shot fault diagnosis method for chillers based on cross-modal information compression. EAAI, 161, 112177. (www)

  @article{ , abstract = {In practical chiller fault diagnosis applications, target fault training data is often inaccessible, which poses significant challenges to the effectiveness of data-driven diagnostic approaches. Generalized zero-shot fault diagnosis (GZSFD) aims to detect and classify all fault types without relying on fault samples from all categories during training. GZSFD models are trained exclusively on data from seen faults—those with available historical data—making them prone to bias toward these seen classes during inference. However, existing methods fail to exploit the fine-grained semantics of fault attributes, resulting in suboptimal performance when addressing feature bias toward seen classes. In this article, an innovative GZSFD framework for chillers based on cross-modal information compression is proposed to overcome this difficulty. This method constructs a cross-modal feature fusion attention (CmFFA) and information compression module, and integrates them into a variational autoencoder with generative adversarial network (VAEGAN), and establishes a CmFFA-VAEGAN network to synthesize high-quality unseen fault samples. Specifically, the CmFFA module effectively aligns local regions of the virtual samples with key words in the textual attributes, enabling the model to attend to fine-grained semantic information and robustly fuse features from different modalities. This facilitates the generation of virtual samples that more accurately reflect the characteristics of the target modality. In addition, an information bottleneck (IB) layer is introduced at the output of the generator to compress redundant information within the fused features, retaining only the key information most relevant to the data augmentation task. This design enhances cross-modal consistency between the synthesized samples and their corresponding attributes, while alleviating the feature shift of unseen class samples toward seen categories. Extensive experiments are designed and executed on the chiller dataset. Experimental results demonstrate that the proposed framework effectively mitigates the bias of synthesized unseen fault samples toward seen categories, leading to a significant improvement in diagnostic accuracy.}, archive = {J_EAAI}, author = {Kexin Jiang and Xuejin Gao and Huayun Han and Huihui Gao and Yongsheng Qi and Xi Zhang and Liang Zhao}, doi = {10.1016/j.engappai.2025.112177}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112177}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generalized zero-shot fault diagnosis method for chillers based on cross-modal information compression}, volume = {161}, year = {2025}, }
• (2025). Enhancing rotation you only look once version 8 for accurate detection of arbitrarily-oriented and multi-scale construction material in complex environment. EAAI, 161, 112176. (www)

  @article{ , abstract = {Construction material detection is integral to effective material management. While deep learning-based methods have advanced automatic detection, challenges such as arbitrarily oriented objects, similar features, large object scale variations, and noise interference still limit accuracy. This study proposes an enhanced detection method based on improved rotation you only look once version 8, incorporating three key innovations. First, the large selective kernel network employs a spatial selection mechanism to dynamically adjust the network's receptive field, capturing key features and contextual information of various materials in complex construction scenarios. This enhances object detection performance in feature-similar environments. Second, the poly kernel inception network combines non-dilated multi-scale convolutions to extract dense texture features from differently sized objects, while using contextual anchor attention to capture long-range information for small objects. These components work together to improve multi-scale object detection. Third, the rectangular self-calibration module uses horizontal and vertical pooling to model rectangular attention regions, capturing axial global context. Its shape self-calibration function adjusts these regions to better fit arbitrarily oriented construction materials, enhancing focus on objects while suppressing noise interference. Experimental results show a mean average precision of 0.871–2.7 % higher than the baseline model and surpassing other state-of-the-art methods—while maintaining real-time detection speed at 28.3 frames per second. The proposed method improves material detection accuracy in complex construction environments, enabling refined material management on-site. This supports material entry and exit tracking, on-site usage monitoring, inventory management, and procurement planning, while also strengthening lean control of construction progress and costs.}, archive = {J_EAAI}, author = {Yujie Lu and Yuanjun Nong and Dayu Zhu and Shuo Wang}, doi = {10.1016/j.engappai.2025.112176}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112176}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing rotation you only look once version 8 for accurate detection of arbitrarily-oriented and multi-scale construction material in complex environment}, volume = {161}, year = {2025}, }
• (2025). Long-term forecasting of monthly reservoir inflow using deep and machine-learning-based algorithms. EAAI, 161, 112175. (www)

  @article{ , abstract = {This study provides a method for dynamic flow forecasting to detect how to move between periods with high and low-flow peaks. In other words, the model continues by gaining a deep insight into the long-term stable changes of the time series. Four-time horizons (5, 10, 15 and 20 years) were chosen to forecast the streamflow. These time horizons represent the ability of this method to detect the continuation of its movement in the short, medium, and long term. In this study, a dynamic and recurrent cycle is used to forecast river streamflow based on long-short-term memory (LSTM) and artificial neural network (ANN) models. The 12-month time delay of precipitation and streamflow of the Zayanderud river from 1970 to 2019 was used as input variables for training and validation of the models. In each stage of the forecasting process, 1 year (12 months) is forecasted and used as input for the next stage. The results show that the ANN model performs better in shorter time horizons and the LSTM model in longer time horizons, respectively. The best performance of the ANN model occurs in the 10-year forecast (Mean Absolute Percentage Error (MAPE) = 44.3, Nash–Sutcliffe efficiency coefficient (NSCE) = 0.58, Root Mean Square Error (RMSE) = 43.75) and the best performance of the LSTM model occurs in the 15-year forecast (MAPE = 48.9, NSCE = 0.54, RMSE = 47.12). Finally, the selected period is used to forecast the streamflow in future periods when observational data are not available.}, archive = {J_EAAI}, author = {Bahram Ghenaati and Mohammad Reza Nikoo and Mohammad G. Zamani}, doi = {10.1016/j.engappai.2025.112175}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112175}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Long-term forecasting of monthly reservoir inflow using deep and machine-learning-based algorithms}, volume = {161}, year = {2025}, }
• (2025). Characteristics detection of surface agglomerated abrasive on electroplated diamond wire: A measurement method based on semantic segmentation. EAAI, 161, 112173. (www)

  @article{ , abstract = {Electroplated Diamond Wire (EDW) serves as the primary tool for cutting semiconductor materials. Abrasive agglomeration on its surface induces sudden fluctuations in sawing force, exacerbates material surface damage, and shortens tool lifespan. However, existing testing methods primarily focus on abrasive counts while generally lacking effective means to quantify key characteristics of agglomerated abrasives (e.g., density, maximum size, and number of abrasives contained), leading to substantial limitations in the EDW quality assessment system. To address this critical gap, this study proposes a U-shape Network (UNet)-based semantic segmentation model for EDW surface abrasive grain segmentation, termed Electroplated Diamond Wire U-shape Network (EDW-UNet). By integrating Atrous Spatial Pyramid Pooling (ASPP) and Mamba-like Linear Attention (MLLA), EDW-UNet enhances the detection of multi-scale targets and the accuracy of complex boundary segmentation, enabling precise segmentation of both single and agglomerated abrasives on EDW surfaces. Experimental results demonstrate that EDW-UNet achieves a mean Intersection over Union of 0.917 and a mean Pixel Accuracy of 0.961 on the test set, outperforming the baseline UNet and other mainstream models, with an inference speed of up to 12.82 frames per second. For the first time, a feature extraction algorithm developed from the segmentation results enables measurement of key parameters, including the density, maximum size, and number of abrasives contained in agglomerates. This study not only addresses the technical gap in quantitative detection of agglomerated abrasive features but also provides a novel high-precision, quantitative standard for optimizing EDW manufacturing processes and assessing product quality, thereby enhancing the cutting efficiency and yield of semiconductor materials. The code is available at: https://github.com/huangwenbin123/EDW-UNet .}, archive = {J_EAAI}, author = {Wenbin Huang and Yufei Gao and Shengtan Hu and Dameng Cheng}, doi = {10.1016/j.engappai.2025.112173}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112173}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Characteristics detection of surface agglomerated abrasive on electroplated diamond wire: A measurement method based on semantic segmentation}, volume = {161}, year = {2025}, }
• (2025). Risk-response coupling in underground structures under liquefiable soil conditions: A causality-informed dynamic bayesian network integrated framework. EAAI, 161, 112171. (www)

  @article{ , abstract = {Underground structures in liquefiable soils face complex seismic risks that can trigger cascading failures. This study proposes a Granger causality-informed Dynamic Bayesian network (G-DBN) framework to capture the temporal propagation of seismic risk in such systems. Firstly, a system risk assessment model integrates multiple performance indicators through Cloud Model (CM) to quantify overall risk levels, considering uncertainties associated with soil liquefaction and structural responses. Subsequently, a structural dynamic risk inference model is established using Dynamic Bayesian network (DBN), combining Granger Causality (GC) analysis with engineering-informed relationships to define the network structure. The input features include key structural state variables such as tunnel cross-section convergence ( T 4 ), tunnel uplift displacement ( T 6 ), station uplift displacement ( S 5 ), and inter-story drift angle ( S 6 ), and the aggregated structural risk indicator serves as the target variable. This framework enables the temporal propagation of risk across interconnected structural nodes, and elucidates the mechanisms by which liquefiable soil deformations and structural responses interact within the soil-structure system. Results showed that the risk characteristic value (Expectation , E x ) decreased from 29.12 % (percentage) to 5.21 % as the Peak Ground Acceleration (PGA, expressed in units of gravitational acceleration g) increased from 0.1 g to 0.7 g. The proposed G-DBN model demonstrates robust predictive capabilities, achieving coefficient of determination ( R 2 ) values exceeding 0.95 across multiple seismic intensity conditions. Additionally, tunnel cross-section convergence ( T 4 ) was identified as the most critical factor affecting risk propagation in the coupled underground systems. By integrating holistic risk quantification with dynamic propagation analysis, this study offers a robust tool for understanding dynamic risk evolution and supports decision-making for seismic resilience of underground infrastructure in liquefiable soils.}, archive = {J_EAAI}, author = {Heqi Kong and Xiaohua Bao and Jun Shen and Xiangcou Zheng and Xiangsheng Chen}, doi = {10.1016/j.engappai.2025.112171}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112171}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Risk-response coupling in underground structures under liquefiable soil conditions: A causality-informed dynamic bayesian network integrated framework}, volume = {161}, year = {2025}, }
• (2025). Neural operator-enabled forward and inverse modeling of laser-induced surface acoustic waves and applications in nondestructive evaluation. EAAI, 161, 112170. (www)

  @article{ , abstract = {Laser-induced surface acoustic wave (SAW)-driven nondestructive evaluation offers high-resolution, non-contact characterization of subsurface microstructures. However, its practical application is often limited by the high computational costs associated with traditional numerical simulation methods. Recently, machine learning has emerged as an attractive alternative to accelerate these simulations. In this paper, we develop a neural operator-enabled framework for both forward and inverse modeling of laser-induced SAW propagation. A general dataset with randomly generated subsurface structures is used to evaluate and quantify the model's performance in both wave propagation and subsurface inversion problems. Three potential applications are then investigated: subsurface crack localization, multilayer structure characterization and polycrystalline grain imaging. The results demonstrate that the neural operator-enabled model achieves satisfactory accuracy even in the presence of noise and source waveform variations, underscoring its potential as an efficient and accurate surrogate model for practical nondestructive evaluation using laser-induced SAWs.}, archive = {J_EAAI}, author = {Zaiwei Liu and Zhongqing Su}, doi = {10.1016/j.engappai.2025.112170}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112170}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural operator-enabled forward and inverse modeling of laser-induced surface acoustic waves and applications in nondestructive evaluation}, volume = {161}, year = {2025}, }
• (2025). A multi-objective double Q-learning-based hyper-heuristic algorithm for aluminum production and transportation integrated scheduling problem. EAAI, 161, 112169. (www)

  @article{ , abstract = {The cooperative process of electrolytic aluminum production and aluminum liquid distribution is a typical distributed integrated scheduling problem. The aluminum production and transportation integrated scheduling problem (APTISP) is a strong coupling problem that is studied with the optimal objectives of total completion time and total energy consumption. A mathematical model is constructed for the APTISP. A multi-objective double Q-learning-based hyper-heuristic (MDQHH) algorithm is designed to solve the APTISP. In the MDQHH algorithm, a heuristic rule-based initialization operation is designed to generate the initial population of the APTISP. Nine problem-specific low-level perturbation heuristics are constructed to explore the APTISP solution space. A dynamically adjustable ε -greedy strategy is designed to select the low-level perturbation heuristics, and ε gradually decreases as the learning process progresses. Two Q-tables are utilized alternately to select actions and update the Q-values based on the current solution in each iteration. The experimental results show that the MDQHH algorithm outperforms several state-of-the-art comparison algorithms in addressing the performance of APTISP.}, archive = {J_EAAI}, author = {Fuqing Zhao and Ting Yang and Tianpeng Xu and Jianlin Zhang}, doi = {10.1016/j.engappai.2025.112169}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112169}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-objective double Q-learning-based hyper-heuristic algorithm for aluminum production and transportation integrated scheduling problem}, volume = {161}, year = {2025}, }
• (2025). Scalable multi-agent reinforcement learning for factory-wide dynamic scheduling in semiconductor manufacturing. EAAI, 161, 112168. (www)

  @article{ , abstract = {Real-time dynamic scheduling in modern manufacturing is highly complex due to frequent disturbances and intricate operational constraints. While reinforcement learning (RL) has shown promise, existing approaches often rely on extensive dispatching rules and struggle to scale to factory-wide settings. This study introduces a scalable multi-agent RL (MARL) framework with a leader–follower structure, enabling decentralized agents to handle sub-problems while maintaining global coordination through abstract goals. To further enhance robustness, a limited rule-based conversion algorithm is proposed to mitigate performance degradation from poor agent decisions. Our experimental results demonstrate that the proposed model outperforms the state-of-the-art deep RL-based scheduling methods in various aspects. Additionally, the proposed model provides the most robust scheduling performance to demand changes. Overall, the proposed MARL-based scheduling model presents a promising solution to the real-time scheduling problem, with potential applications in various manufacturing industries.}, archive = {J_EAAI}, author = {Jaeyeon Jang and Diego Klabjan and Han Liu and Nital S. Patel and Xiuqi Li and Balakrishnan Ananthanarayanan and Husam Dauod and Tzung-Han Juang}, doi = {10.1016/j.engappai.2025.112168}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112168}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Scalable multi-agent reinforcement learning for factory-wide dynamic scheduling in semiconductor manufacturing}, volume = {161}, year = {2025}, }
• (2025). Research on global path planning algorithm based on indoor map partition preprocessing. EAAI, 161, 112167. (www)

  @article{ , abstract = {Mobile robots utilize Simultaneous Localization and Mapping (SLAM) technology to generate environmental maps and determine their locations within these environments. Subsequently, they employ path planning algorithms to complete navigation tasks. Although existing path planning algorithms are relatively mature, they still exhibit inefficiencies in complex indoor environments. To address this issue, this paper introduces an Indoor Map Partitioning Preprocessing (IMPP) algorithm, which identifies and segments irregularly shaped, complex rooms to accelerate the path planning process. The method initially utilizes the Robot Operating System (ROS) to construct an indoor map dataset and subsequently applies an image segmentation model to identify and enclose rooms. By combining image processing techniques with path planning algorithms, this method can obtain room index information and successfully exclude irrelevant areas from the path planning process. Ultimately, the IMPP algorithm is integrated with a variety of global path planning algorithms. Experimental results demonstrate that in complex indoor environments, this method significantly surpasses existing partitioning methods in terms of room recognition accuracy. Moreover, it decreases the number of expansion points in global path planning algorithms, significantly enhancing processing speed and efficiency.}, archive = {J_EAAI}, author = {Jifan Yang and Xiaoling Li and Xiaoyang Liu and Xunding Pan and Lei Wang}, doi = {10.1016/j.engappai.2025.112167}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112167}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on global path planning algorithm based on indoor map partition preprocessing}, volume = {161}, year = {2025}, }
• (2025). An ensemble transfer learning strategy for short-term electricity load forecasting of data-sparse buildings. EAAI, 161, 112166. (www)

  @article{ , abstract = {The “data silos” caused by the lack of available historical or shared information is an important reason affecting forecasting performance of building’s short-term power demand, thus limits the flexibility of demand response of building group. Knowledge’s transfer learning (TL) from similar domains has been proved an efficient tool to enhance target prediction’s performance. To improve data-sparse building’s energy forecasting accuracy and robustness simultaneously, an ensemble strategy based on different TL models is proposed in this study. Firstly, a two-stage similar data selection method is proposed for source domain construction. The multi-kernel maximum mean discrepancy (MK-MMD) is used for selecting similar buildings to the target one; the nearest neighbor search (NNS) method is adopted for picking the most helpful data from the similar buildings to form the dataset of source domain. To determine the optimal number of source domains for transfer learning, an objective function is also formulated. Then, two effective TL models, i.e., instance-based TL (IBTL) and model-based TL (MBTL) are both built as sub models. On this basis, a parallel ensemble framework is designed with model weights adjusted by multiple linear regression. For case study, a public data set containing 1,449 buildings is treated as candidate source domain set for knowledge transfer, and two educational buildings with sparse historical data are treated as target buildings for hourly electricity load forecasting. Forecasting results show that compared with the previous reported TL models, the proposed ensemble strategy always achieves the best performance in two cases and the prediction accuracies are improved by 41.8% and 29.3% (MAPE) respectively. The impact of source domain number and target data amount on prediction performance are also analyzed and discussed in details.}, archive = {J_EAAI}, author = {Xingqiao Liu and Borui Wei and Wenping Xue and Xiaoying Li and Kangji Li}, doi = {10.1016/j.engappai.2025.112166}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112166}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An ensemble transfer learning strategy for short-term electricity load forecasting of data-sparse buildings}, volume = {161}, year = {2025}, }
• (2025). Intelligent multispectral image recognition by using credibility-based square root cubature kalman filters and broad particle swarm learning. EAAI, 161, 112165. (www)

  @article{ , abstract = {This paper proposes intelligent multispectral image recognition by using credibility-based Square Root Cubature Kalman Filters (SRCKF) and broad particle swarm learning to address the issue of low recognition accuracy in multispectral target detection using Kalman filtering. This method initially enhances the estimation accuracy of SRCKF by integrating credibility theory. Secondly, the regression parameters of a Broad Learning System (BLS) are tuned with Particle Swarm Optimization (PSO), forming a PSO-BLS network that offers fast, lightweight feature learning. The PSO-BLS acts as an regressor that compensates residual errors produced by the credibility-based SRCKF, thereby refining the overall filtering precision. Subsequently, a particle swarm-optimized broad learning neural network is employed to rectify the SRCKF results from a compensatory standpoint, thereby enhancing the system’s filtering precision. Ultimately, the adaptive search window generated by Camshift(Continuously adaptive Mean shift) can be employed as an input for the credibility SRCKF, thereby facilitating the acquisition of more reliable real-time observation data. Finally, in the experimental part, the proposed algorithm is validated by the MNIST(Modified National Institute of Standards and Technology) dataset , CIFAR-10(Canadian Institute For Advanced Research) dataset, GTOT(Grayscale-Thermal Object Tracking) dataset and OTB2015(Object Tracking Benckmark 2015) dataset respectively for its generalization ability and robustness in static classification tasks and the effectiveness of dynamic target detection and tracking in real lighting scenes, Occlusion and motion blur scenes, while the performance is more advantageous than some mainstream algorithms.}, archive = {J_EAAI}, author = {Quanbo Ge and Zijian Xue and Hao Wang and Yuanliang Wang}, doi = {10.1016/j.engappai.2025.112165}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112165}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent multispectral image recognition by using credibility-based square root cubature kalman filters and broad particle swarm learning}, volume = {161}, year = {2025}, }
• (2025). Accidents analysis of mining industry through semantic text representation and dimensionality reduction: An integrated clustering framework. EAAI, 161, 112164. (www)

  @article{ , abstract = {Despite significant improvements in worker health and safety in recent decades, the mining industry still experiences fatal and non-fatal accidents. This underscores the critical need for a more nuanced understanding of accident causation patterns through accident data analysis. While conventional analytical approaches have yielded valuable insights, the extensive information embedded within text-based accident narratives remains underutilized. To address this gap, this study presents an artificial intelligence (AI)-based framework that integrates transformer-based natural language processing, nonlinear dimensionality reduction, and unsupervised machine learning to analyze and cluster accident narratives from the U.S. mining industry. Specifically, this study uses Sentence-BERT (SBERT), a sentence embedding model based on Bidirectional Encoder Representations from Transformers (BERT), to extract the high-dimensional semantical representation of accident narratives. These embeddings are then mapped to a low-dimensional space using the Uniform Manifold Approximation and Projection (UMAP) technique, followed by clustering with the k-means algorithm and subsequent hazard-focused cluster analysis. The primary contribution to AI lies in demonstrating the effectiveness of combining modern sentence embedding techniques with dimensionality reduction and clustering for the semantic analysis of safety-related narratives. From an engineering standpoint, this framework enables the identification of latent accident patterns that can inform hazard detection and guide safety interventions in the mining industry. The resulting clusters reveal diverse accident patterns across mining operations. In clusters associated with underground mining, a high proportion of incidents (ranging from 84 to 98 %) involved no equipment, with distinct injury patterns: torso injuries (67 %) from over-exertion, lower extremity injuries (58 %) from slips/falls, and upper extremity injuries from over-exertion (95 %) and material handling (85 %). Equipment-related clusters revealed strong associations with drilling tools (92 %), loaders (98 %), and bolting equipment (96 %). Clusters associated with strip/quarry/open pit operations exhibited a high frequency of vehicle-related accidents (98 % transportation, 99 % loaders), often resulting in multiple body part injuries. Milling operation clusters show 52–97 % no-equipment accidents, with injury patterns similar to those in underground mining. Additionally, noise-induced hearing loss (96–97 %) was observed in clusters spanning all mining operation types. These findings offer actionable insights for safety professionals and support data-driven, targeted interventions in the mining industry.}, archive = {J_EAAI}, author = {Abid Ali Khan Danish and Snehamoy Chatterjee and Kumar Vaibhav Raj and Rennie Kaunda and Hugh Miller and Aref Majdara}, doi = {10.1016/j.engappai.2025.112164}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112164}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Accidents analysis of mining industry through semantic text representation and dimensionality reduction: An integrated clustering framework}, volume = {161}, year = {2025}, }
• (2025). A lightweight and improved you only look once model using GhostWise convolution and attention mechanism for accurate plant disease detection. EAAI, 161, 112163. (www)

  @article{ , abstract = {Artificial intelligence enables early plant disease detection to support food security, yet achieving high accuracy with low computational cost on multi-class datasets remains challenging. To bridge this gap and solve the problem, we propose an improved version of the “You Only Look Once version 8 nano” (YOLOv8n) model that integrates a convolutional block attention module and custom coarse-to-fine modules into the baseline YOLOv8n architecture. To improve both efficiency and detection performance, the proposed model incorporates a combination of Ghost module and Depthwise convolution for lightweight feature extraction. Additionally, the attention modules emphasize important spatial and channel-wise features, helping the model focus better on critical information. The performance of the proposed model is compared with YOLO models including YOLOv5n, YOLOv5s, YOLOv7, YOLOv8n, and YOLOv8s. The experimental results show that the improved model outperforms the benchmark YOLO models. In addition, the improved model achieves mean Average Precision at Intersection over Union threshold of 0.5 ( mAP@0.5 ) 0.742, representing an improvement of 3.37 % over the baseline YOLOv8n. For the metric mean Average Precision at thresholds from 0.5 to 0.95 ( mAP@0.5:0.95 ), the improved model shows an improvement of 5.13 % over YOLOv8n. The precision measure reveals an improvement of 8.99 % and F1 score of 0.6856 outperforms YOLOv8n by 5.85 %. In addition, generalization experiments are conducted using commonly preferred plant disease datasets. These results validate the robustness, accuracy and stability of the improved model in effectively detecting plant leaf diseases.}, archive = {J_EAAI}, author = {Bunyamin Dikici and Mehmet Fatih Bekciogullari and Hakan Acikgoz and Serkan Ozbay}, doi = {10.1016/j.engappai.2025.112163}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112163}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight and improved you only look once model using GhostWise convolution and attention mechanism for accurate plant disease detection}, volume = {161}, year = {2025}, }
• (2025). Hybrid optimization approach for syngas-fueled gas turbines: Integrating inverse problem solving and machine learning techniques. EAAI, 161, 112162. (www)

  @article{ , abstract = {This study investigates the efficiency of gas turbines powered by syngas derived from gasification, combining a detailed thermodynamic model with a hybrid optimization approach that integrates inverse problem solving via the Levenberg-Marquardt (LM) algorithm and supervised machine learning using Gradient Boosted Trees (GBT). The LM method is used to estimate turbine and compressor performance parameters based on efficiency maps extracted through a custom image processing routine. These maps are applied to calibrate isentropic efficiencies and define the system's operational boundaries. The resulting performance curves enhance the accuracy of the thermodynamic cycle model. To identify optimal operating conditions that maximize thermal efficiency and minimize entropy generation, the inverse problem formulation is integrated with the GBT model, enabling data-driven optimization based on simulated system behavior. All modeling and simulations are conducted in Wolfram Mathematica version 14.0, while verification is performed with the commercial software Gasturb version 14.0, using syngas composition data available in the literature. As a case study to demonstrate the method's applicability, syngas obtained from sewage sludge gasification is analyzed. The findings after optimization indicated an average thermal efficiency of 40 % when using syngas. The analysis revealed that the fuel mass flow rate contributes approximately 45 % to the efficiency gains and more than 70 % to the exergy reduction, while the excess air ratio contributes around 50 % to the cycle's efficiency. The study demonstrates the value of integrating LM-based modeling and GBT optimization, indicating significant potential for enhancing the performance of syngas-fueled turbines. Moreover, it highlights syngas from sewage sludge as a promising sustainable energy source.}, archive = {J_EAAI}, author = {Hiago David Zogbi Silva Oliveira and Vítor Caldeira de Andrada Bastos and York Castillo Santiago and Isabela Florindo Pinheiro}, doi = {10.1016/j.engappai.2025.112162}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112162}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid optimization approach for syngas-fueled gas turbines: Integrating inverse problem solving and machine learning techniques}, volume = {161}, year = {2025}, }
• (2025). Frequency-wavelet adaptive basis network for long-term time series forecasting. EAAI, 161, 112161. (www)

  @article{ , abstract = {Transformer-based models have demonstrated remarkable efficacy in long-term time series forecasting. Nevertheless, their fixed-scale feature extraction and static spectral processing limit adaptability to inherent multi-scale temporal dependencies. Additionally, conventional self-attention mechanisms struggle to effectively model hierarchical temporal structures. To address these limitations, we propose the Frequency-Wavelet Adaptive Basis Network (FWBNet), a novel framework that integrates three synergistic components including a Multi-scale Adaptive Basis Module (MS-ABM) for multi-resolution feature extraction, a Dual-Stream Wavelet Cross-Attention (DWCA) system with learnable wavelet transformations, and a computationally efficient Frequency-domain Multi-Layer Perceptron bottleneck (FreMLP bottleneck) for selective spectral processing. The MS-ABM innovatively integrates multi-scale features of time series through an adaptive basis function selection mechanism, dynamically fusing characteristics at different scales to form hierarchical representations. These multi-dimensional representations are then processed by the DWCA system, which establishes an efficient modeling framework for cross-scale temporal dependencies by combining bi-directional attention with learnable wavelet transforms. Building upon this foundation, the FreMLP bottleneck layer strategically performs selective feature optimization in the frequency domain, extracting key frequency information through spectral sparsification techniques to form an end-to-end multi-scale temporal feature learning pipeline. Experimental evaluations on six benchmark datasets demonstrate that FWBNet significantly outperforms state-of-the-art models, achieving mean squared error (MSE) reductions of 14% for multivariate and 21% for univariate predictions. This study presents a comprehensive framework for effective multi-scale temporal information integration in complex time series forecasting applications.}, archive = {J_EAAI}, author = {Qiang Lai and Yang You}, doi = {10.1016/j.engappai.2025.112161}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112161}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Frequency-wavelet adaptive basis network for long-term time series forecasting}, volume = {161}, year = {2025}, }
• (2025). Balanced sample repository for knowledge distillation in data-free image classification scenario. EAAI, 161, 112160. (www)

  @article{ , abstract = {To address strict data security and privacy requirements in the field of knowledge distillation for image classification, generative network-based data-free knowledge distillation produces fake images for training student models to perform image classification tasks. However, we discover that the datasets consisting of those fake images often suffer from imbalances, such as unclear class margins, uneven intra-class distribution, and skewed label distribution, impeding classification accuracy. To this end, this paper proposes the balanced sample repository method for data-free knowledge distillation in the context of image classification, comprising three components: dynamic pseudo-supervised loss weight (DLW), biased label generator (BLG), and intra-class homogeneity operator (IHO). Specifically, DLW adjusts the pseudo-supervised loss weight to maintain distinct class margins, BLG uses the roulette algorithm to balance label distribution, and IHO employs local outlier factors to remove outliers and homogenize intra-class distributions. The fake images generated by the generative network are initially stored in the sample repository, and those three components achieve a balanced dataset by influencing this sample repository. The experiments conducted across five recognized image classification datasets demonstrate that the proposed method achieves competitive results and produces more balanced samples.}, archive = {J_EAAI}, author = {Yafeng Sun and Xingwang Wang and Junhong Huang and Shilin Chen}, doi = {10.1016/j.engappai.2025.112160}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112160}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Balanced sample repository for knowledge distillation in data-free image classification scenario}, volume = {161}, year = {2025}, }
• (2025). A lightweight detection model for green navel oranges in natural environments. EAAI, 161, 112157. (www)

  @article{ , abstract = {In natural environments, the detection of green navel oranges in the unstructured plantings of Gannan navel oranges presents challenges, including low model accuracy and excessive parameter size. This paper integrates existing technologies to propose a lightweight green tangerine detection model based on You Only Look Once version 5s (YOLOv5s). The aim is to improve the accuracy of green tangerine identification and detection in natural environments through model improvements. Firstly, the original YOLOv5 network model is optimized by replacing the original CSPDarknet53 (Cross Stage Partial Darknet53) backbone network with MobileNetV3 in order to reduce model parameters. Concurrently, the Squeeze-and-Excitation (SE) module in the network is substituted with the Convolutional Block Attention Module (CBAM), resulting in a lightweight network. Secondly, CBAM is integrated into the C3 network (cross-stage partial bottleneck with three convolutional layers) in the neck region to enhance the model's ability to extract green navel orange features after lightweighting. Finally, improve the loss function of the model to reduce the positioning error of the model for low-quality targets. The experimental results demonstrate that the enhanced model attains a 3.6 % enhancement in precision, 2.3 % in recall, and 2.1 % in mean Average Precision (mAP) in comparison with the original YOLOv5 model. The model's size is 3.9 megabytes (MB), representing a reduction of 72.9 % from the original 14.4 MB. Research shows that our method provides an efficient and accurate solution for identifying green navel orange targets in complex environments, with potential application value in agricultural automation for orchard management and monitoring.}, archive = {J_EAAI}, author = {Dongyuan Zhang and Qiusheng Li and Xindi Yuan}, doi = {10.1016/j.engappai.2025.112157}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112157}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight detection model for green navel oranges in natural environments}, volume = {161}, year = {2025}, }
• (2025). Efficient real-time instance segmentation of garment for intelligent robot tie-dye based on you only look once version 11 network. EAAI, 161, 112156. (www)

  @article{ , abstract = {Tie-dye, as an intangible cultural heritage with a long history, has gained global popularity due to its unique artistic value and versatile craftsmanship. However, its reliance on manual operation leads to low production efficiency and inconsistent quality. To address this, we propose an improved You Only Look Once version-11 (YOLOv11) instance segmentation model, Garment-YOLO, to efficiently segment the garment regions of robotic tie-dye by artificial intelligence technology. First, the C3k2_DynamicMixFormer (C3k2_DMF) module introduces dynamic kernel weight selection mechanism and multi-scale fusion to effectively balance local details and global information. Meanwhile, the Dual-Cross Recalibration Feature Pyramid Network (DCR-FPN) is proposed to enhance the detail preservation of edge region by selectively aggregating semantic and boundary information. Furthermore, the Superior-Head replaces depthwise convolution (DWConv) with part convolution (PConv), significantly reducing model complexity while maintaining performance. Experimental results demonstrate that Garment-YOLO achieves 167.4 frames per second (FPS) and maintains an optimal trade-off between inference speed and segmentation accuracy, reaching 91.9 % mean average precision (mAP)50–95 (Box), 91.5 % mAP50-95 (Mask). To further validate its practical performance, we conducted a 72-h production-line comparison, showing that compared to the baseline, we increase the product qualification rate by 24.6 %, reduce dye waste by 19.4 %, and increase the total production by 52.5 % compared to manual tie-dye. This study not only provides a feasible solution for the intelligent transformation of traditional tie-dye but also contributes to the preservation and development of this intangible cultural heritage. The code will be released on GitHub ( https://github.com/fudifu123/GARMENT-YOLO ). A video that intuitively introduces our research ( https://www.bilibili.com/video/BV13MgczAEkB ).}, archive = {J_EAAI}, author = {Difei Feng and Qihong Zhou and Lei Xiao and Kunfeng Ge and Hangzhou Ma and Li Zhou}, doi = {10.1016/j.engappai.2025.112156}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112156}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient real-time instance segmentation of garment for intelligent robot tie-dye based on you only look once version 11 network}, volume = {161}, year = {2025}, }
• (2025). A deep patch network with spatiotemporal meta-parameter learning for soft sensor modeling of industrial processes. EAAI, 161, 112155. (www)

  @article{ , abstract = {Soft sensing techniques have been widely employed in industrial processes to predict hardly measurable quality variables using easily measurable process variables. Effective modeling strategies for representing spatiotemporal features are highly desired in soft sensing applications for dynamic industrial processes. Deep networks, such as graph convolutional network and gated recurrent unit, are frequently employed to capture the spatial or temporal characteristics of sequential data. Nevertheless, effectively modeling and utilizing spatiotemporal heterogeneity remains a challenging problem. Thereby, a p atch network with spatiotemporal m eta- p arameter l earning (PMPL) is proposed in this paper, which presents an uniform framework of modeling spatiotemporal heterogeneous feature representations for soft sensing. In PMPL, a patching strategy is utilized to segment times series into sub-series-level patches, allowing the retention of local semantic information while capturing long-term dependencies. Next, a novel spatiotemporal meta-parameter learning approach is proposed to generate specific parameters for a spatial encoder, temporal encoder and spatiotemporal decoder through establishing corresponding embedding layers. In this way, the spatial, temporal and spatiotemporal dependencies within the process data can be adaptively captured and comprehensively refined from a uniform perspective. Extensive experiments are conducted based on two real production processes, which illustrate the feasibility and efficacy of the proposed model.}, archive = {J_EAAI}, author = {Xudong Shi and Kangping Du and Weili Xiong and Humberto Morales and Adriana Amicarelli}, doi = {10.1016/j.engappai.2025.112155}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112155}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep patch network with spatiotemporal meta-parameter learning for soft sensor modeling of industrial processes}, volume = {161}, year = {2025}, }
• (2025). Deep bioinspired evolutionary stacking algorithm for unpaired multimodal cell classification calibration. EAAI, 161, 112153. (www)

  @article{ , abstract = {Single-modality deep learning approaches for cell image classification exhibit inherent limitations in informational diversity when processing cross-institutional datasets acquired under varied imaging protocols. In contrast, multimodal cell imaging has emerged as a promising alternative for addressing data heterogeneity through comprehensive information integration. This study introduces a novel multimodal alternate training decision-making architecture based on a stacking algorithm for unpaired multimodal cell classification calibration. The method leverages deep bioinspired evolutionary networks combined with kernel-based support vector machines. Specifically, deep base classifiers incorporating multimodal concepts are derived from heterogeneous cell datasets. Each base classifier employs a bioinspired strategy to perform alternate training between two evolutionary densely connected attention networks. To mitigate class imbalance, where diseased cells are significantly outnumbered by normal cells, we incorporate a Shannon entropy loss term. Finally, multiple kernel-based support vector machines serve as meta classifiers, transforming high-level multimodal concepts into a separable feature space for robust decision-making. Experimental results demonstrate the superiority of our approach over existing algorithms for unpaired multimodal cell image classification. Our findings emphasize the importance of alternate training intra-modality classifiers and inter-modality fusion calibration for accurate and reliable medical image classification. All source code for this work will be publicly available on GitHub.}, archive = {J_EAAI}, author = {Lili Zhao and Di Xu and Xueping Tan and Jinzhao Yang and Weiping Ding and Hengde Zhu and Lichi Zhang and Qian Wang}, doi = {10.1016/j.engappai.2025.112153}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112153}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep bioinspired evolutionary stacking algorithm for unpaired multimodal cell classification calibration}, volume = {161}, year = {2025}, }
• (2025). Temporal-frequency joint hierarchical transformer with dynamic windows for speech emotion recognition. EAAI, 161, 112152. (www)

  @article{ , abstract = {Speech Emotion Recognition (SER) aims to identify the emotional state of a speaker from speech signals, serving as a critical prerequisite for achieving natural human–computer interaction. In speech signals, emotional information is inherently distributed across diverse frequency bands and temporal scales, with emotional cues in distinct regions exhibiting varying levels of heterogeneity or interdependence. Existing Transformer-based methods face limitations in precisely localizing salient temporal-frequency regions and modeling their inter-regional relationships. To address these challenges, a temporal-frequency joint hierarchical Transformer with dynamic window mechanisms, abbreviated as TF-DWFormer, is proposed to capture critical emotional cues and their contextual dependencies across temporal-frequency dimensions. It operates through several principal phases: Firstly, a feature reconstruction module is designed to extract temporal, frequency, and temporal-frequency representations of emotional speech. Secondly, a high-low frequency-based emotion-aware partitioning strategy is designed to achieve the division of emotional regions. Thirdly, a local window within a hierarchical Transformer analyzes static intra-region correlations to capture fine-grained emotional patterns, while a dynamic window adaptively models temporal evolution across regions, learning dynamic inter-region relationships. Lastly, a dual-cross-attention mechanism is employed to synergize comprehensive emotion representation from different domains. Our evaluation experiments demonstrate that the proposed TF-DWFormer method achieves recognition accuracies of 73.68%, 91.67%, 92.59%, 74.42%, and 50.54% on the datasets IEMOCAP, CASIA, EMODB, eNTERFACE05, and MELD, respectively, outperforming existing SER methods. These results confirm the capability of TF-DWFormer to precisely localize salient regions, robustly model inter-region dependencies, and effectively fuse multi-domain information, providing a promising solution for advancing SER technology.}, archive = {J_EAAI}, author = {Yonghong Fan and Heming Huang and Huiyun Zhang and Ziqi Zhou}, doi = {10.1016/j.engappai.2025.112152}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112152}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Temporal-frequency joint hierarchical transformer with dynamic windows for speech emotion recognition}, volume = {161}, year = {2025}, }
• (2025). Three-dimensional reconstruction and fracture segmentation based on X-ray and computed tomography paired dataset. EAAI, 161, 112151. (www)

  @article{ , abstract = {In some orthopedic surgeries, the use of three-dimensional (3D) computed tomography (CT) scanning technology is not feasible due to scene limitations, leaving doctors to rely on two-dimensional (2D) X-ray images for real-time diagnosis. However, X-ray images lack 3D information, making accurate diagnosis challenging. Developing an algorithm to convert 2D X-ray images into 3D CT images, while simultaneously combining high-quality 3D reconstruction with precise fracture segmentation, offers a promising solution to the problem. In this study, we propose a novel artificial intelligence (AI)-driven framework named 3D reconstruction and segment anything model (3DRecSAM). The reconstruction image enhancer (RIE) is designed to achieve high-precision 3D reconstruction and provide high-quality feature initialization for fracture segmentation. Meanwhile, the mamba segment anything model (MSAM), based on the segment anything model (SAM) architecture, is developed for accurate fracture segmentation. We introduce a Kolmogorov–Arnold network (KAN)-based attention fusion module (KAF), which facilitates the joint optimization of the RIE reconstruction network and the MSAM segmentation network. Furthermore, the selective scanning mamba with KAN (SKM) is incorporated to enhance feature extraction for both RIE and MSAM. Mamba efficiently captures long-range dependencies and sequential patterns, while KAN’s learnable activation functions facilitate adaptive feature fusion and non-linear representation. To train and evaluate 3DRecSAM, we introduce the real X-ray and CT paired dataset (XCPData), which is publicly available on GitHub: https://github.com/YuanGao1201/XCPData .}, archive = {J_EAAI}, author = {Yuan Gao and Yuan Zhou and Da Chen and Jiachen Li and Mingle Zhou and Gang Li and Yunbo Gu and Jean-Louis Coatrieux and Yang Chen}, doi = {10.1016/j.engappai.2025.112151}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112151}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-dimensional reconstruction and fracture segmentation based on X-ray and computed tomography paired dataset}, volume = {161}, year = {2025}, }
• (2025). Unsupervised domain adaptation for lithology classification using dynamic entropy-based prototype learning. EAAI, 161, 112150. (www)

  @article{ , abstract = {Lithology classification plays a crucial role in geological exploration and resource evaluation. However, the significant distribution discrepancies between source and target domains, coupled with the unavailability of source domain data, pose substantial challenges to traditional domain adaptation methods. To address these challenges, we propose an innovative framework, Unsupervised Domain Adaptive Dynamic Entropy Prototype Learning (UDADEPL), which leverages a source-free unsupervised domain adaptation strategy for lithology classification. The UDADEPL framework consists of a frozen source pre-trained model and a trainable target model, incorporating a dynamic entropy-based prototype learning matrix for reliable sample selection and centroid-based pseudo-label learning for iterative optimization. Additionally, an information maximization loss and source domain regularization loss are integrated into a curriculum learning strategy to balance feature extraction and domain adaptation. This approach enables the model to effectively handle complex lithological boundaries and class imbalances in the target domain. Extensive experiments on datasets from the Tarim Oilfield and Hugoton–Panoma field demonstrate the superiority of UDADEPL over traditional machine learning and advanced deep learning models. UDADEPL achieves superior classification accuracy, outperforming the best baseline models, especially in cross-domain adaptation and complex lithology identification.}, archive = {J_EAAI}, author = {Hengxiao Li and Yahui Liu and Lu Liu}, doi = {10.1016/j.engappai.2025.112150}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112150}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised domain adaptation for lithology classification using dynamic entropy-based prototype learning}, volume = {161}, year = {2025}, }
• (2025). Transforming physics-informed machine learning to convex optimization. EAAI, 161, 112149. (www)

  @article{ , abstract = {Physics-Informed Machine Learning (PIML) offers a powerful paradigm of integrating data with physical laws to address important problems in engineering, such as parameter estimation, inferring hidden physics, equation discovery, and state prediction. However, PIML, such as Physics-Informed Neural Networks (PINNs), still lack the necessary accuracy, stability, and interpretability when applying in practical engineering due to many serious optimization challenges including the spectral bias, non-convex optimization, multi-objective optimization, and non-smooth optimization. In this study, we propose the Convex-PIML based on convex optimization and basis functions widely used in well-established numerical solvers to overcome all these limitations. The linear combination of B-splines is utilized to approximate the data, promoting the convexity of the loss function. By dividing variables into blocks and replacing the non-convex loss terms with convex approximations, the problem is further converted into a sequence of successively refined approximated convex optimization problems. This conversion known as Block Successive Convex Approximation (BSCA) allows the use of well-established convex optimization algorithms, obtaining solutions effectively and efficiently. Furthermore, an adaptive knot optimization method is introduced to mitigate the spectral bias issue of PIML, further improving the performance. The proposed fully adaptive framework by combining the adaptive knot optimization and BSCA is tested in scenarios with distinct types of physical prior. The results indicate that optimization problems are effectively solved in these scenarios, highlighting the potential of the framework for broad applications. Note that the Convex-PIML is also flexible since many other basis functions can also be incorporated to handle different systems.}, archive = {J_EAAI}, author = {Letian Yi and Siyuan Yang and Ying Cui and Zhilu Lai}, doi = {10.1016/j.engappai.2025.112149}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112149}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transforming physics-informed machine learning to convex optimization}, volume = {161}, year = {2025}, }
• (2025). Finite-time fuzzy tracking control for networked suspension systems under gilbert-elliott fading channels. EAAI, 161, 112148. (www)

  @article{ , abstract = {In this paper, we address the finite-time ℓ 2 − ℓ ∞ fuzzy tracking control problem for networked suspension systems operating under Gilbert-Elliott fading channels. Firstly, the suspension system is established based on interval type-2 fuzzy rules to effectively describe the nonlinearity and uncertainty inherent in the model. Secondly, the Gilbert-Elliott model is employed to characterize the random switching behavior of channel states (good channel and fading channel) under real-world conditions. The timely identification of the current channel state is achieved through a mode detection method. To address channel fading and communication constraints, a novel mode-dependent controller is proposed. Simultaneously, we aim to minimize unnecessary and redundant communication. The hybrid scheduling strategy of the FlexRay protocol is utilized to transmit the measurement output data to the observer, and a saturation function is introduced in the observer to minimize the impact of outliers. Subsequently, the design conditions for stochastic finite-time boundedness and ℓ 2 − ℓ ∞ performance of the augmented system is proposed. Finally, simulation results demonstrate that the proposed algorithm is practically viable in dealing with complex environments, such as Gilbert-Elliott fading channels, outliers, and communication constraints.}, archive = {J_EAAI}, author = {Li-Juan Cai and Xiao-Heng Chang and Xin-Yue Zhao}, doi = {10.1016/j.engappai.2025.112148}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112148}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Finite-time fuzzy tracking control for networked suspension systems under gilbert-elliott fading channels}, volume = {161}, year = {2025}, }
• (2025). RT-FedFlow: An efficient framework for real-time traffic signal optimization using federated multi-agent reinforcement learning. EAAI, 161, 112147. (www)

  @article{ , abstract = {The optimization of Traffic Signal Control (TSC) is a critical component of the Intelligent Transportation System (ITS) in smart cities, aiming to eliminate traffic congestion and improve urban mobility. However, traditional approaches to TSC often rely on centralized data processing, limiting scalability and adaptability to real-time traffic dynamics while raising concerns about data privacy, communication, and computation. To address these challenges, this paper proposes the RT-FedFlow, a novel federated multi-agent reinforcement-learning (FMARL) framework for real-time traffic signal optimization. The framework employs enhanced FMARL techniques to enable decentralized collaboration among multiple traffic signal agents, optimizing traffic flow while seamlessly balancing local adaptability and global coordination. Federated model aggregation enables agents to collaboratively learn a global policy while preserving local data privacy, with each reinforcement learning (RL) agent independently optimizing its policy. Next, an advanced reward mechanism is designed to minimize congestion and reduce waiting times. Moreover, a hierarchical communication and coordination strategy is adopted to facilitate efficient local interaction among agents and global model aggregation to improve overall system performance. In addition, the framework enables real-time optimization and prioritization of emergency vehicles through a dedicated preemption module. Compared to existing TSC models, RT-FedFlow significantly improves traffic signal coordination and response times by dynamically adapting to complex, real-time traffic conditions. The proposed RT-FedFlow framework is implemented and evaluated using the SUMO simulator. Extensive simulation results demonstrate that RT-FedFlow outperforms all benchmark models, achieving a 15.49 % reduction in average inference time, a 3.07 % reduction in average queue length, a 13.04 % reduction in average intersection delay, a 0.20 % improvement in fuel consumption, and a 5.35 % emergency vehicle clearance time.}, archive = {J_EAAI}, author = {Ayaz Akbar and Syed Sajid Ullah and Abdul Malik and Saeed Mian Qaisar}, doi = {10.1016/j.engappai.2025.112147}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112147}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {RT-FedFlow: An efficient framework for real-time traffic signal optimization using federated multi-agent reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). A hybrid data-driven model for state of health estimation of lithium-ion battery with capacity recovery. EAAI, 161, 112146. (www)

  @article{ , abstract = {The failure of lithium-ion batteries has attracted the attention of researchers. Monitoring the degradation process through sensor-based state-of-health (SOH) assessment enables early detection of anomalies and facilitates timely interventions to prevent catastrophic failures. During battery degradation, the phenomenon of capacity recovery makes it complicated to accurately evaluate the SOH. However, existing works usually neglect this phenomenon. In this paper, we propose a hybrid model focusing on capacity recovery. The model consists of four modules: a feature extraction module, a data decomposition module, a global trend assessment module, and a local fluctuation assessment module. Specifically, in the feature extraction module, a convolutional neural network extracts features from the input data that effectively characterize the capacity degradation process. The data decomposition module applies empirical mode decomposition (EMD) to separate local fluctuations caused by capacity recovery from the global degradation trend. These components are then input into their respective assessment modules. Additionally, the local fluctuation assessment module incorporates an attention mechanism that adaptively identifies the similarity between extracted features and health states, thereby enhancing the evaluation of local fluctuations. The proposed model is validated using the NASA (National Aeronautics and Space Administration) battery dataset, demonstrating its effectiveness in addressing the challenges of SOH prediction under capacity recovery conditions.}, archive = {J_EAAI}, author = {Yu Lin and Luo Zhou and Jianhai Yan and Shuguang He}, doi = {10.1016/j.engappai.2025.112146}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112146}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid data-driven model for state of health estimation of lithium-ion battery with capacity recovery}, volume = {161}, year = {2025}, }
• (2025). A novel localization algorithm integrated with sensor fusion for position estimation of mobile robot in the road following and roundabout environment. EAAI, 161, 112145. (www)

  @article{ , abstract = {A novel Artificial Intelligence algorithm based autonomous mobile robot localization in road environments is presented in this paper. The sensor fusion data, employing camera and odometry, are used to build the local maps for road environment. The localization of the robot is performed using a novel Artificial Intelligence algorithm, called Priori-Posteriori-Laser-Simulator, to find the position and heading-angle of robot along movement in two phases, namely priori and posteriori stages as 1st phase and Kalman filter in 2nd phase. In the priori stage, the path of the robot is planned in the built-map using Laser Simulator through generating a series of points as lines to drive the robot in the middle of the road environments. In the posteriori-stage, the position and robot heading angle are measured using sensor fusion system and the laser-simulator updates the position accordingly through a comparison between the planned and actual positions. The Kalman filter is used in 2nd phase to remove noise and enhance to enhance the localization accuracy. The proposed algorithm has been tested in indoor and outdoor road environments to show its effectiveness and performance for localizing robot in such environments. It is also compared with fuzzy-logic, adaptive H-infinity filter, fuzzy logic with Kalman filter and Oriented-Rotated Binary Simultaneous Localization and Mapping 2 algorithms on localizing robot at road environment. Results show the capability of the proposed algorithm to enable the robot to move effectively in road environments, recognize the features of road terrains and localize the robot during autonomous navigation in road environments.}, archive = {J_EAAI}, author = {Mohammed A.H. Ali and Nukman Yusoff and Bushroa Abd Razak and Sherzod Turaev and Rawad Abdulghafor and Aisha Muhammad}, doi = {10.1016/j.engappai.2025.112145}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112145}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel localization algorithm integrated with sensor fusion for position estimation of mobile robot in the road following and roundabout environment}, volume = {161}, year = {2025}, }
• (2025). Electroencephalography-based emotion recognition using a dual-stream multi-scale spatiotemporal convolutional capsule network. EAAI, 161, 112144. (www)

  @article{ , abstract = {In recent years, emotion recognition using electroencephalogram (EEG) signals has gained significant attention. However, shallow convolutional neural networks used in EEG emotion recognition are ineffective at capturing spatial relationships between features, adversely affecting model performance. To solve this problem, this study proposes a dual-stream multi-scale spatiotemporal convolutional capsule network aimed at improving EEG-based emotion recognition. The novelty of our approach is the dual-stream multi-scale spatiotemporal design, which enables parallel extraction and fusion of temporal and spatial EEG features at multiple scales, offering a richer and more discriminative representation for emotion recognition. Specifically, a dual-stream feature construction module is developed to extract multi-scale temporal and spatial features from raw EEG signals. A hybrid spatiotemporal attention mechanism enhances feature fusion, while a capsule-based classifier improves recognition accuracy by modeling relationships between local and global features. Experimental results using subject-dependent 10-fold cross-validation show average accuracies of 97.72%, 97.56%, and 97.82% for valence, arousal, and dominance on the Dataset for Emotion Analysis using Physiological signals(DEAP), with average F1 scores of 97.83%, 97.51%, and 97.68%. For the Dataset for Emotion Analysis using Physiological signals(DREAMER 1 ), the average accuracies are 96.48%, 96.32%, and 96.35%, with corresponding F1 scores of 96.23%, 95.74%, and 95.66%. The proposed method outperforms existing state-of-the-art approaches on both datasets, reducing the number of parameters by 58.07% and decreasing inference time by approximately 50.86% and 33.39%, compared to Residual Network 18. Additionally, in the subject-independent leave-one-subject-out cross-validation, the proposed method demonstrated results that significantly outperformed the baseline model across both datasets. Experiments show that the proposed method enhances spatiotemporal feature extraction and improves emotion recognition accuracy. This method reduces computational resource consumption and enhances recognition accuracy, thereby facilitating efficient algorithm development and deployment for real-time emotion monitoring and related applications.}, archive = {J_EAAI}, author = {Han Cai and Pengfei Lu and Xiaofang Wang and Yuxing Wang}, doi = {10.1016/j.engappai.2025.112144}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112144}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Electroencephalography-based emotion recognition using a dual-stream multi-scale spatiotemporal convolutional capsule network}, volume = {161}, year = {2025}, }
• (2025). Inverse structural design with generative and probabilistic autoencoders and diffusion models. EAAI, 161, 112143. (www)

  @article{ , abstract = {Traditional structural design is a forward trial-and-error process. Designers need to iterate through different design solutions and conduct structural analysis until the design meets the codes and standards. This study proposes and investigates a generative machine learning (ML) framework for inverse design of continuous beam systems. Three generative ML models, including conditional variational autoencoder (CVAE), conditional autoencoder with maximum likelihood estimation (CAE-MLE), and denoising diffusion models (DDMs) are trained and fine-tuned on the CBeamXP (Continuous Beam Cross-section Predictors) dataset with 1,000,000 beam sections to generate the cross sectional properties. Research results show that CAE-MLE achieves the highest generation accuracy and robustness, while CVAE offers more variability through latent space sampling. DDMs provide controllable generation variability via a stochasticity parameter in the inverse diffusion process. The proposed framework enables efficient generation of multiple design solutions and can potentially accelerate the conceptual design workflows in structural engineering. This work also demonstrated the feasibility toward artificial intelligence (AI)-assisted structural design using generative approaches and tabular datasets.}, archive = {J_EAAI}, author = {Bozhou Zhuang and Adrien Gallet and Danny Smyl}, doi = {10.1016/j.engappai.2025.112143}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112143}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Inverse structural design with generative and probabilistic autoencoders and diffusion models}, volume = {161}, year = {2025}, }
• (2025). Multi-objective generative design framework and realization for quasi-serial manipulator: Considering kinematic and dynamic performance. EAAI, 161, 112142. (www)

  @article{ , abstract = {This paper proposes a framework for optimizing the linkage mechanism of a quasi-serial manipulator for target tasks. The process is illustrated through a case study of a two-degree-of-freedom (2-DOF) linkage mechanism, which significantly influences the workspace of the quasi-serial manipulator. First, a diverse set of quasi-serial mechanisms is generated with workspaces that satisfy the target task and is converted into three-dimensional computer-aided design (3D CAD) models. Then, kinematic and dynamic analyses are conducted to compute workspace and payload torque labels for surrogate model training. Adaptive sampling is employed to identify an appropriate amount of training data for accurate prediction, while ensemble models are utilized to enhance robustness. After model training, a multi-objective optimization problem is formulated under the mechanical and dynamic constraints of the manipulator. The design objective is to recommend quasi-serial mechanisms with optimized kinematic (workspace) and dynamic (payload torque) performance that fulfill the target task. To explore the underlying physics of the Pareto solutions obtained via the Non-Dominated Sorting Genetic Algorithm (NSGA-II), various data mining techniques are applied to extract design rules that offer practical guidance. Finally, a detailed manipulator is realized using 3D-printed parts with topology optimization, and its performance is verified through a payload test. These results demonstrate the potential of the proposed framework for broader mechanism applications and its capability to support practical design decisions through design rule extraction.}, archive = {J_EAAI}, author = {Sumin Lee and Sunwoong Yang and Namwoo Kang}, doi = {10.1016/j.engappai.2025.112142}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112142}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective generative design framework and realization for quasi-serial manipulator: Considering kinematic and dynamic performance}, volume = {161}, year = {2025}, }
• (2025). Blockchain and deep learning-enabled IoT device-to-device authentication approach for smart cities using 5th generation technology. EAAI, 161, 112141. (www)

  @article{ , abstract = {As smart cities evolve, the increasing number of Internet of Things (IoT) devices requires secure authentication mechanisms for device-to-device (D2D) communication, especially within 5G networks. To address security challenges in D2D communication, Blockchain (BC) technology is utilized. Existing methods face issues like single-point failure attacks, high implementation costs, and data privacy concerns. This work aims to develop a secure and efficient authentication model leveraging BC technology and a Deep Q Network (DQN) for key generation to enhance IoT security in smart city applications. A novel authentication model, Deep Q Network-SecAuth (DQN-SecAuth), is proposed for secure D2D communication using BC. The model involves key entities such as the Registration Authority Center (RAC), blockchain, and IoT devices. The authentication process includes six stages: setup, registration, key generation, authentication, new device addition, and formal verification. In the key generation stage, the secret key is generated using a Deep Learning (DL) model named DQN. The proposed model also incorporates encryption, hash functions, Chebyshev polynomials, and XOR operations to ensure security. The DQN-SecAuth model achieved a minimum computational time of 9.525 s and memory usage of 42.897 megabytes (Mb), consensus delay of 0.420 s, energy consumption of 0.715 J, latency of 0.469 s, power consumption of 0.270 kW-hour (kWh), and throughput of 0.804 megabits per second (Mbps). The key novelty of this work is the use of a DQN for adaptive key generation, integrated into a BC-based authentication framework. This combination enhances security, reduces computational overhead, and outperforms traditional static key-generation and authentication methods in smart city IoT environments.}, archive = {J_EAAI}, author = {K. Prabhu Chandran and Bhuvaneswari P and Sivasankaran V and Vimala S}, doi = {10.1016/j.engappai.2025.112141}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112141}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Blockchain and deep learning-enabled IoT device-to-device authentication approach for smart cities using 5th generation technology}, volume = {161}, year = {2025}, }
• (2025). Emergency medical facility site selection in drone-based relief operations using an enhanced T-spherical fuzzy frank combined compromise solution method. EAAI, 161, 112140. (www)

  @article{ , abstract = {Unmanned aerial vehicles (UAVs) play a vital role in disaster response operations, emphasizing the need for efficient site selection for emergency medical facilities (EMFs). This study presents a structured framework for evaluating candidate sites and identifying critical criteria for establishing EMFs in drone-based relief operations. It proposes a novel T-spherical fuzzy (T-SF) multi-criteria group decision-making framework for EMF site selection. The framework integrates the T-SF-Entropy method and a hybrid subjective–objective weighting scheme that combines stepwise weight assessment ratio analysis and the method based on the removal effects of criteria. This integration enables a more reliable aggregation of expert opinions. The framework applies T-SF Frank aggregation operators in the modified combined compromise solution method to rank potential sites under multi-disaster conditions. A real-world case study evaluates alternative sites using the critical criteria and demonstrates the practical utility of the proposed model in the field of emergency engineering logistics. Quantitative analysis shows that the “urban logistics hub” achieves the highest compromise index due to its advantages in proximity, infrastructure, and supply chain access. Sensitivity analysis confirms that variations in parameters do not affect the stability of rankings. This study uses artificial intelligence to support intelligent decision-making in disaster response. It provides an effective engineering application that optimizes UAV-based healthcare deployment, enhances resource allocation, and improves the speed and accuracy of emergency medical responses.}, archive = {J_EAAI}, author = {Rajdip Mahajan and Saptadeep Biswas and Vladimir Simic and Dragan Pamucar and Abhijit Baidya and Uttam Kumar Bera}, doi = {10.1016/j.engappai.2025.112140}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112140}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Emergency medical facility site selection in drone-based relief operations using an enhanced T-spherical fuzzy frank combined compromise solution method}, volume = {161}, year = {2025}, }
• (2025). Feature description attention: Channel-independent local–global fusion for multi-scale feature representation. EAAI, 161, 112139. (www)

  @article{ , abstract = {In recent deep convolutional networks, attention mechanisms rely heavily on global information to generate attention weights, but few methods can effectively recalibrate global features based on local object characteristics. To address this gap, we introduce a masked-averaging strategy that adaptively selects regions of interest from the feature map, allowing local features to encode and reflect global information. By combining these local descriptors with global averages and maximums into a multi-scale Bag-of-Visual-Words (BoVW), our method jointly captures salient point, local region, and global context, resulting in richer and more discriminative feature representations. Additionally, we propose Feature Batch Normalization (FBN) to facilitate cross-channel interactions, further enhancing the performance of attention modules. In terms of engineering applications, to overcome the limitations of traditional channel-dependent attention under model compression and pruning, we design a channel-independent feature recalibration mechanism based on one-dimensional depthwise convolution, termed Feature Description Attention (FDA). It leverages the properties of BoVW and FBN to provide a flexible and lightweight attention module with minimal computational overhead and robust performance under quantization and pruning. Our method demonstrates consistent gains across three vision tasks: classification accuracy improves by 0.72–1.68%, object detection performance increases by 1.8–3.0% mean average precision, and segmentation accuracy rises by 1.9%. These results across different benchmarks confirm the effectiveness and generalizability of our approach. FDA offers a scalable, hardware-friendly solution for modern vision applications, especially valuable in resource-constrained settings such as mobile deployment, autonomous systems, and medical image analysis.}, archive = {J_EAAI}, author = {Yuanyang Zhu and Guangjie Han and Hongbo Zhu and Fan Zhang}, doi = {10.1016/j.engappai.2025.112139}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112139}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature description attention: Channel-independent local–global fusion for multi-scale feature representation}, volume = {161}, year = {2025}, }
• (2025). Improving robustness of transformers for power quality disturbance classification via optimized relevance maps. EAAI, 161, 112138. (www)

  @article{ , abstract = {Power quality disturbances (PQDs) are a critical area of research in the power systems domain. The growing complexity of modern power systems increases the variability of PQDs that may occur, making it challenging to manage power quality using traditional methods, and motivates the inclusion of machine learning models. However, these models are vulnerable to adversarial attacks. The predominant defense approach is adversarial training, which increases robustness at the price of significant increase in computational complexity of the training process, and can often lead to reduced accuracy on non-adversarial examples. In this light, this paper introduces a distinct approach to increase the robustness of Transformer-based PQD classifiers by leveraging their intrinsic explainability derived from the self-attention mechanism. By incorporating the resulting explainable features in the form of relevance maps into the loss function during the fine-tuning stage, we guide the model to focus on salient disturbance features. This method enhances the robustness against adversarial attacks, without compromising accuracy, and avoids the additional computational overhead associated with adversarial training. Experimental results show that when no adversarial attack is applied, the fine-tuning stage increases the PQD classification accuracy from 97.62% to 99.26%, thus overcoming the accuracy-robustness trade-off. Furthermore, the Transformer with the proposed fine-tuning stage is more adversarial robust than a state-of-the-art deep Convolutional neural network (DeepCNN) PQD classifier. The proposed Transformer maintains accuracy and F1 scores slightly above 88%, whereas fo DeepCNN both scores are slightly below 66.60%, thus highlighting the effectiveness of our approach.}, archive = {J_EAAI}, author = {Itamar Kapuza and Elinor Ginzburg-Ganz and Ram Machlev and Yoash Levron}, doi = {10.1016/j.engappai.2025.112138}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112138}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving robustness of transformers for power quality disturbance classification via optimized relevance maps}, volume = {161}, year = {2025}, }
• (2025). Progressive epipolar geometry for robust light field super-resolution. EAAI, 161, 112137. (www)

  @article{ , abstract = {Light field technology captures both spatial and angular information, densely sampled high-resolution light field images contain abundant 3-dimensional geometric information, enabling widespread applications in various industrial fields. However, existing methods for joint spatial and angular super-resolution of light field images face significant challenges when reconstructing scenes with large disparities and significant occlusions. To address this issue, we propose a geometric information enhancement module that enables efficient extraction and preservation of geometric information, and then we propose a non-disparity-based, one-stage approach that allows for the direct reconstruction of densely sampled high-resolution light field images from sparsely sampled low-resolution counterparts. Specifically, we decompose the original 4-dimensional light field image into three distinct 2-dimensional representations: spatial, angular, and epipolar plane image. Among these representations, the epipolar plane image contains abundant geometric information, which inspires us to propose a more efficient feature extractor. In addition, we introduce a progressive feature fusion strategy that better preserves geometric information extracted from epipolar plane images. Finally, to avoid errors introduced by warping operations that complicate the process of enhancing geometric information, we introduce a spatial-angular integrated upsampling module. Extensive experimental results on public datasets demonstrate that our proposed method significantly outperforms state-of-the-art approaches both quantitatively and qualitatively. Specifically, on the Occlusions dataset, our method achieved significant improvement in performance while reducing inference time by approximately 80% compared to the current best method. This efficiency gain is particularly beneficial for practical applications of the artificial intelligence algorithm.}, archive = {J_EAAI}, author = {Hao Zhang and Hao Sheng and Rongshan Chen and Da Yang and Ruixuan Cong and Zhenglong Cui and Xuefei Huang and Guanqun Su}, doi = {10.1016/j.engappai.2025.112137}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112137}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Progressive epipolar geometry for robust light field super-resolution}, volume = {161}, year = {2025}, }
• (2025). Exploring knee joint moment prediction models for landing task with various model architecture: Leveraging inertial measurement unit sensors data and subject-specific attributes. EAAI, 161, 112136. (www)

  @article{ , abstract = {This study presents a comparative evaluation of deep learning models for predicting Three-Dimensional (3D) knee joint moments during landing tasks using Inertial Measurement Unit (IMU) data. We assess a range of architectures, from simple feedforward networks to complex Three-Dimensional Convolutional Neural Networks (3D-CNN), to balance predictive accuracy with computational cost. The 3D-CNN model achieved the highest performance, with relative Root Mean Squared Error (rRMSE) values of 6.08 % for the knee abduction moment, 5.3 % for the external rotation moment, and 4.9 % for the flexion moment. We further explored the effect of adding subject-specific features, which yielded modest performance gains. While these features provided some improvement, model performance primarily depended on sensor data, reinforcing the need to integrate both biomechanical and Artificial Intelligence-driven insights. From an engineering application perspective, our findings guide researchers and clinicians in selecting the most suitable model and sensor placement based on both accuracy and computational complexity. Specifically, we highlight thigh-mounted IMU as the most effective placement for capturing landing dynamics. These insights support the practical deployment of knee joint moment prediction systems in real-world settings, such as gyms or training centers, where compact sensors and limited computational resources are essential.}, archive = {J_EAAI}, author = {Tommy Sugiarto and Yi-Jia Lin and Ya-Wen Tu and Hsiao-Liang Tsai and Lin-Fen Hsieh and Chi-Tien Sun and Patrik Kutilek and Wei-Chun Hsu}, doi = {10.1016/j.engappai.2025.112136}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112136}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploring knee joint moment prediction models for landing task with various model architecture: Leveraging inertial measurement unit sensors data and subject-specific attributes}, volume = {161}, year = {2025}, }
• (2025). Key region semantic information augmented transformer for image captioning. EAAI, 161, 112135. (www)

  @article{ , abstract = {Existing image captioning models often face difficulties in capturing inter-object relationships and generating description that comprehensively understands the entire image content, either relying on object detectors that overlook contextual information or depending on grid features that fail to adequately model spatial interactions. This paper proposes two solutions to these challenges. The first is the introduction of a module for mining semantic information from key regions. Based on the spatial proximity and high co-occurrence between objects, this module identifies the public region covered by these objects as a key region, mines their semantic information, and incorporates it into the modeling process, which compensates for the limitations of grid features. Second, we improve the standard Transformer decoder’s architecture by innovatively introducing an adaptive gating mechanism that dynamically adjusts the alignment between textual and visual features, enhancing the model’s overall comprehension of the image. To validate our approach, we applied these modules to the Transformer framework and proposed a novel method for image captioning, called Key region Semantic information Augmented Transformer (KSAT) for Image Captioning. Extensive experiments on benchmark datasets show that the proposed method outperforms many models. Specifically, our method achieves a score of 139.6% on the offline test, and 138.4% on the official online test server on the Consensus-based Image Description Evaluation (CIDEr) metric. In qualitative evaluation, our method also outperforms other methods at generating captions for complex scenes. Overall, these results confirm the validity of our method and advance the field of artificial intelligence.}, archive = {J_EAAI}, author = {Fuyun Deng and Wei Li and Zhixin Li}, doi = {10.1016/j.engappai.2025.112135}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112135}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Key region semantic information augmented transformer for image captioning}, volume = {161}, year = {2025}, }
• (2025). Enhanced neural-network-based iterative learning control considering iterative uncertainties for piezoelectric actuated micro-positioning platform. EAAI, 161, 112134. (www)

  @article{ , abstract = {This research aims to developing a new enhanced data-driven sliding-mode iterative learning control (E-DDSILC) strategy for piezoelectric actuated micro-positioning (PAMP) platforms. For the first time, the analysis demonstrating that errors converge to 0 in E-DDSILC is successfully extended from strictly repetitive systems to systems with non-strictly repetitive initial conditions. This generalization expands the practical application range of E-DDSILC. Simultaneously, iterative uncertainties are considered, which are the major factor affecting the performance of iterative learning control. To address these uncertainties, a diagonal recurrent neural network is employed to fit and compensate for them within a dynamic linearization model, thereby further enhancing the tracking accuracy and practicability of E-DDSILC. Finally, Several experiments are performed on a PAMP platform to compare the developed E-DDSILC method with both classical DDSILC and traditional E-DDSILC schemes. Comparative experimental results prove the superiority of the developed controller.}, archive = {J_EAAI}, author = {Miaolei Zhou and Yulong Sun and Xiuyu Zhang and Wei Gao and Chun-Yi Su}, doi = {10.1016/j.engappai.2025.112134}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112134}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced neural-network-based iterative learning control considering iterative uncertainties for piezoelectric actuated micro-positioning platform}, volume = {161}, year = {2025}, }
• (2025). A data-driven multi-stage stochastic optimization for sustainable humanitarian supply chain using machine learning algorithms. EAAI, 161, 112133. (www)

  @article{ , abstract = {The frequency and severity of natural disasters have intensified, resulting in significant human, financial, and emotional consequences. Earthquakes, in particular, have caused severe economic losses, deaths, and homelessness among millions. This study is designed to establish a comprehensive plan for managing pre- and post-disaster phases, including preparations, responses, and recovery efforts. It introduces a Multi-Stage Stochastic Programming (MSSP) model for sustainable humanitarian relief operations, optimizing location, allocation, and inventory management. The first and third stages concentrate on minimizing environmental impacts, while the second stage centers on enhancing social welfare. Simultaneously, economic cost reduction is consistently pursued in all three stages. The model's primary advantages include optimized inventory management to avoid shortages and flexible logistics strategies for timely and cost-effective delivery of relief items. Additionally, it ensures continuous aid, addressing both short-term and long-term needs to improve disaster management effectiveness and resilience. The multi-objective model is solved using Augmented Epsilon-Constraint (AEC). Furthermore, this paper employs Multi-Criteria Decision-Making (MCDM) methods to rank suppliers, leveraging Machine Learning (ML) algorithms to enhance ranking precision, thereby leading to a more responsive Supply Chain (SC). A real-world case study is then provided to illustrate the applicability and validity of the proposed model. Focusing on achieving balanced sustainability across all three stages, ensuring seamless logistics for all humanitarian supplies and affected individuals, and addressing uncertainties, the model determines the optimal quantities of all relief items to store. Through comprehensive analysis, the results provide key insights into the importance of MSSP in disaster management plans, enhancing understanding of the model's effectiveness.}, archive = {J_EAAI}, author = {Farnaz Ansari and Ali Bozorgi-Amiri and Hossein Shakibaei}, doi = {10.1016/j.engappai.2025.112133}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112133}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A data-driven multi-stage stochastic optimization for sustainable humanitarian supply chain using machine learning algorithms}, volume = {161}, year = {2025}, }
• (2025). A theme music generation model based on hybrid variational autoencoders and conditional generative adversarial networks. EAAI, 161, 112131. (www)

  @article{ , abstract = {The musical theme is the core melodic element that runs through a work. It is the main source of recognizable features and the basis of the overall structure of the score. However, existing music generation systems often face problems with missing thematic features and monotonous structure. These systems find it difficult to capture the complex thematic changes and long-term dependencies in music, resulting in a lack of dynamic changes, artistic expression in phrasing, and structural coherence in the generated results. In this paper, we propose a thematic music generation model based on a generative adversarial network, named Thematic Music Conditional Generative Adversarial Network (TM-CGAN), which solves these challenges through three key innovations. First, we propose a thematic fusion degree calculation module based on multi-dimensional feature metrics to overcome the theoretical limitations of existing methods, which lack explicit modeling of thematic features. Subsequently, we construct a two-dimensional image representation learning framework that preserves Musical Instrument Digital Interface (MIDI) symbolic sequences and employs convolutional neural networks (CNNs) to effectively model the multi-dimensional feature dependencies inherent in musical structures. Finally, we design a hybrid variational autoencoder-conditional generative adversarial network architecture that processes latent space modeling of thematic features through variational inference mechanisms while simultaneously optimizing generation quality through adversarial training. We conducted comprehensive experiments across three diverse MIDI music datasets to validate our approach. The experimental results demonstrate that TM-CGAN significantly outperforms state-of-the-art baseline models on multiple evaluation metrics, including generated music quality, theme representativeness, and structural integrity.}, archive = {J_EAAI}, author = {Fangzhu Jin and Peng Li and Xiaojun Wu}, doi = {10.1016/j.engappai.2025.112131}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112131}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A theme music generation model based on hybrid variational autoencoders and conditional generative adversarial networks}, volume = {161}, year = {2025}, }
• (2025). Malicious detection and trust calculation using residual recurrent neural network for trust with quality of service-aware multicast routing in mobile ad-hoc network system. EAAI, 161, 112130. (www)

  @article{ , abstract = {A Mobile Ad-hoc Network is a collection of mobile nodes without any proper infrastructure. In this work, a novel trust and Quality of Service-aware multicast routing technique for Mobile Ad-hoc Network is introduced. The key scope of this research paper is to evaluate the trust with Quality of Service -aware multicast routing process in Mobile Ad-hoc Network by detecting malicious nodes. For performing the optimal routing without any suspicious attacks initially the malicious node detection is performed by Residual Recurrent Neural Network. Further, if the Mobile Ad-hoc Network model is normal and free from malicious nodes, then the true value is calculated by utilizing the outcome of malicious node detection. If the node is free from malicious then, the trust value becomes high or else the trust value becomes low. Once, the trust calculation is completed the optimal routing is performed in the Mobile Ad-hoc Network with the support of the Enhanced Artificial Rabbits Optimization algorithm. Moreover, different constraints like hop count, throughput, Packet Delivery Ratio, delay, and energy consumption are derived. Further, different experiments are evaluated to prove the effectiveness of the implemented model against several baseline technique. Hence, the developed model accomplishes superior efficiency in detecting the malicious and performs the multicast routing in the Mobile Ad-hoc Network.}, archive = {J_EAAI}, author = {Sanjaya Kumar Sarangi and Rasmita Lenka and Janmejaya Mishra and Ritarani Sahu and Arabinda Nanda}, doi = {10.1016/j.engappai.2025.112130}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112130}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Malicious detection and trust calculation using residual recurrent neural network for trust with quality of service-aware multicast routing in mobile ad-hoc network system}, volume = {161}, year = {2025}, }
• (2025). Energy management system scheduling optimization based on an improved generative adversarial network deep reinforcement learning algorithm. EAAI, 161, 112129. (www)

  @article{ , abstract = {Households, as electricity consumers, play a critical role in achieving the carbon peak and carbon neutrality goals. The development of smart grids provides technical support for the efficient integration and distribution of renewable energy, gradually extending to household users. This has led to higher demands for the stability of electricity supply to address the growing electricity demand and the uncertainties associated with renewable energy. To address this, this paper proposes an improved generative adversarial network and an enhanced deep reinforcement learning algorithm to improve the scheduling capability of the energy management system. First, we introduce an improved wasserstein generative adversarial network that combines stochastic differential equations and autocorrelation penalty terms with the generator. The experimental results demonstrate that the proposed method can generate high-quality time series data. The generated data were used to train our scheduling model, effectively enhancing its generalization capability. Secondly, We introduced the Minmax mechanism to address Q-value estimation bias by utilizing multiple Q-networks. This mechanism first divides the target Q-values into several groups, selects the maximum value from each group, and then takes the minimum among these maxima as the final target Q-value. We applied this mechanism to improve deep reinforcement learning algorithms based on multi-Q-value evaluation. Comparison experiments show that this improvement significantly enhances the algorithm’s performance, outperforming traditional algorithms in terms of convergence, volatility, and final rewards. The energy management system demonstrates stronger adaptability when handling uncertainties arising from renewable energy variations, ensuring reliable power supply and achieving balanced energy management.}, archive = {J_EAAI}, author = {Weipeng Chao and Yuanbo Shi and Yushuai Li and Meng Liu and Xiaoling Leng}, doi = {10.1016/j.engappai.2025.112129}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112129}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Energy management system scheduling optimization based on an improved generative adversarial network deep reinforcement learning algorithm}, volume = {161}, year = {2025}, }
• (2025). Deep ensemble learning and error correction method for remaining useful life prediction of rolling bearings. EAAI, 161, 112128. (www)

  @article{ , abstract = {Rolling bearings are critical components of rotating machinery, and the prediction of their remaining useful life (RUL) is important for system reliability and operating efficiency. A novel RUL prediction method based on deep ensemble learning and error correction is here proposed. Firstly, the moving average filter (MAF) is applied to preprocess vibration signals for removing outliers and noise. Then, time-domain features are extracted from the processed vibration signals, and optimized to obtain imperative feature signals. Subsequently, a deep ensemble learning model is built with gated recurrent unit (GRU), bidirectional GRU (BiGRU), long short-term memory (LSTM) and bidirectional LSTM (BiLSTM) as base learners, and the overall performance of the prediction model is enhanced by introducing an error correction strategy. The MAF method is also used to smooth the trend of the RUL prediction outcomes. Finally, the proposed method is applied to two full-lifecycle rolling bearing datasets: the Prognostics and Health Management 2012 (PHM2012) dataset and the Intelligent Maintenance System (IMS) dataset. It is evaluated using mean square error (MSE), mean absolute error (MAE), and the R-square coefficient of determination (R 2 ). The test results demonstrate that the method achieves highly accurate RUL predictions: on the PHM2012 dataset, the MSE, MAE, and R 2 reach 0.000380, 0.013695, and 0.994716, respectively; on the IMS bearing dataset, the corresponding values are 0.001056, 0.015403, and 0.978346. In addition, the method outperforms traditional single models (GRU, BiGRU, LSTM, BiLSTM) as well as the Transformer model in both cases, with R 2 improvements over the Transformer of 0.011065 and 0.008744, respectively. These results highlight the superior generalization capability and robustness of the proposed method, making it applicable to industrial environments requiring reliable RUL prediction.}, archive = {J_EAAI}, author = {Wenzhe Yin and Hong Xia and Enrico Zio and Xueying Huang}, doi = {10.1016/j.engappai.2025.112128}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112128}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep ensemble learning and error correction method for remaining useful life prediction of rolling bearings}, volume = {161}, year = {2025}, }
• (2025). Maritime supply chain optimization using robust adversarial reinforcement learning. EAAI, 161, 112127. (www)

  @article{ , abstract = {This study explores novel strategies for analyzing and managing port productivity in a multi-stage supply chain network by integrating synchronization and reinforcement learning (RL) techniques. Current port management systems face issues with nonlinearity, high interdependency, and vulnerability to market disruptions, which might destabilize port operations. To tackle these issues, seaport operations are examined in four stages of implementation: the terminal operator, inland carrier, inland terminal operator, and consignee. Brownian motion is applied to characterize stochastic disruptions in volatile markets, and the port performance under market disruptions is analyzed using nonlinear data analytics. The dynamical analysis reveals that port management systems exhibit highly coupled nonlinear dynamics with a tendency towards instability. The complex nature of maritime port logistics requires innovative strategies to analyze container handling volumes, optimize the strategic planning of port management, and improve overall efficiency. A novel optimal policy for port operations is realized by integrating a deep deterministic policy gradient into robust adversarial deep learning. A deep reinforcement learning algorithm is employed to learn adaptively from historical port-related data and real-time container handling feedback, enabling intelligent strategies to make informed decisions and dynamically adjust policies in response to stochastic disruptions or changing market conditions. Quantitative results demonstrate that the proposed strategy achieves up to 97.78 % operating efficiency despite disturbances, and the adversarial attacks are shown to decrease port productivity by up to 77.78 % in scenarios without robust deep learning support. This study contributes to the growing field of intelligent port operations by paving the way for more adaptive and smart solutions in the maritime supply chain network.}, archive = {J_EAAI}, author = {Truong Ngoc Cuong and Sam-Sang You and Le Ngoc Bao Long and Hwan-Seong Kim and Duy Anh Nguyen and Nguyen Duy Tan}, doi = {10.1016/j.engappai.2025.112127}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112127}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Maritime supply chain optimization using robust adversarial reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). Employing dual-path structure and soft attention mechanism to enhance recognition and classification of wild medicinal licorice in xinjiang. EAAI, 161, 112126. (www)

  @article{ , abstract = {Licorice is highly valued in traditional Chinese medicine for its anti-inflammatory, antiviral, and immunomodulatory properties, and is widely used in the pharmaceutical, food, and cosmetic industries. Xinjiang, the largest licorice-producing region in China, faces severe overharvesting of wild licorice due to increasing market demand, threatens natural populations and fragile ecosystems. Accurate identification and classification of licorice species are crucial for environmental protection and sustainable resource utilization, as traditional methods relying on experience are inefficient, subjective, and prone to errors. This study builds on the Inception-Residual Network-Version 2 (Inception-ResNet-V2) architecture and proposes an advanced licorice recognition model called Inception-ResNet-V2-Soft Attention, Dual-path Structure, and Focal Loss (IRV2-SDF). The IRV2-SDF model integrates a soft attention mechanism that focuses on key regions, a dual-path structure for multi-scale feature extraction, and a focal loss function to address class imbalance. It aims to improve the identification and classification of three wild licorice species ( Glycyrrhiza glabra , Glycyrrhiza inflata , and Glycyrrhiza uralensis ) and associated weeds in complex environments. Trained on 3,653 images collected from Xinjiang, the model achieves an average recognition accuracy of 91.79%, surpassing traditional models, with accuracy improvements of 4.27%, 2.08%, 2.76%, and 6.36% for G. glabra , G. inflata , G. uralensis , and weeds, respectively. By effectively reducing background noise and enhancing detection capabilities, the model overcomes the limitations of traditional methods and provides a robust solution for wild licorice recognition. This research offers a technical foundation for licorice conservation and sustainable utilization and can serve as a reference for identifying other medicinal plants in complex environments.}, archive = {J_EAAI}, author = {Yuan Qin and Jianguo Dai and Guoshun Zhang and Miaomiao Xu and Jing Yang and Jinglong Liu}, doi = {10.1016/j.engappai.2025.112126}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112126}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Employing dual-path structure and soft attention mechanism to enhance recognition and classification of wild medicinal licorice in xinjiang}, volume = {161}, year = {2025}, }
• (2025). Real-time cooperative target tracking in cluttered environments using multiple drone swarms with adaptive fuzzy emotional learning. EAAI, 161, 112125. (www)

  @article{ , abstract = {This paper presents a real-time adaptive trajectory prediction framework for cooperative unmanned aerial vehicle (UAV) swarms engaged in dynamic target tracking within cluttered environments. The proposed system introduces a novel artificial intelligence (AI)-based control architecture combining fuzzy inference, neuro-emotional learning, and distributed multi-agent coordination. At the core of the approach is a Bidirectional Fuzzy Brain Emotional Learning Prediction (BFBEL-P) model, which integrates fuzzy logic and an online adaptive neural structure to enable trajectory forecasting without pre-training or prior knowledge of the environment. From an engineering perspective, this AI model is deployed in UAV swarm navigation, where robust decision-making, predictive coordination, and obstacle avoidance are essential for target interception missions. In contrast to conventional prediction methods, such as curve fitting, nonlinear model predictive control (MPC), and deep learning-based Long Short-Term Memory (LSTM) networks, the BFBEL-P framework offers fast convergence, low computational cost, and high adaptability. The system incorporates multi-threaded data fusion across the swarm to achieve consensus-driven predictions and maintain situational awareness, even under sensor failures or occlusions. Simulation results show that BFBEL-P improves short-term prediction accuracy by 82.2%, reduces prediction time by 15%, and achieves a 100% tracking success rate across benchmark scenarios. These results establish BFBEL-P as a reliable AI technique for distributed control in real-world UAV swarm applications, offering a promising tool for search-and-rescue, surveillance, and defense operations.}, archive = {J_EAAI}, author = {Lucas William Page and Vu Phi Tran and Duy Luan Nguyen}, doi = {10.1016/j.engappai.2025.112125}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112125}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time cooperative target tracking in cluttered environments using multiple drone swarms with adaptive fuzzy emotional learning}, volume = {161}, year = {2025}, }
• (2025). Hierarchical multi-scale matched masked autoencoder for industrial multi-rate time series modeling. EAAI, 161, 112122. (www)

  @article{ , abstract = {In practical industrial processes, due to sensor hardware limitations, the sampling rates of different variables often vary, leading to multi-rate time series (MRTS) data. However, the distribution of multi-scale dynamics in MRTS data typically follows a step-like pattern, with intricate scale transitions from fine to coarse and complex scale-consistent dependencies across rates. Additionally, the inherent characteristics of MRTS data often result in label scarcity. Both factors present significant challenges for MRTS modeling. To address these issues, we propose a novel self-supervised learning strategy, called H ierarchical M ulti-Scale M atched M asked A utoencoder (H3MAE). Specifically, we design a scale-matching input fusion mechanism where each layer is hierarchically aligned to a specific scale, with the scale-matching integration from two sources, effectively capturing the multi-scale dynamics and cross-rate scale-consistent dependencies in MRTS data. Besides, we introduce a novel auxiliary task that imputes masked positions in the encoded representation space at each layer, aiming to achieve MRTS representation learning and mitigate label scarcity. Furthermore, we propose a unique encoder-imputer structure in each layer to enable multi-scale self-supervised learning while generating temporally aligned features satisfying the input requirements of the next layer. Experimental results on three benchmark datasets and two industrial multi-rate tasks demonstrate that our framework yields better performance in MRTS modeling. The code is publicly available at https://github.com/monolithycq/H3MAE .}, archive = {J_EAAI}, author = {Changqing Yuan and Yongfang Xie and Shiwen Xie and Jie Wang}, doi = {10.1016/j.engappai.2025.112122}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112122}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical multi-scale matched masked autoencoder for industrial multi-rate time series modeling}, volume = {161}, year = {2025}, }
• (2025). Unsupervised feature selection via latent feature representation and modified graph embedding. EAAI, 161, 112121. (www)

  @article{ , abstract = {Data in practical applications are correlated not only between samples but also between high-dimensional features. Latent representations can effectively represent such correlations. Existing latent representation methods mainly consider inter-instance relationships and rely on constructing similarity graphs. However, the interconnection information obtained by latent representations is redundant due to the noise and irrelevance of the raw data. Therefore, to address the above issues, this paper proposes an unsupervised feature selection by latent feature representation and modified graph (LFRMG, for short) embedding. First, a latent feature representation norm for learning self-representation structure is designed to explore the interconnections among features. Mining feature relationships via self-representation can solve the redundancy caused by fixed similarity graph relationships. Second, the latent feature representation is combined with modified graph regularization to achieve good feature interconnections while maintaining local data information. In addition, the l 2 , 0 -norm is utilized to select the full-row sparse projection of salient features to avoid the drawbacks of sparsity limitation and parameter tunability. A scheme for solving the closed form of l 2 , 0 is constructed. Finally, the presented approach is superior to many state-of-the-art unsupervised methods through comprehensive experiments on nine existing datasets.}, archive = {J_EAAI}, author = {Jialing Yan and Gang Hu and Guo Wei}, doi = {10.1016/j.engappai.2025.112121}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112121}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised feature selection via latent feature representation and modified graph embedding}, volume = {161}, year = {2025}, }
• (2025). Coarse-to-fine dual-branch network for ship target recognition in complex environments. EAAI, 161, 112120. (www)

  @article{ , abstract = {Harsh sea conditions and the complex and variable positions of ships significantly impact the capacity of imaging devices to capture high-quality ship images, making ship target recognition challenging in the application of artificial intelligence. Many scholars have recently proposed cascaded recognition models to address this issue. Following this method, in this paper, we propose a novel method for ship target recognition in complex environments called the coarse-to-fine dual-branch (CFDB) network. The CFDB model designs a dual-branch network from coarse to fine to lock the target area fine features and then uses peer-to-peer communication to extract and exchange learning of the target region’s final discriminative contour features, assisting in predicting ship classes in the complex environment. The proposed method is evaluated on the constructed complex in background ships (CIB-ships) dataset and the publicly available Marine Argos Recognition Ships (MAR-ships) and Game-of-Ships datasets. Compared with the suboptimal method, the proposed CFDB network exhibits improvements of 2.11%, 1.33%, and 1.24% accuracy on the CIB-ships, MAR-ships, and Game-of-ships datasets, respectively. The results demonstrate that the proposed method provides useful ideas for the dynamic monitoring of ships in real environments. Our code will be published at https://github.com/yangt1013/CFDB-master .}, archive = {J_EAAI}, author = {Yang Tian and Hao Meng}, doi = {10.1016/j.engappai.2025.112120}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112120}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Coarse-to-fine dual-branch network for ship target recognition in complex environments}, volume = {161}, year = {2025}, }
• (2025). A multi-scale sparse channel transformer network for image reconstruction of astronomical bright source contamination. EAAI, 161, 112119. (www)

  @article{ , abstract = {Bright source contamination has long been a challenging issue in the field of image processing, particularly in applications such as astronomical observations, satellite imaging, and nighttime surveillance. To address this issue, this paper proposes a novel Multi-Scale Sparse Channel Transformer Network (MSCformer) aimed at achieving high-quality image reconstruction under the influence of bright source contamination. The network integrates a Top-k Sparse Attention mechanism with a Channel Attention module, enabling selective focus on the most informative features and adaptive weight allocation across channels. Additionally, a Multi-Scale Dual-Gate Feedforward Network is designed to further enhance the expression of valuable features while suppressing redundant information. Experimental results demonstrate that the proposed method exhibits outstanding performance in practical applications on the Sloan Digital Sky Survey (SDSS) photometric image dataset. Compared to existing state-of-the-art techniques, MSCformer achieves significant performance improvements, with a Peak Signal-to-Noise Ratio (PSNR) of 45.093 decibel(dB), a Structural Similarity Index Measure(SSIM) of 0.978, and a Pixel Average Absolute Error (PAAE) of 0.675. This not only significantly enhances the removal of bright source contamination in the field of astronomy but also provides important reference value for subsequent research in related domains.}, archive = {J_EAAI}, author = {Yajuan Zhang and Congcong Shen and Xia Jiang and Bo Qiu and Ali Luo and Fuji Ren and Yuanlu Chen}, doi = {10.1016/j.engappai.2025.112119}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112119}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-scale sparse channel transformer network for image reconstruction of astronomical bright source contamination}, volume = {161}, year = {2025}, }
• (2025). Seismic damage assessment of reinforced concrete frame structures based on transfer-residual networks. EAAI, 161, 112118. (www)

  @article{ , abstract = {Post-earthquake building damage assessment is a critical research focus in civil engineering due to its vital role in guiding timely and informed rescue operations. This study proposes a rapid seismic damage assessment method for reinforced concrete frame structures based on transfer-residual networks. The methodology involves pre-training a base model on a large dataset generated from computationally efficient simplified model simulations, followed by transfer learning on a smaller, high-fidelity dataset derived from refined finite element model simulations. This approach significantly enhances prediction accuracy while reducing computational costs for seismic damage assessment in new building structures. Optimization of model parameters was performed to find the optimal residual network, which achieves an accuracy of 91.2 % with a remarkable 32-fold speedup over conventional nonlinear time-history analysis methods. Moreover, the transfer-residual network enhances accuracy by 10 % compared to training from scratch. The results conclusively demonstrate the feasibility of utilizing large datasets from simplified models to improve the training accuracy of refined models with limited data, providing a valuable reference for advancing transfer learning-based frameworks in seismic damage assessment of buildings.}, archive = {J_EAAI}, author = {Chen Xiong and Zhijie Luo and Jie Zheng and Linlin Xie and Liu Mei and Lixiao Li and Wujian Long}, doi = {10.1016/j.engappai.2025.112118}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112118}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Seismic damage assessment of reinforced concrete frame structures based on transfer-residual networks}, volume = {161}, year = {2025}, }
• (2025). Machine learning-enabled performance-based design of three-dimensional printed engineered cementitious composites. EAAI, 161, 112117. (www)

  @article{ , abstract = {The superior tensile ductility of engineered cementitious composites (ECC) offers a promising solution to the challenge of integrating conventional steel reinforcement in three-dimensional (3D) concrete printing (3DCP). However, the widespread adoption of 3D printed ECC (3DP-ECC) is hindered by the reliance on trial-and-error design process. The complex material component and inherent anisotropy of 3DP-ECC pose challenges for accurate property prediction and inverse design. This paper introduces a performance-based design strategy for 3DP-ECC, leveraging machine learning (ML) and multi-objective optimization. The anisotropic-mechanical properties including compressive strength and flexural strength were experimentally and statistically investigated; further, ML prediction models conbined with multi-objective optimization algorithm were developed to inversely design 3DP-ECC for specific mechanical performance requirements, while reducing carbon footprint and material cost. Specifically, an extensive database was assembled, followed by grey relational analysis (GRA) to identify the parametric sensitivity of the mechanical properties of 3DP-ECC. Three representative ML techniques were employed, with the back-propagation artificial neural network (BPANN) demonstrating superior predictive accuracy. Model interpretability analyses uncovered the importance of input parameters and their influence on predicted outcomes. Lastly, non-dominated Sorting Genetic Algorithm II (NSGA-II) integrated with the BPANN models was applied to perform the inverse design of 3DP-ECC, showing good effectiveness and accuracy. This work offers an efficient and viable avenue for performance-based design for 3DP-ECC, along with the potential to develop low-carbon cost-effective 3DP-ECC.}, archive = {J_EAAI}, author = {Wenguang Chen and Long Liang and Junhong Ye and Lingfei Liu and Neven Ukrainczyk and Liqiang Yin and Jiangtao Yu and Kequan Yu}, doi = {10.1016/j.engappai.2025.112117}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112117}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-enabled performance-based design of three-dimensional printed engineered cementitious composites}, volume = {161}, year = {2025}, }
• (2025). Personalized text-to-image generation with large language and vision assistant enhanced training. EAAI, 161, 112116. (www)

  @article{ , abstract = {Personalized image generation aims to synthesize images of a specific identity. The identity, denoted as V ∗ , refers to an entity with distinctive visual attributes, such as a dog-shaped backpack. However, existing methods like DreamBooth and Custom Diffusion often struggle to generate images of V ∗ that accurately match the input prompts. In this work, we analyze two key issues underlying this limitation: (1) the overbinding problem , where the prompt tokens used to represent V ∗ unintentionally bind to irrelevant visual details from the reference image during training; and (2) the low language prior problem , where insufficient use of pre-trained language prior limits the model’s ability to faithfully generate all the prompt words. To overcome these challenges, we propose LLaVA-Booth, a novel personalization method for diverse, identity-preserving image generation, based on Large Language and Vision Assistant (LLaVA) enhanced training. Our method alleviates the overbinding problem by disentangling background information and solves the low language prior problem by enriching the language context. Additionally, we introduce two auxiliary objectives: (1) an identity (ID) binding loss to strengthen the identity binding and (2) a prior preservation loss to prevent language drift and encourage generation diversity. Experiments demonstrate that LLaVA-Booth effectively mitigates overbinding and enhances language priors to improve prompt fidelity, then generates diverse, high-quality, and identity-preserving images of V ∗ .}, archive = {J_EAAI}, author = {Junsheng Luan and Zhanjie Zhang and Wei Xing and Lei Zhao}, doi = {10.1016/j.engappai.2025.112116}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112116}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Personalized text-to-image generation with large language and vision assistant enhanced training}, volume = {161}, year = {2025}, }
• (2025). A multi-view new energy vehicle form generation design method combining kansei imagery and deep learning. EAAI, 161, 112115. (www)

  @article{ , abstract = {In the competitive landscape of new energy vehicles, exterior design has become a crucial differentiator amid functional homogenization. User preferences are central to shaping vehicle appearance, yet most perceptual design methods rely on a single viewpoint, limiting insights into complex preference patterns. This study proposes a multi-perspective mapping approach that integrates Kansei engineering with deep learning. Firstly, user core imagery is collected and mined through big data. Secondly, Kernels Network (KNet) semantic segmentation model, Residual Networks (ResNet) tri-view (front/side/rear) score prediction model and fully connected network (FCN) feature fusion model are integrated to construct a multi-view feature mapping system. Finally, the optimal combination of morphological elements is explored based on the Elite Genetic Algorithm (EGA), and the scheme is validated through generative artificial intelligence (AI) workflow. The experimental results demonstrate that, employing “Cool” as a case study, the three-view scheme and the combination scheme devised by this research process exhibit substantial superiority over the majority of the samples. Under identical parameters, the scheme with decision constraints surpasses the randomly generated scheme in terms of perceptual scores and stability. The performance of the test set and the experimental results collectively substantiate the model’s validity. This workflow—covering preference extraction, morphological decomposition, AI-driven generation, and validation—provides a scalable framework for new energy vehicle exterior design. It also demonstrates novel applications of Kansei engineering in multi-view fusion and generative form design.}, archive = {J_EAAI}, author = {Zihao Wang and Le Xi and Yifan Ding and Wenjie Fang and Kaiming Wang and Hongliang Zuo}, doi = {10.1016/j.engappai.2025.112115}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112115}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-view new energy vehicle form generation design method combining kansei imagery and deep learning}, volume = {161}, year = {2025}, }
• (2025). Explainable remaining useful life uncertainty prediction method for rolling bearing. EAAI, 161, 112114. (www)

  @article{ , abstract = {Rolling bearing remaining useful life prediction is the core technology of equipment maintenance. Although deep learning-based prediction methods have made significant breakthroughs, the problems of insufficient model explainability and prediction result uncertainty quantification have seriously constrained the credibility of maintenance decisions. Therefore, this research combines prediction uncertainty quantification with model explanation to propose an explainable uncertainty prediction method. The method includes multi-dimensional feature extraction, remaining useful life uncertainty prediction, and Shapley additive explanations interpreter. For feature extraction, multi-dimensional feature vectors are constructed as network inputs by extracting time-domain features and frequency-domain features. Then, the remaining useful life prediction interval for the rolling bearing is compressed by the proposed gated temporal quantile network. Finally, the prediction model is explained using the Shapley additive explanations interpreter. The multi-case validation results based on the Xi'an Jiaotong University and Changxing Sumyoung Technology Co., Ltd. (XJTU-SY) and Intelligent Maintenance Systems (IMS) rolling bearing full life cycle datasets show that the proposed model has an average interval coverage of 93.96 % and an average interval width of 9.92 %, which indicates that the model maintains high accuracy and robustness in different cases. The nonlinear mapping relationship between the prediction results and the features is clarified by analyzing the Shapley values. Finally, the Shapley values are used to rank the importance of the features to locate the position that may cause rolling bearing performance degradation, which provides credible decision support for the development of the predictive maintenance strategy.}, archive = {J_EAAI}, author = {Ting Zhang and Honglei Wang}, doi = {10.1016/j.engappai.2025.112114}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112114}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable remaining useful life uncertainty prediction method for rolling bearing}, volume = {161}, year = {2025}, }
• (2025). Machine learning-based estimation of sound transmission loss in single and double-layered rectangular functionally graded plates. EAAI, 161, 112112. (www)

  @article{ , abstract = {This study presents a robust machine learning (ML) framework for predicting sound transmission loss (STL) in temperature-dependent functionally graded (FG) single- and double-layered plates, combining analytical modeling and data-driven approaches. The FG plates are modeled with material properties varying along the thickness via a power-law distribution, incorporating nonlinear temperature effects. A high-fidelity analytical solution is derived using first-order shear deformation theory (FSDT) and acoustic velocity potential, with governing equations formulated via Hamilton's principle and solved using the weighted residual method. To enable ML training, a large-scale parametric study generates 189,000 data points, covering variations in plate geometry, temperature, acoustic cavity depth, gradient index, and incident wave angles. ML algorithms are trained on this data to predict STL, with the extreme gradient boosting (XGBoost) algorithm demonstrating the highest accuracy (coefficient of determination R 2 = 99.38 % for training data and R 2 = 99.11 % for test data). The validated ML model is then employed to investigate key parameters—temperature, power-law index, incidence angle, cavity depth, and plate thickness—revealing their nonlinear interactions and impact on STL performance. The developed artificial intelligence (AI) framework provides an efficient tool for acoustic design optimization, offering real-time parameter tuning capabilities for engineers working with functionally graded acoustic barriers.}, archive = {J_EAAI}, author = {Chunfeng Jiang and Masoud Babaei}, doi = {10.1016/j.engappai.2025.112112}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112112}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-based estimation of sound transmission loss in single and double-layered rectangular functionally graded plates}, volume = {161}, year = {2025}, }
• (2025). Cubic linguistic T-spherical fuzzy aggregation operator-based multi-attribute group decision-making model and its application to food waste treatment technique selection. EAAI, 161, 112111. (www)

  @article{ , abstract = {Food waste presents significant environmental, economic, and social challenges globally, contributing to resource depletion and greenhouse gas emissions. Effective treatment techniques are essential for minimizing these impacts, promoting sustainability, and supporting circular economy practices. However, selecting the most suitable food waste treatment technique is a complex multi-attribute group decision-making (MAGDM) problem that involves the simultaneous consideration of environmental, economic, and social factors under significant uncertainty. To address this challenge, this study proposes a novel cubic linguistic T-spherical fuzzy sets (CLT-SFSs) framework, which integrates linguistic T-spherical fuzzy sets and linguistic interval-valued T-spherical fuzzy sets to enhance decision-making precision under uncertainty. The proposed CLT-SF framework offers a unified structure capable of capturing both qualitative expert opinions and quantitative uncertainty ranges, offering unprecedented expressiveness for complex sustainability assessments. First, the formal definition and basic operations for CLT-SF numbers, including addition, multiplication, scalar multiplication, and scalar power, and a comparison law are established. Next, to efficiently aggregate cubic linguistic T-spherical fuzzy information, we propose the cubic linguistic T-spherical fuzzy weighted averaging and the cubic linguistic T-spherical fuzzy weighted geometric aggregation operators. These aggregation operators can effectively and comprehensively aggregate attribute values in MAGDM problems. Subsequently, utilizing these operators, a CLT-SFS-based combinative distance-based assessment model is developed to address MAGDM problems. The model's applicability and robustness are demonstrated through a real-world case study on the selection of food waste treatment techniques. A parameter analysis is also conducted to examine the sensitivity of ranking outcomes. Finally, a comparative analysis with existing methods confirms the proposed model's effectiveness, feasibility, and advantages in addressing complex decision-making scenarios. This research not only advances the theoretical framework of fuzzy decision-making but also provides a practical tool for stakeholders in waste management to make informed and sustainable choices.}, archive = {J_EAAI}, author = {Shahid Hussain Gurmani and Weiping Ding and Rana Muhammad Zulqarnain and Jiangfeng Hao}, doi = {10.1016/j.engappai.2025.112111}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112111}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cubic linguistic T-spherical fuzzy aggregation operator-based multi-attribute group decision-making model and its application to food waste treatment technique selection}, volume = {161}, year = {2025}, }
• (2025). Contextual and orientation correction modules enhance weakly-supervised aerial object detection in remote sensing images. EAAI, 161, 112110. (www)

  @article{ , abstract = {This paper addresses the challenges of context and orientation ambiguity in weakly supervised aerial object detection. While current research focuses on improving detection accuracy and efficiency, it often encounters difficulties with contextual and rotational variations in aerial imagery. We propose a novel Context and Orientation Correction (COC) framework, which includes two innovative modules: a context correction module and an orientation correction module. The context correction module utilizes style normalization to guide the model in identifying atypical objects within specific contextual scenes by mitigating contextual disparities between instances and refining contextual information. Additionally, the orientation correction module aims to reduce feature distance between instances with varying orientations, leveraging contrastive learning to ensure consistent object representations. Furthermore, we introduce a category-aware aggregation loss to enhance similarity in feature representations of objects from the same category, thereby addressing the class collision issue commonly associated with contrastive learning. Our COC framework achieves 27.6% mean Average Precision and 59.8% mean Average Precision on the Detection in Optical Remote Sensing Image (DIOR) and Northwestern Polytechnical University Very High Resolution 10. v2 (NWPU VHR-10.v2) datasets, respectively, demonstrating its significant effectiveness.}, archive = {J_EAAI}, author = {Le Yang and Shunzhou Wang and Xuerong Wang and Shutong Wang and Yuting Lu and Binglu Wang}, doi = {10.1016/j.engappai.2025.112110}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112110}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Contextual and orientation correction modules enhance weakly-supervised aerial object detection in remote sensing images}, volume = {161}, year = {2025}, }
• (2025). Towards detection of anomalous cosmic ray signals for observations acquired from cosmic ray extremely distributed observatory mobile detectors. EAAI, 161, 112109. (www)

  @article{ , abstract = {In this work, we propose and test a method that allows the detection of anomalous cosmic ray signals acquired using Complementary Metal-Oxide-Semiconductor detectors. The method uses unsupervised embedding based on Principal Component Analysis which we named Eigenhits in apparent analogy to Eigenfaces. The embedding generated using Eigenhits allows the detection of potential anomalies, defined as images whose position described by the embedding relative to a given measure is above a certain distance threshold from other images. Thus, the problem of anomaly detection was reduced to the problem of detecting outliers which can be solved, for example, using clustering algorithms. We conducted tests of our approach on the Cosmic Ray Extremely Distributed Observatory (CREDO) dataset containing 13168 images and obtained satisfactory results demonstrating the stability and effectiveness of the method. The embedding generation method we propose in this paper and the evaluation of its effectiveness in detecting anomalies in images of cosmic ray events from CREDO dataset is a pioneering study with many critical applications.}, archive = {J_EAAI}, author = {Tomasz Hachaj and Łukasz Bibrzycki and Marcin Piekarczyk and Olaf Bar and Michał Niedźwiecki and Sławomir Stuglik and Piotr Homola and Dmitriy Beznosko and David Alvarez-Castillo and Bożena Poncyljusz and Ophir Ruimi and Oleksandr Sushchov and Krzysztof Rzecki and CREDO collaboration}, doi = {10.1016/j.engappai.2025.112109}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112109}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards detection of anomalous cosmic ray signals for observations acquired from cosmic ray extremely distributed observatory mobile detectors}, volume = {161}, year = {2025}, }
• (2025). An uncertainty-aware safe-evolving reinforcement learning algorithm for decision-making and control in highway autonomous driving. EAAI, 161, 112108. (www)

  @article{ , abstract = {Rule-based and optimization-based approaches face challenges in decision-making and control for autonomous vehicles (AVs) in dynamic and complex highway scenarios. In contrast, reinforcement learning (RL) offers a more flexible and adaptable data-driven solution by allowing AVs to learn optimal actions through interactions with the environment, without requiring predefined rules or explicit programming. However, in real-world highway environments characterized by uncertainty, RL algorithm encounter difficulties in ensuring stability and safety. To address these challenges, this paper proposes an uncertainty-aware safe-evolving RL algorithm that integrates internal stability, external stability, and provable safety mechanisms. The internal stability mechanism ensures consistent performance improvements with high probability during policy updates itself, while the external stability leverages a benchmark policy as a reference to ensure the current policy performs at least as well as, if not better than, the benchmark. Furthermore, an action projection mechanism and a mixed learning procedure are incorporated to make minimal modifications to the learned policy, ensuring safety while supporting stable learning from both safe and original actions. The results show that the proposed algorithm maintains stability and safety throughout the learning process, achieves final performance comparable to traditional RL methods, and delivers higher training efficiency in a complex dynamic highway scenario in simulation. This suggests that the algorithm offers a viable solution for self-evolving systems in uncertain real-world environments, where traditional approaches may struggle.}, archive = {J_EAAI}, author = {Ping Lu and Sunan Zhang and Feihong Tan and Fulin Zhang and Yuxiang Feng and Bo Hu}, doi = {10.1016/j.engappai.2025.112108}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112108}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An uncertainty-aware safe-evolving reinforcement learning algorithm for decision-making and control in highway autonomous driving}, volume = {161}, year = {2025}, }
• (2025). A deep learning-driven network for joint low-light enhancement and deblurring in maritime surveillance systems. EAAI, 161, 112107. (www)

  @article{ , abstract = {—Shipboard video surveillance systems are crucial for maritime environmental perception. However, shipborne cameras are often affected by challenges such as ship rolling and low illumination, causing motion blur and low-light degradation in captured images. These degradations significantly reduce the accuracy and robustness of sea-surface obstacle detection. To address this challenge, this paper presents a deep learning-based joint network model designed to simultaneously deblur and enhance low-light maritime images. Unlike existing step-by-step approaches that treat deblurring and low-light enhancement as separate tasks, the proposed model employs an encoder-decoder architecture to jointly address both tasks within a unified framework, thereby overcoming the limitations of separate processing. Furthermore, the model incorporates a novel adaptive wavelet curve attention mechanism and an extended spatial pyramid pooling module with dynamic global context to mitigate detail loss caused by sea surface reflections and artifacts in dynamic backgrounds. Additionally, a joint loss function is exploited to further enhance image restoration quality. Experiments on a self-constructed synthetic maritime dataset demonstrate that the proposed model achieves a Peak Signal-to-Noise Ratio (PSNR) of 31.81, a Structural Similarity Index Measure (SSIM) of 0.812, and a Learned Perceptual Image Patch Similarity (LPIPS) of 0.246, significantly outperforming existing competing methods. Moreover, evaluations on real-world degraded maritime images confirm the model's strong generalization ability and robustness. This advancement offers practical benefits for improving obstacle perception in autonomous ships navigating complex maritime environments.}, archive = {J_EAAI}, author = {Zeyuan Shao and Yong Yin and Hongguang Lyu and Qianfeng Jing and Tao Cheng}, doi = {10.1016/j.engappai.2025.112107}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112107}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep learning-driven network for joint low-light enhancement and deblurring in maritime surveillance systems}, volume = {161}, year = {2025}, }
• (2025). A double hierarchy hesitant fuzzy forecasting model considering the influence of investor emotion and the co-movement of stock markets. EAAI, 161, 112106. (www)

  @article{ , abstract = {The stock market is a complex system influenced by various internal and external factors. Effective stock index forecasting models must balance interpretability with the ability to integrate multi-source information. While traditional fuzzy models offer high interpretability, they struggle to incorporate multiple dynamic factors. This study proposes a novel forecasting model based on double hierarchy hesitant fuzzy linguistic term sets (DHHFLTS), which integrates investor sentiment and international market co-movement. By constructing fuzzy logic rules, the model captures complex interactions underlying index fluctuations and enhances predictive accuracy in volatile environments. To evaluate its effectiveness, the model was tested on time series data from three major indices: the Shanghai Stock Exchange Composite Index (SSEC), Taiwan Capitalization Weighted Stock Index (TAIEX), and Financial Times Stock Exchange 100 Index (FTSE 100). The Root Mean Square Error (RMSE) achieved was 32.06 for the SSEC, 69.74 for the TAIEX, and 25.27 for the FTSE 100. Compared with existing benchmark methods, the proposed model demonstrates superior predictive accuracy and generalization performance. These findings confirm the model's capability in capturing multidimensional uncertainty and temporal patterns, highlighting its potential for intelligent financial forecasting and decision-making.}, archive = {J_EAAI}, author = {Liu Zhengmin and Du Chuantao and Zhang Jihao and Gao Shanshan and Liu Peide}, doi = {10.1016/j.engappai.2025.112106}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112106}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A double hierarchy hesitant fuzzy forecasting model considering the influence of investor emotion and the co-movement of stock markets}, volume = {161}, year = {2025}, }
• (2025). Style prompt tuning for bridging visual gaps in autonomous driving. EAAI, 161, 112105. (www)

  @article{ , abstract = {Artificial intelligence models for semantic segmentation and image classification in autonomous driving must maintain reliability across adverse conditions such as rain, fog, snow, and nighttime. However, models trained on only clear daytime images often fail to generalize under such domain shifts. Existing unsupervised domain adaptation (UDA) methods employ image-level style transfer using generative adversarial networks (GANs) or diffusion models, which necessitate paired data and risk altering content. Therefore, this study proposes Style Prompt Tuning, a novel UDA framework that utilizes image–text models to automatically generate and optimize textual prompts representing target-domain styles. These prompts guide a U-Net-based style network to synthesize source images in the target style while preserving their semantic content. Our approach employs clustering within the Contrastive Language—Image Pretraining (CLIP) embedding space and a composite loss function, including content, style, transfer, patch, and total variation terms to optimize prompt quality. The generated stylized images augment the source dataset and are used to train more robust task models. Experiments on semantic segmentation benchmarks (Cityscapes-to-Adverse Conditions Dataset with Correspondences (ACDC), DarkZurich, BDD100k-night, and Nighttime Driving) and image classification (Visual Domain Adaptation 2017) reveal our approach to achieve improvements of +3.4 mean intersection-over-union (mIoU) and +0.8% accuracy over prior UDA methods. These results highlight our method’s practical effectiveness for real-world autonomous driving applications under visually challenging scenarios.}, archive = {J_EAAI}, author = {Suyeon Cha and Giyun Choi and Minji Kwak and Jongwon Choi}, doi = {10.1016/j.engappai.2025.112105}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112105}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Style prompt tuning for bridging visual gaps in autonomous driving}, volume = {161}, year = {2025}, }
• (2025). Enhancing fuzzy cognitive map convergence through supervised and unsupervised learning algorithms: A case study of operational risk assessment in power distribution networks. EAAI, 161, 112104. (www)

  @article{ , abstract = {Fuzzy Cognitive Maps (FCMs) are commonly used for modeling complex systems. However, their convergence challenges significantly limit their accuracy and reliability, especially in operational risk assessment. To address this issue, the current study proposes a novel approach that integrates advanced supervised and unsupervised learning algorithms: specifically, the Mesh Adaptive Direct Search (MADS) and Genetic Algorithm (GA). To meet the critical need for accurate risk modeling in power distribution networks, the proposed methodology utilizes ten years of time-series data from the Yazd Power Distribution Network as a real-life case study. This optimizes risk relationships and reduces convergence errors. The main contributions of this research are: (1) the development of an integrated FCM-based framework that improves convergence stability and accuracy through advanced learning algorithms, (2) the demonstration of the effectiveness of MADS in achieving faster convergence and lower error rates compared to GA, and (3) the provision of a data-driven, scalable solution for risk prioritization and decision-making in complex and dynamic systems. The results show a significant decrease in convergence error from 0.126 to 0.005, allowing for more precise risk analysis and mitigation strategies.}, archive = {J_EAAI}, author = {Elham Fallah Baghemoortini and Davood Shishebori and Mustafa Jahangoshai Rezaee and Armin Jabbarzadeh}, doi = {10.1016/j.engappai.2025.112104}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112104}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing fuzzy cognitive map convergence through supervised and unsupervised learning algorithms: A case study of operational risk assessment in power distribution networks}, volume = {161}, year = {2025}, }
• (2025). Semi-supervised generative adversarial network for plant leaf disease detection. EAAI, 161, 112103. (www)

  @article{ , abstract = {Fully supervised plant leaf disease segmentation methods based on convolutional neural networks (CNN) require large quantities of labeled images for training, which are time-consuming and labor-intensive to obtain in practical scenarios. Moreover, most diseases of plant leaf exhibit indistinct edge information and background noise interference, which hinders precise disease localization. To overcome these issues, a semi-supervised generative adversarial network (SSGAN) is proposed for plant leaf disease inspection. Firstly, to learn comprehensive semantic features from limited data, a semi-supervised method based on generative adversarial networks is developed to enhance the interaction of features among labeled and unlabeled disease images. Secondly, a boundary feature attention module (BFAM) is proposed to enhance edge detail representation, which makes the model pay more attention to the boundary feature of plant leaf diseases. Finally, a background noise suppression module (BNSM) is proposed to bolster the differentiation between the normal and disease areas so as to alleviate the adverse impact of background noise. The effectiveness of SSGAN is verified on two plant leaf disease datasets. The mean intersection over union (mIoU) on the apple and tomato leaf datasets improves by 2.07 % and 2.28 % respectively, compared with the suboptimal method. The testing results of experiments show that SSGAN achieves great performance on plant leaf disease segmentation under limited labeled data.}, archive = {J_EAAI}, author = {Lixiang Zhao and Jun Hao and Demin Li and Jianbo Yu}, doi = {10.1016/j.engappai.2025.112103}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112103}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised generative adversarial network for plant leaf disease detection}, volume = {161}, year = {2025}, }
• (2025). Industrial device-aided data collection for real-time rail defect detection via a lightweight network. EAAI, 161, 112102. (www)

  @article{ , abstract = {Rail defect detection is challenging due to the diverse and irregular nature of defects, along with the limited availability of high-quality datasets. Existing methods struggle with effectively capturing multi-scale features for proper feature allocation and preserving crucial details in deep networks, leading to incomplete defect representation and reduced accuracy. To address these limitations, we propose Rail Defect Detection Network (REDNet), a lightweight deep learning model specifically designed for real-time rail defect detection in manufacturing and maintenance applications. We design the Multi-Scale Deep Feature Aggregation (MSDFA) module to enhance semantic consistency modeling and achieve more precise feature fusion. We develop the Adaptive Task Decomposition Head (ATDH) to address dynamic feature allocation, and we introduce the Reversible Column Network (RevCol) as the backbone to enhance feature extraction and ensure information reconstruction. Additionally, we developed a high-quality dataset using specialized equipment to address data scarcity and utilized it for training. REDNet achieved a high mean Average Precision at an Intersection over Union (IoU) threshold of 0.50 (mAP50) of 94.1% with exceptional real-time performance at 204.1 frames per second (FPS), while keeping an efficient design of 5.70 million parameters and surpassing state-of-the-art (SOTA) methods in accuracy and speed. These features make it suitable for defect detection, facilitating engineering deployment, and improve quality control in rail manufacturing and maintenance. Generalization tests on the public Microsoft Common Objects in Context (MS COCO) dataset yielded a mean Average Precision across IoU thresholds from 0.50 to 0.95 (mAP50–95) of 46.8%, further confirming the effectiveness of REDNet.}, archive = {J_EAAI}, author = {Qing Dong and Tianxin Han and Gang Wu and Lina Sun and Min Huang and Fu Zhang}, doi = {10.1016/j.engappai.2025.112102}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112102}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Industrial device-aided data collection for real-time rail defect detection via a lightweight network}, volume = {161}, year = {2025}, }
• (2025). A quantitative multi-dimensional resilience framework in process industries. EAAI, 161, 112100. (www)

  @article{ , abstract = {Process industries face significant safety challenges arising from the handling of hazardous materials, tightly coupled operations, and complex system interdependencies. Addressing these challenges requires management strategies that are both proactive and holistic rather than purely reactive. Resilience engineering embodies this approach by emphasizing a system's ability to anticipate disturbances, adapt under stress, and recover effectively. This study introduces a comprehensive, multi-dimensional framework for assessing resilience in process industries. The framework was developed through content analysis of expert interviews, a systematic literature review, and application of the spherical fuzzy Delphi method. Results revealed five primary dimensions of resilience: the organizational dimension, which encompasses leadership commitment, safety culture, and communication processes; the human resources dimension, focusing on workforce competence, training, and well-being; the individual dimension, addressing cognitive readiness, situational awareness, and decision-making skills; the risk management system dimension, which includes hazard identification, risk assessment, emergency preparedness, and continuous monitoring; and the technical dimension, covering equipment reliability, system redundancy, automation safeguards, and maintenance practices. Within these dimensions, factors such as leadership, effective communication, workforce capability, robust risk governance, and technical robustness emerged as particularly influential. By integrating these dimensions into a unified framework, our study advances theoretical understanding of resilience engineering and offers practical guidance for enhancing safety performance in process plants. Furthermore, this framework lays a foundation for future research aimed at developing standardized assessment tools, evaluating long-term outcomes of resilience interventions, and exploring the integration of emerging technologies—such as artificial intelligence, the internet of things, and advanced automation—into resilience engineering practices.}, archive = {J_EAAI}, author = {Mojtaba Emkani and Moslem Alimohammadlou and Esmaeil Zarei and Mehdi Jahangiri and Mojtaba Kamalinia}, doi = {10.1016/j.engappai.2025.112100}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112100}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A quantitative multi-dimensional resilience framework in process industries}, volume = {161}, year = {2025}, }
• (2025). Mutual transfer learning for cuff-less blood pressure estimation using photoplethysmography-based visibility graphs. EAAI, 161, 112099. (www)

  @article{ , abstract = {Cuff-less blood pressure (BP) estimation is critical to cardiovascular disease prevention and management. Photoplethysmography (PPG)-based monitoring technology offers advantages over cuff-based devices, including portability, lower power consumption, and faster measurements. However, current deep learning methods for BP estimation from PPG signals are limited by their analysis only from one-dimensional perspectives, failing to exploration of the underlying physiological patterns of cross-domain visual representations based on higher dimensional perspectives. Furthermore, conventional knowledge distillation techniques necessitate unidirectional knowledge transfer from pre-trained models, rendering it challenging to obtain feedback on the learning state of small networks for optimizing and adjusting the training process. This is inadequate for acquiring a profound understanding of the intricate mapping relationship between PPG and BP values. Therefore, this work presents a novel Transformer-based mutual transfer learning framework (MTL) that estimates BP values from phase-space reconstructed PPG signals using a multi-field complementary approach. The proposed MTL method leverages four phase-space reconstruction techniques to convert PPG signals into visibility graphs (VGs) that provide rich time-variant information. Furthermore, the joint optimization strategy with multiple losses of the soft label and structural knowledge learning enables us to transfer pre-trained knowledge from heterogeneous Transformer models and obtain cumulative multi-field complementary VG features during the fine-tuning process. We evaluate our MTL on three datasets of 1375 subjects using a subject-wise data-splitting paradigm based on five-fold cross-validation, achieving a state-of-the-art performance with estimation errors of 0.50 ± 4.94 millimeter of mercury (mmHg) and 0.21 ± 2.63 mmHg for systolic and diastolic BP, respectively. Our proposed end-to-end MTL offers a computationally efficient solution and elegant generalization ability for BP estimation using PPG-based VGs, providing a novel and innovative approach to BP monitoring that can advance cardiovascular disease prevention and management.}, archive = {J_EAAI}, author = {Chenbin Ma and Zhenchang Liu and Peng Zhang and Lishuang Guo and Haonan Zhang and Zeyu Liu and Guanglei Zhang}, doi = {10.1016/j.engappai.2025.112099}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112099}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mutual transfer learning for cuff-less blood pressure estimation using photoplethysmography-based visibility graphs}, volume = {161}, year = {2025}, }
• (2025). Groundwater parameters estimation: A hybrid approach of convolutional layers with asynchronous and distributed bio-inspired algorithms. EAAI, 161, 112098. (www)

  @article{ , abstract = {This research focuses on aquifer hydraulic parameters estimation using bio-inspired algorithms since they can tackle groundwater model non-linearities. We propose two novel hybrid frameworks that combine the advantages of convolutional layers (CL) to enhance pattern recognition with heuristic search of Particle Swarm Optimization (PSO) and Differential Evolution (DE) algorithms. These integrations are implemented using an asynchronous and distributed approach to address efficiency issues in large-scale problems, resulting in ADPSO-CL (Asynchronous and Distributed Particle Swarm Optimization with Convolutional Layers) and ADDE-CL (Asynchronous and Distributed Differential Evolution with Convolutional Layers). The distributed method employs virtual machines, where a server generates and assigns particles to workers, which run in parallel with asynchronous iterative solution exchanges. We assess different algorithm configurations in an integrated water management model by coupling two software: Water Evaluation and Planning (WEAP) and MODFLOW. Results indicate that ADPSO-CL outperforms ADDE-CL by demonstrating more stable asynchronous communication, with fewer incomplete experiments (more than one worker was disconnected before completing all iterations), 33% in contrast to 71%. Additionally, produces results closer to the expected values, with mean absolute percentage error (MAPE) values of 78.25% for hydraulic conductivity ( K ) and 55.56% for specific yield ( S y ), compared to 299% and 209% in ADDE-CL. Moreover, ADPSO-CL has the fastest convergence rate, achieving efficient results in about half of the total iterations. This study introduces a novel and scalable architecture for intricate simulation–optimization problems, demonstrating its potential for future applications in real-world water resources planning and management.}, archive = {J_EAAI}, author = {Kiara Tesen and Hermilo Cortés and Sebastián Vicuña and Edmundo Molina-Perez and Francisco Suárez}, doi = {10.1016/j.engappai.2025.112098}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112098}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Groundwater parameters estimation: A hybrid approach of convolutional layers with asynchronous and distributed bio-inspired algorithms}, volume = {161}, year = {2025}, }
• (2025). Forecasting power generation: A novel two-dimensional logistic fractional grey model. EAAI, 161, 112097. (www)

  @article{ , abstract = {Accurate prediction of power generation is essential for the stable operation of power systems, sustained economic growth, and harmonious social development. However, power systems exhibit chaotic characteristics such as complex and nonlinear dynamics during operation; for this reason, this study exploits the fact that two-dimensional logistic mapping helps with the recognition and control of the complex high-dimensional dynamics of the system. Additionally, the fractional-order chaotic system offers a more universal means to describe the chaotic phenomena, etc., and introduces the modelling mechanism of the two-dimensional logistic model and the fractional-order derivative theory into the grey model. A novel two-dimensional logistic fractional-order grey model is developed. The new model can simultaneously analyze two parallel major factors that evolve and are interrelated, and the fractional-order derivatives and nonlinear terms accurately portray the historical state and dynamic evolution process of the system, which can explore the evolution law in the time series of power generation more adequately. Furthermore, the particle swarm optimization algorithm is used to optimize the fractional order derivatives and parameters so that the adjustable model parameters can better capture the change pattern of the original power generation system data. The new model is applied to China's power generation prediction, and its validity is verified from three different perspectives through two different power generation modes as the research objects. Moreover, the new model outperforms the other seven models in its calculations. Finally, the new model effectively forecasts China's power generation for the next five years.}, archive = {J_EAAI}, author = {Mingyue Weng and Huiming Duan and Derong Xie}, doi = {10.1016/j.engappai.2025.112097}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112097}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forecasting power generation: A novel two-dimensional logistic fractional grey model}, volume = {161}, year = {2025}, }
• (2025). A deep reinforcement learning optimization algorithm based on heterogeneous graph neural network for hybrid flow shop scheduling problem with finite transportation resources. EAAI, 161, 112096. (www)

  @article{ , abstract = {In many actual process industries, transportation equipment represented by automated guided vehicles (AGVs) is widely used. However, current related research rarely considers the coordinated scheduling optimization of AGVs and production equipment. To this end, the hybrid flow shop scheduling problem (HFSP) with finite transportation resources (HFSP-FTR) in process industries is considered in this paper. Since it is necessary to consider the coordinated scheduling optimization of machines and AGVs at the same time, an end-to-end deep reinforcement learning scheduling method based on a heterogeneous graph neural network is proposed. First, a heterogeneous graph model capable of expressing any HFSP-FTR instance is constructed to visually represent the allocation relationship among jobs, machines, and AGVs. Second, a Markov decision process is formulated for solving HFSP-FTR, which specifically encompasses state features based on the heterogeneous graph model and a novel representation method for composite scheduling actions. Then, a novel heterogeneous graph neural network framework that can parallelly embed nodes of various types is proposed, aiming to derive graph-level embeddings for HFSP-FTR instances. Finally, a policy network is designed to obtain the probability distribution of each composite scheduling action being executed, and a deep reinforcement learning framework based on the multi-actor network is proposed for training. The results of numerical simulation experiments conducted on test instances with varying types demonstrate that the proposed method can acquire effective and highly generalized scheduling strategies in terms of makespan performance, outperforming other scheduling algorithms that are widely applied regarding the quality of scheduling solutions and other aspects.}, archive = {J_EAAI}, author = {Yejian Zhao and Xiaochuan Luo and Weixiang Xu and Yulin Zhang}, doi = {10.1016/j.engappai.2025.112096}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112096}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep reinforcement learning optimization algorithm based on heterogeneous graph neural network for hybrid flow shop scheduling problem with finite transportation resources}, volume = {161}, year = {2025}, }
• (2025). Unknown fault incremental learning based on shapelet prototypical network for streaming industrial signals. EAAI, 161, 112094. (www)

  @article{ , abstract = {Unknown faults represent faults that have never occurred before, they are constantly emerging due to the changing environments and operations in industrial processes. It is a challenge for existing fault diagnosis methods to continually detect unknown faults and effectively classify known faults in streaming industrial signals. This article proposes an unknown fault incremental learning method for streaming industrial signals. In this work, shapelet prototypical embedding combined with a memory distance matrix is employed to embed streaming industrial signals into a discriminative feature space. Therefore, the category information in the signals can be extracted and is not limited by the size of the sliding window. Besides, a new training paradigm based on meta-learning by sampling simulated-incremental tasks is proposed to obtain generalizable shapelets. Moreover, based on the new training paradigm, the meta-discovery module is proposed to continually detect unknown faults, and the meta-calibrate module can calibrate all prototypes into a distinguishable space. Experiments on the simulated streaming time series, benchmark Tennessee Eastman process, and real-world aluminum electrolysis process illustrate the superiority of the proposed method in terms of accuracy and interpretability. The code is available in https://github.com/XiaoxueWan/UFIL.git .}, archive = {J_EAAI}, author = {Xiaoxue Wan and Lihui Cen and Xiaofang Chen and Yongfang Xie and Zhaohui Zeng}, doi = {10.1016/j.engappai.2025.112094}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112094}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unknown fault incremental learning based on shapelet prototypical network for streaming industrial signals}, volume = {161}, year = {2025}, }
• (2025). Adaptive attention graph convolution network with normalized embedded gaussian for rapid serial visualization presentation decoding. EAAI, 161, 112093. (www)

  @article{ , abstract = {Graph convolutional networks (GCNs) have been widely used in Brain Computer Interface (BCI) and have shown great prowess in identifying electroencephalogram (EEG) spatiotemporal features. However, most GCNs learn channels topological relationship by fixed adjacency matrix. This lacks connectivity strength information and ignores the data dependency. This paper proposes a data-driven adjacency matrix based on normalized embedded Gaussian function, and constructs a Gaussian-Adaptive Attention Graph Convolution Network (Gaussian-AAGCN). Brain regions connectivity is calculated by normalized embedded Gaussian function, and the topological relationship is adaptively learned by input data in a data-driven manner. This data-driven adaptive adjacency matrix avoids brain activity information loss caused by fixed adjacency matrix and improves the flexibility of graph construction. Convolutional block attention module (CBAM) is introduced to adaptive feature refinement in two independent dimensions, improving model representation ability. Experimental results show that the average area under curve (AUC), true positive rate (TPR) and false positive rate (FPR) of Gaussian-AAGCN on 14 subjects are 93.52 %, 91.59 %, and 4.58 % respectively. Compared to Transformer, Event-Related Potential Capsule Network (ERP-CapsNet), Electroencephalogram Convolutional Neural Network (EEGNet), and Multi-Granularity Information Fusion Network (MGIFNet), the AUC of Gaussian-AAGCN is higher by 18.02 %, 5.32 %, 2.62 %, and 1.82 %, respectively. Using the adaptive adjacency matrix, the model AUC and TPR are increased by about 4.8 % and 7.8 % respectively. After integrating CBAM, the AUC and TPR increased by about 3.5 % and 8 % respectively.}, archive = {J_EAAI}, author = {Mengyuan Zhao and Qingsong Ai and Kun Chen and Quan Liu and Sheng Quan Xie and Li Ma}, doi = {10.1016/j.engappai.2025.112093}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112093}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive attention graph convolution network with normalized embedded gaussian for rapid serial visualization presentation decoding}, volume = {161}, year = {2025}, }
• (2025). GREEN: Graph reasoning enhanced encoder network for social intention-aware forecast of vessel navigating trajectory. EAAI, 161, 112092. (www)

  @article{ , abstract = {Recent studies have demonstrated that graph neural network (GNN) excel in interactive modeling, particularly by enhancing reasoning capabilities through the construction of interaction graphs, and the application of graph convolutional network (GCN). However, these methods remain heavily dependent on prior features, which limits the ability of vessels to accurately infer each other’s decision-making intentions in encounter situations. Therefore, this paper proposes graph reasoning enhanced encoder network (GREEN), which enables social community insight on multi-ships interactions, and latent intention aware for interactive trajectory forecast. The architecture of GREEN includes two layers of trajectory-graph-embedded encoders layer (TGEEL) and multi-graphs-gated fusion layer (MGGFL). In TGEEL, we respectively design intention trajectory generator unit for graph-embedded multi-ships trajectory representation via random walk process, and latent intention generator unit to forecast future navigating intentions via variational autoencoder. In MGGFL, we also design two units, where one unit is in charge of social aware construction and spatio-temporal features extraction based on interactive multi-graphs, and the other one is responsible for gated fusion of multi-scale features and forecast future trajectory using spatio-temporal attention. Experimental results across five datasets were collected from real-world maritime environments, which demonstrated that GREEN achieved improvements of 39.98%, 38.57%, 38.65%, and 63.33% in average displacement error , final displacement error, maximum displacement error, and miss rate, compared with state-of-the-art methods. The paper highlights GREEN robust potential in complex navigational scenarios, and provides pivotal support for advancing the development of intelligent maritime systems.}, archive = {J_EAAI}, author = {Junhao Jiang and Yi Zuo}, doi = {10.1016/j.engappai.2025.112092}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112092}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {GREEN: Graph reasoning enhanced encoder network for social intention-aware forecast of vessel navigating trajectory}, volume = {161}, year = {2025}, }
• (2025). DiffAT: Effective data augmentation with diffusion models for time series forecasting. EAAI, 161, 112091. (www)

  @article{ , abstract = {Data augmentation offers a promising solution to data scarcity in deep learning-based time series forecasting. However, current approaches face dual limitations (1) Hand-designed methods (e.g., cropping/masking): often disrupt the continuity of vital temporal patterns (such as seasonal and trend) by introducing abrupt pattern discontinuities; (2) Generative models often face difficulties in preserving task-critical features that are essential for prediction, especially when aiming to generate diverse augmented series. To tackle these dilemmas, we propose a novel conditional diffusion-based data augmentation framework, named DiffAT, for time series forecasting tasks. DiffAT synergizes: (1) Patch-wise masking reconstruction to capture structural invariants (such as autocorrelation and causality), and (2) encoding hand-designed augmentation prototypes for guiding diversity-preserving generation. DiffAT achieves dual enhancement: maintaining continuity of temporal patterns through progressive denoising process and exposing latent invariant patterns via guided diversity injection. We validate the efficacy of DiffAT through extensive experiments on seven real-world datasets, by comparing DiffAT with six state-of-the-art time series data augmentation methods. The results indicate our method can boost the forecasting performance of Autoformer by up to 6.49 % in 26/28 cases and improve forecasting performance of LightTS by up to 3.11 % in 23/28 cases on 7 real-world benchmarks. Extensive experiments also indicate that DiffAT can improve the accuracy of forecasting models in few-shot scenario (with 1 % training data) in 54/60 cases. We will release the source code upon publication.}, archive = {J_EAAI}, author = {Yang Yu and Ruizhe Ma and Wenbo Gu and Zongmin Ma}, doi = {10.1016/j.engappai.2025.112091}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112091}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DiffAT: Effective data augmentation with diffusion models for time series forecasting}, volume = {161}, year = {2025}, }
• (2025). Evaluating reinforcement learning-based neural controllers for quadcopter navigation in windy conditions. EAAI, 161, 112090. (www)

  @article{ , abstract = {Accurate quadcopter navigation under windy conditions remains challenging for traditional control methods, especially in the presence of unpredictable wind gusts and strict navigational constraints. This paper evaluates Deep Reinforcement Learning (DRL) based controllers under such conditions, analysing the impact of wind domain randomisation, multi-goal training, enhanced state representations with explicit wind information, and the use of temporal data to capture affecting dynamics over time. Experiments in the AirSim simulator across four trajectories — evaluated under both no-wind and windy conditions — demonstrate that DRL-based controllers outperform classical methods, particularly under stochastic wind disturbances. Moreover, we show that training a DRL agent with domain randomisation improves robustness against wind but reduces efficiency in no-wind scenarios. However, incorporating wind information into the agent’s state space enhances robustness without sacrificing performance in wind-free settings. Furthermore, training with stricter waypoint constraints emerges as the most effective strategy, leading to precise trajectories and improved generalisation to wind disturbances. To further interpret the learned policies, we apply Shapley Additive explanations analysis, revealing how different training configurations influence the agent’s feature importance. These findings underscore the potential of DRL-based neural controllers for resilient autonomous aerial systems, highlighting the importance of structured training strategies, informed state representations, and explainability for real-world deployment.}, archive = {J_EAAI}, author = {Alain Andres and Aritz D. Martinez and Sümer Tunçay and Ignacio Carlucho}, doi = {10.1016/j.engappai.2025.112090}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112090}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evaluating reinforcement learning-based neural controllers for quadcopter navigation in windy conditions}, volume = {161}, year = {2025}, }
• (2025). Pixel-level semantics boosted fine-grained bird image classification. EAAI, 161, 112089. (www)

  @article{ , abstract = {Fine-grained bird image classification (FBIC) is crucial for endangered bird conservation and biodiversity research. However, existing methods often struggle to capture detailed features and manage the interference caused by complex backgrounds. To address these challenges, we propose a novel Pixel-Level Semantic Boosted Fine-Grained Bird Image Classification (PFIC) framework, which enhances fine-grained bird image classification by incorporating pixel-level semantic information. PFIC consists of two core components: the Grouped Detail Enhancement (GDE) module and the Background–Foreground Enhancement (BFE) strategy. GDE integrates multi-level pixel-level semantic information, derived from a segmentation feature extractor, into classification features via two submodules: grouped aggregation and detail enhancement. This approach enhances the model’s ability to capture fine-grained details. BFE augments training samples by restricting background ranges and applying random shifts to foreground objects, thereby improving the model’s capability to recognize foreground objects in complex environments. Experimental results demonstrate that our method achieves state-of-the-art performance on the CUB-200-2011 and NABirds datasets. Additionally, further experiments on the Stanford Cars dataset validate the framework’s potential for generalization to other fine-grained image classification tasks.}, archive = {J_EAAI}, author = {Haoxiang Ma and Yongjian Deng and Bochen Xie and Jian Liu and Hai Liu and Youfu Li and Zhen Yang}, doi = {10.1016/j.engappai.2025.112089}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112089}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pixel-level semantics boosted fine-grained bird image classification}, volume = {161}, year = {2025}, }
• (2025). Transformer-based soft sensor of five day biochemical oxygen demand indicators in the municipal wastewater treatment process. EAAI, 161, 112088. (www)

  @article{ , abstract = {Five day biochemical oxygen demand (BOD 5 ) is usually used to measure whether wastewater treatment meets the standard. Previously, the development of machine learning technology was applied to predict BOD 5 for future planning of wastewater treatment. In real wastewater treatment, the monitoring of BOD 5 often lags behind for a long time, so neural network-based soft sensor methods are widely used in this field. However, the currently used BOD 5 soft sensor model has the problem of ignoring the synergistic effect of the model and stable information transmission. In order to monitor the BOD 5 concentration more accurately, a Transformer network model with embedded Long Short-Term Memory (LSTM) network is proposed in this article : LSTM is embedded into the calculation of the transformer multi-head attention, and the output method of the multi-head attention is improved so that the attention heads are output one by one in a serial manner, so it is named LSTM-Serial-Transformer (LSformer). Compared with previous models, LSformer optimizes input information through LSTM and Transformer to achieve synergistic effect, thus better capturing the characteristics of BOD 5 time series data, while reducing the risk of gradient vanishing or gradient exploding during data transmission to improve the stability of the model. Finally, the model was verified to have better soft sensor performance on a benchmark simulation model and a real-world dataset.}, archive = {J_EAAI}, author = {Xiaoling Zhang and Zhi Qi and Peng Chang and Zhiqi Hu}, doi = {10.1016/j.engappai.2025.112088}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112088}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer-based soft sensor of five day biochemical oxygen demand indicators in the municipal wastewater treatment process}, volume = {161}, year = {2025}, }
• (2025). ForgetMe: Benchmarking the selective forgetting capabilities of generative models. EAAI, 161, 112087. (www)

  @article{ , abstract = {The widespread adoption of diffusion models in image generation has increased the demand for privacy-compliant unlearning. However, due to the high-dimensional nature and complex feature representations of diffusion models, achieving selective unlearning remains challenging, as existing methods struggle to remove sensitive information while preserving the consistency of non-sensitive regions. To address this, we propose an Automatic Dataset Creation Framework based on prompt-based layered editing and training-free local feature removal, constructing the ForgetMe dataset and introducing the Entangled evaluation metric. The Entangled metric quantifies unlearning effectiveness by assessing the similarity and consistency between the target and background regions and supports both paired ( Entangled-D ) and unpaired ( Entangled-S ) image data, enabling unsupervised evaluation. The ForgetMe dataset encompasses a diverse set of real and synthetic scenarios, including CUB-200-2011 (Birds), Stanford-Dogs, ImageNet, and a synthetic cat dataset. We apply LoRA fine-tuning on Stable Diffusion to achieve selective unlearning on this dataset and validate the effectiveness of both the ForgetMe dataset and the Entangled metric, establishing them as benchmarks for selective unlearning. Our work provides a scalable and adaptable solution for advancing privacy-preserving generative AI. Code is available at: https://github.com/YuZhenyuLindy/ForgetMe .}, archive = {J_EAAI}, author = {Zhenyu Yu and Mohd Yamani Idna Idris and Pei Wang and Yuelong Xia and Yong Xiang}, doi = {10.1016/j.engappai.2025.112087}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112087}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ForgetMe: Benchmarking the selective forgetting capabilities of generative models}, volume = {161}, year = {2025}, }
• (2025). A novel data generation scheme for surrogate modelling with deep operator networks. EAAI, 161, 112086. (www)

  @article{ , abstract = {Operator-based neural network architectures such as deep operator networks have emerged as a promising tool for the surrogate modelling of physical systems. In general, the training data for operator surrogate modelling are generated by solving partial differential equations using the finite element method. The computationally intensive nature of data generation is one of the major bottlenecks in deploying these surrogate models, hindering the deployment of these surrogate models in practical applications. In this study, we propose a novel methodology to alleviate the computational burden associated with training data generation for deep operator networks. Unlike the existing literature, the proposed framework for data generation does not use any partial differential equation integration strategy, such as the finite element method. In the proposed strategy, the primary field consistent with the boundary conditions is first generated randomly using Gaussian process regression. Thereafter, from the primary field, the input source field is calculated using second-order accurate finite difference techniques. The computation of the derivatives from the finite difference scheme is significantly less computationally intensive and easier to implement compared to integrating the underlying governing equations, thereby reducing the computational cost associated with generating training datasets. To validate the proposed approach, we employ heat equations as a model problem and develop the surrogate model for numerous boundary value problems.}, archive = {J_EAAI}, author = {Shivam Choubey and Shailendra Rahi and Birupaksha Pal and Manish Agrawal}, doi = {10.1016/j.engappai.2025.112086}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112086}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel data generation scheme for surrogate modelling with deep operator networks}, volume = {161}, year = {2025}, }
• (2025). Interpretable generalization diagnosis: Variational causal disentanglement-based coalitional game attribution network. EAAI, 161, 112084. (www)

  @article{ , abstract = {Interpretability and controllability are the keys to intelligent fault diagnosis in industrial scenarios. However, most intelligent methods have focused on post-hoc feature importance explanations, which cannot constrain the uncontrollable learning behavior or form an interpretable diagnosis process consistent to match expert experience. To address this problem, this paper proposes a variational causal disentanglement-based coalitional game attribution network (VCN) to achieve a controllable and interpretable modeling paradigm that can draw diagnostic conclusions with conforming to expert experience. This model enables independent causal factor disentanglement and model regularization during the training process, as well as fast explanation computation at test time. Variational causal disentanglement is defined and do-operations and variational inference processes are introduced to generate training data with independent causal factors for subsequent attribution. The coalitional game attribution establishes a coalitional game relationship between the model and the independent causal factors, and regularizes the model so that it converges to a state that conforms to logical axioms. Both laboratory and real-world data demonstrate the proposed method enables the provision of interpretable diagnostic conclusions while maintaining controllable performance. The code is available at: https://github.com/cyber-dogy/VCN-code .}, archive = {J_EAAI}, author = {Junwei Gu and Changhua Hu and Yu Wang and Mingquan Zhang and Yanzhuo Lin and Shangjing Peng and Dongdong Li}, doi = {10.1016/j.engappai.2025.112084}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112084}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpretable generalization diagnosis: Variational causal disentanglement-based coalitional game attribution network}, volume = {161}, year = {2025}, }
• (2025). A novel photovoltaic cell defect detection method: A deep learning model based on multi-scale enhanced feature extraction. EAAI, 161, 112083. (www)

  @article{ , abstract = {With the increasingly widespread application of clean energy technology, the global demand for photovoltaic cells is gradually increasing, and the status and defects of photovoltaic cells are also being taken seriously. However, there are various types of defects in photovoltaic cells, including those that are difficult to detect, making detection work somewhat challenging. This article proposes a novel improved model for defect detection in photovoltaic cells based on You Only Look Once (YOLO). Firstly, the Multi-Head Self-Attention (MHSA) mechanism is adopted to compensate for the shortcomings of the Cross Stage Partial Bottleneck with 2 Convolutions (C2f) module in channel feature extraction. Secondly, the Receptive-Field Coordinate Attention Convolutional (RFCAConv) operation module is utilized to expand the receptive field of defect detection in the model and increase the range of feature extraction. Finally, the Bi-directional Feature Pyramid Network (BiFPN) and Concat modules are integrated to enhance the feature relationships between the output connections of each module and strengthen the output results. The final experimental results demonstrate that the improved model achieves an accuracy of 90.3%, representing a 5.2% increase compared to the original model. Through experimental verification, the improved model proposed in this paper can not only detect defects in photovoltaic cells in electroluminescence (EL) images but also in infrared thermal images. The model has strong generalization ability.}, archive = {J_EAAI}, author = {Yu Gao and Zhanying Li and Yinghao Zhang and Kangye Zhang and Haoyang Yu}, doi = {10.1016/j.engappai.2025.112083}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112083}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel photovoltaic cell defect detection method: A deep learning model based on multi-scale enhanced feature extraction}, volume = {161}, year = {2025}, }
• (2025). Wind power prediction based on hybrid deep learning and monte carlo simulation. EAAI, 161, 112082. (www)

  @article{ , abstract = {This study proposes a hybrid deep learning model combining Variational Mode Decomposition (VMD), Convolutional Neural Networks (CNN), and a Three-Dimensional Gated Neural Network (TGNN) to enhance wind power prediction accuracy. VMD decomposes wind power time series into intrinsic mode functions, CNN extracts deep features, and TGNN captures temporal dependencies with error feedback control. A Monte Carlo simulation based on the Central Limit Theorem is introduced to evaluate predictive uncertainty and interval coverage. Experimental results from three wind farms in China demonstrate that the proposed model significantly outperforms baseline hybrid models. Specifically, it achieves a coverage probability (CP) of 0.887, an average interval width (AIW) of 200.870, and a normalized mean square error (NMSE) of 0.057. Compared with the VMD-CNN-LSTM model, the proposed VMD-CNN-TGNN reduces root mean error (RMSE) by approximately 50%, indicating its superior accuracy, robustness, and practical value in wind power forecasting.}, archive = {J_EAAI}, author = {Zhiyong Guo and Qiaoli Han and Fangzheng Wei and Wenkai Qi}, doi = {10.1016/j.engappai.2025.112082}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112082}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Wind power prediction based on hybrid deep learning and monte carlo simulation}, volume = {161}, year = {2025}, }
• (2025). A parallel memetic algorithm for qubit mapping on noisy intermediate-scale quantum machines. EAAI, 161, 112081. (www)

  @article{ , abstract = {Quantum computer technologies are advancing rapidly, but current devices are still limited to the realm of Noisy Intermediate Scale Quantum (NISQ) systems. Due to the limited set of gates and scarce physical qubit connectivity, most quantum circuit-based programs need to be transpiled to be executed on such hardware. This transpilation, which consists in converting the quantum circuit into a hardware-compliant one includes the critical qubit mapping step, where logical qubits are mapped onto physical qubits to meet the hardware connectivity constraints. In this paper, we investigate the application of Artificial Intelligence (AI) to the qubit mapping problem. More exactly, we propose a Parallel Memetic Algorithm for Qubit Mapping (PMA-QM), designed to tackle this challenge efficiently. Using a fine-tuned parallel model to accelerate this inherently computationally expensive hybrid approach, our algorithm takes problem-specific knowledge into account to optimize circuit depth, reducing execution time, and minimizing error rates consequently. PMA-QM has been experimented using various medium-to-large scale circuit benchmarks. PMA-QM outperforms the widely used SWAP-based BidiREctional (SABRE) algorithm, delivering consistently better solutions.}, archive = {J_EAAI}, author = {Jérôme Rouzé and Nouredine Melab and Jan Gmys and Daniel Tuyttens}, doi = {10.1016/j.engappai.2025.112081}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112081}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A parallel memetic algorithm for qubit mapping on noisy intermediate-scale quantum machines}, volume = {161}, year = {2025}, }
• (2025). Enhancing non-orthogonal multiple access systems: A reconfigurable machine learning classification approach. EAAI, 161, 112080. (www)

  @article{ , abstract = {Non-Orthogonal Multiple Access (NOMA) is being considered as a Multiple Access (MA) technique for the next-generation systems, due to factors such as a high per-user spectral resource allocation, grant-free transmission, support for millimeter Waves (mmWaves) and massive-MIMO (mMIMO). A practical limitation of the NOMA systems is imperfect Successive Interference Cancellation (SIC); and the involved decoding delay. This work develops a data-driven Machine Learning (ML) model providing the functionality of a SIC receiver. While most of the Deep Learning (DL) algorithms have high complexity and training times, the given ML approach utilizes the received symbol as the only primary predictor. The other two predictors are derived from the primary predictor, and utilized for the Power-Domain (PD)-NOMA symbol-decoding process. The model utilizes a developed low-complexity NOMA-ML-based Decoder (MLbD) dataset for the same. The extensive test simulations confirm the reconfigurable ML-based receiver to be at par with the existing Maximum Likelihood (MLH) decoder in terms of decoding accuracy. Still, the former supports its integration into the next-generation systems due to its reconfigurable nature and removes the drawbacks related to SIC in the NOMA systems.}, archive = {J_EAAI}, author = {Saurabh Srivastava and Rampravesh Kumar and Prajna Parimita Dash}, doi = {10.1016/j.engappai.2025.112080}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112080}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing non-orthogonal multiple access systems: A reconfigurable machine learning classification approach}, volume = {161}, year = {2025}, }
• (2025). Collaborative energy-saving path planning of unmanned surface vehicle cluster based on multi-head attention mechanism and multi-agent deep reinforcement learning. EAAI, 161, 112078. (www)

  @article{ , abstract = {This paper proposes a Multi-Agent Deep Reinforcement Learning (MADRL) algorithm to address the overlooked issue of energy consumption optimization in existing cluster path planning methods, enabling collaborative energy-efficient path planning for Unmanned Surface Vehicle (USV) clusters. Meanwhile, it provides an alternative MADRL approach to address practical engineering challenges. We not only consider the problems of avoiding danger, reaching the target point and avoiding path conflicts between USV, but also further consider the energy consumption and speed planning of USV cluster. Specifically, firstly, we establish the USV motion model and the USV cluster conflict model, and propose a new energy consumption model, which considers the relationship among speed, marine environment and propulsion load. Secondly, we propose a multi-head local state relevance capture mechanism-multi-agent proximal policy optimization (MLSRC-MAPPO) algorithm. This algorithm can use multi-head attention mechanism (MHA) to capture the potential dependencies between USV, thus enhancing the convergence performance of multi-agent. Finally, in order to reduce the training difficulty, we propose a multi-dimensional action space method for action networks. The experimental results demonstrate that the proposed multi-dimensional action space method has achieved significant success in the lightweighting of the action network: the number of network parameters was reduced by 89.10%, and the computational complexity was decreased by 88.94%, significantly enhancing training efficiency. Meanwhile, the MLSRC-MAPPO algorithm, by incorporating a multi-head attention mechanism, greatly improved the convergence performance of the multi-agent system. In test scenarios with both identical and different starting points, the method reduced energy consumption by 26.24% and 38.47% respectively, fully validating its effectiveness and superiority. Furthermore, comparative experiments with existing cluster energy-saving path planning methods show that the proposed method exhibits clear advantages in terms of energy consumption and path planning efficiency, further verifying its superiority. The corresponding code for this paper is as follows: https://github.com/xhpxiaohaipeng/Multi_USV_Trajectory_energy_plan}, archive = {J_EAAI}, author = {Haipeng Xiao and Lijun Fu and Chengya Shang and Yunfeng Lin and Longfei Yue and Yaxiang Fan}, doi = {10.1016/j.engappai.2025.112078}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112078}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Collaborative energy-saving path planning of unmanned surface vehicle cluster based on multi-head attention mechanism and multi-agent deep reinforcement learning}, volume = {161}, year = {2025}, }
• (2025). Identification of surface subsidence risk in deep foundation pits using a mamba fusion model. EAAI, 161, 112077. (www)

  @article{ , abstract = {This study proposes a novel artificial intelligence–driven neural network model, CNN-Mamba-LSTM-SA, for predicting surface settlement induced by deep foundation pit excavation. The model integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), a Self-Attention (SA) mechanism, and the Mamba architecture to capture both spatial and long-range temporal dependencies in multi-source monitoring data. Bayesian optimization is employed for hyperparameter tuning, and Variational Mode Decomposition (VMD) is used for data denoising, resulting in improved prediction accuracy. To enhance model interpretability, Shapley Additive Explanations (SHAP) are applied to identify key deformation drivers, revealing groundwater level and building settlement as the most influential factors. Model performance is validated using monitoring data from the Nanjing Gemini excavation project, where it achieves superior results compared to conventional models. The CNN-Mamba-LSTM-SA model reduces Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) by up to 69.77 %, 62.45 %, and 79.45 %, respectively. Further analysis through ablation experiments confirms the contribution of each module. Interestingly, the combined removal of CNN and SA results in greater performance degradation than the sum of their individual effects. Finally, a risk warning framework is developed to enable the dynamic transformation of predicted and observed settlement values into actionable risk levels. This integrated artificial intelligence approach offers a robust and interpretable tool for managing excavation-induced risks in urban geotechnical engineering.}, archive = {J_EAAI}, author = {Chenhe Ge and Pengfei Li and Mingju Zhang and Meng Yang}, doi = {10.1016/j.engappai.2025.112077}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112077}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identification of surface subsidence risk in deep foundation pits using a mamba fusion model}, volume = {161}, year = {2025}, }
• (2025). Risk management in maintenance processes: A spherical fuzzy-based failure mode and effect analysis approach in the glass processing industry. EAAI, 161, 112076. (www)

  @article{ , abstract = {Quality management systems are essential tools that aim to increase the competitiveness of businesses and ensure customer satisfaction by creating reliable quality control processes. In this study, the importance of the "risk-based thinking" approach of the International Organization for Standardization (ISO) 9001:2015 standard within the scope of quality management systems was emphasized, and a risk assessment was made for the maintenance process of a business in the glass processing sector. In order to eliminate the deficiencies of the failure mode and effect analysis (FMEA) method, the integration of spherical fuzzy sets (SFSs) and the Technique of Order Preference Similarity to the Ideal Solution (TOPSIS) method was used. Within the scope of the study, 17 different risks were determined by five experts from the maintenance, repair, and quality assurance departments. The experts evaluated these identified risks according to occurrence, severity, and detection factors. Then, these risks were ranked using the Spherical Fuzzy TOPSIS (SF-TOPSIS) method. Finally, different scenarios were created, and their results were discussed to provide a more comprehensive and sensitive risk management approach. This study focuses on maintenance processes in the glass processing industry as a case study. However, the risks related to the maintenance process defined in the study (e.g., machine failures, maintenance inefficiencies, spare parts shortages) are common in the manufacturing sector. Therefore, it can be applied to the sector where the application was carried out and to many different sectors and enterprises.This methodology also serves as a guide for businesses that want to manage process risks within the scope of ISO 9001:2015.}, archive = {J_EAAI}, author = {Buse Duygu Dağıdır and Barış Özkan}, doi = {10.1016/j.engappai.2025.112076}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112076}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Risk management in maintenance processes: A spherical fuzzy-based failure mode and effect analysis approach in the glass processing industry}, volume = {161}, year = {2025}, }
• (2025). Few-shot augmentation based on variational auto-generative adversarial network with moving losses: Application to the variable stiffness prediction in composites. EAAI, 161, 112075. (www)

  @article{ , abstract = {The dataset of carbon fiber reinforced plastics (CFRP) materials presents characteristics of high-dimensional, low-rank, and sparse, which pose difficulties in the combination of mechanical modeling. In this paper, a Variational Auto-Generative Adversarial Network (VAGAN) with moving losses is proposed as a data augmentation method, which extends the size of the CFRP dataset covering components, processes, elasticity, and strengths factors, and increases the information conveyed in the surrogate modeling. A compression strength prediction model for CFRP laminates was constructed by combining the component, process, and mechanical tensor with a neural network optimized by the search algorithm. Combined with the data augmentation strategy, not only was the amount of data expanded, but the prediction accuracy was also significantly improved. The allowable value of compression strength is analyzed and calculated by the predicted values, which brings direct benefits in simplifying the test.}, archive = {J_EAAI}, author = {Zhicen Song and Yunwen Feng and Cheng Lu and Jiaqi Liu}, doi = {10.1016/j.engappai.2025.112075}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112075}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Few-shot augmentation based on variational auto-generative adversarial network with moving losses: Application to the variable stiffness prediction in composites}, volume = {161}, year = {2025}, }
• (2025). Multivariate time series forecasting based on temporal decomposition and graph neural network. EAAI, 161, 112074. (www)

  @article{ , abstract = {It is quite challenging to forecast the Multivariate Time Series (MTS) accurately due to the high dimensionality of MTS and the entangled correlation between variables. Recently, graph-based networks have been demonstrated to be an effective model to handle the complex correlations between MTS. However, all existing graph-based methods construct the graph model of the MTS using only the shallow correlations from the raw MTS data, ignoring the deep-rooted correlations hidden in the features. In this paper, we propose for the first time to construct a comprehensive graph model of MTS that incorporates both shallow correlations from raw data and hidden correlations from decomposed temporal properties. Then, we propose a novel graph-based MTS forecasting framework, which optimizes the graph structure jointly with the model parameters. By doing so, the graph structure can adaptively model the correlations of MTS at a deep level, while the joint optimization can make the constructed graph compatible with the forecasting tasks of MTS, contributing to a globally optimal solution. Finally, we conduct extensive experiments on seven real-world datasets, the results demonstrate the superiority of our method on MTS forecasting over the state-of-the-art baselines. The source codes of the experiments with datasets are available at https://github.com/ironweng/MF-TDGNN .}, archive = {J_EAAI}, author = {Yan Qiao and Pei Zhao and Junjie Wang and Rongyao Hu and Minyue Li and Xinyu Yuan and Meng Li and Zhenchun Wei and Cuiying Feng}, doi = {10.1016/j.engappai.2025.112074}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112074}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multivariate time series forecasting based on temporal decomposition and graph neural network}, volume = {161}, year = {2025}, }
• (2025). A unified framework for lung disease screening using chest X-ray images. EAAI, 161, 112073. (www)

  @article{ , abstract = {Chest X-ray (CXR) imaging is a crucial diagnostic tool for identifying respiratory diseases. The existing methods face challenges such as misclassification caused by overlapping radiologic patterns and the shortage of trained radiologists. However, previous studies have utilized deep learning (DL) models to overcome these limitations, but they struggle with imbalanced datasets, and noise sensitivity. In addition, most of the studies are based on the binary classification of CXR images. Thus, the proposed framework integrates segmentation and multi-class classification for improved generalization and robustness. In this work, artificial intelligence techniques like encoder–decoder and ensemble learning are utilized for segmentation and classification of CXR images that offers reliable, and efficient solution. The output of the presented framework classifies the CXR images into four different classes, along with their corresponding class score. The proposed classification model is ensemble of three pre-trained DL architectures. The last five layers of these base models are fine tuned to match the classification process and combined through a multi-layer perceptron classifier for improved accuracy. The proposed model achieves overall accuracy of 88.98%, and area under curve value of 0.9753, outperforming several state-of-the-art models. It is a lightweight model and demonstrates significant robustness under noisy conditions compared to other models. Moreover, the proposed segmentation and ensemble models are trained and tested on different datasets to obtain greater robustness. The data augmentation is also employed to address class imbalance nature of dataset and enhance the generalization capability. Further, statistical analysis is performed to present the comprehensive comparison among the models.}, archive = {J_EAAI}, author = {Bhavana Singh and Pushpendra Kumar}, doi = {10.1016/j.engappai.2025.112073}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112073}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A unified framework for lung disease screening using chest X-ray images}, volume = {161}, year = {2025}, }
• (2025). A hybrid bayesian model updating and non-dominated sorting genetic algorithm framework for intelligent mix design of steel fiber reinforced concrete. EAAI, 161, 112071. (www)

  @article{ , abstract = {Steel fiber reinforced concrete (SFRC) improves the strength and toughness of conventional concrete, but the high cost and carbon footprint of fibers challenge the balance among performance, cost and sustainability. To address this, an intelligent mix design framework is proposed to optimize compressive and splitting tensile strengths, cost and emissions. Based on 671 experimental records, posterior models were built using Markov Chain Monte Carlo sampling and Bayesian model updating, enabling accurate strength predictions. Compared to traditional regression methods, R 2 scores improved by 15.7 % and 12.4 %, confirming its predictive advantage. Cost-wise, materials dominate, while emissions mainly arise from production, transport and mixing. A non-dominated sorting genetic algorithm identified optimal designs under given constraints. Results show that reducing water-to-cement and aggregate-to-cement ratios, and increasing sand ratio and fiber reinforcement index, enhances SFRC strength. Larger coarse aggregates reduce compressive strength but have limited effect on tensile strength. Optimization suggests potential cost and emission reductions of up to 60 %. Moreover, for compression-prone components, fiber use is inefficient due to high cost and emissions, whereas for crack-resistant or strength-balanced elements, fiber inclusion offers a more sustainable alternative to merely lowering the water-to-cement ratio. The proposed framework enables tailored SFRC mix designs, guiding the efficient use of steel fibers.}, archive = {J_EAAI}, author = {Yong Yu and Jie Su and Bo Wu}, doi = {10.1016/j.engappai.2025.112071}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112071}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid bayesian model updating and non-dominated sorting genetic algorithm framework for intelligent mix design of steel fiber reinforced concrete}, volume = {161}, year = {2025}, }
• (2025). Three-dimensional human pose estimation based on multi-scale spatial–temporal transformer. EAAI, 161, 112068. (www)

  @article{ , abstract = {Recently, transformer-based methods have become dominant in the domain of three-dimensional (3D) human pose estimation, yet the U-net model based on convolutional neural networks (CNN) struggles to model long temporal sequences, and sequence-to-frame (Seq2frame) and sequence-to-sequence (Seq2seq) approaches often fail to preserve dependencies at the start and end of sequences. To address these challenges, this paper proposes a Multi-Scale Spatial–Temporal Transformer network (MSST). This network utilizes Sequence Padding Module (SPM) to extract edge features of the first and last frames, and employs Spatial–Temporal Transformer (STT) to model the spatial–temporal correlations of keypoints. Additionally, we design a Multi-Scale Module (MSM) that analyzes the human skeletal topology to extract multi-scale features of keypoints, local information, and global information, and fuse semantic information at different scales. Finally, we utilize regression heads to project the processed keypoint feature information into 3D space. We conduct quantitative evaluations on two benchmark datasets using four evaluation metrics and design multiple sets of comparative experiments to validate the effectiveness of the proposed modules. Experimental results demonstrate that the proposed network achieves excellent performance.}, archive = {J_EAAI}, author = {Xiaogang Song and Yongxin Cui and Jichen Chen and Xinhong Hei}, doi = {10.1016/j.engappai.2025.112068}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112068}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-dimensional human pose estimation based on multi-scale spatial–temporal transformer}, volume = {161}, year = {2025}, }
• (2025). SharpEdge: High-quality data-driven monocular depth estimation for enhanced boundary precision. EAAI, 161, 112067. (www)

  @article{ , abstract = {While existing monocular depth estimation methods have achieved commendable performance, they often fall short in accurately distinguishing object boundaries. This deficiency largely stems from the inherent noise in dataset acquisition, such as unclear edges and missing depth information. To address these challenges, this paper introduces a novel, high-quality, data-driven monocular depth estimation method tailored for autonomous driving. The approach significantly enhances depth predictions with clearer object boundaries and reduced noise, making it well-suited for real-time, safety-critical applications. Central to our approach is the Self-Adaptive Consistency Filtering mechanism, which dynamically selects high-quality training samples, ensuring that the model learns from the most reliable data and reducing the impact of noise. Additionally, we introduce a Dual-Prior Learning strategy that combines geometric and semantic edge priors. Unlike traditional methods that rely solely on raw depth maps, our approach enhances boundary detection by providing detailed guidance on object contours. This leads to more accurate depth estimation, especially in complex regions where other methods struggle. Empirical evaluations on popular benchmark datasets show that our approach leads to performance improvements of 1.2% on the autonomous driving dataset KITTI and 1.8% on the indoor scene dataset NYU. Compared with recent state-of-the-art methods such as DPT and NewCRFs, our approach achieves superior performance, particularly in recovering fine object boundaries and maintaining spatial consistency across diverse scenes. These results highlight the strong generalization ability of our method, demonstrating that it can enhance depth estimation quality across diverse environments — improving edge precision and spatial coherence, which are critical for autonomous vehicles navigating both complex and dynamic scenarios.}, archive = {J_EAAI}, author = {Mengke Song and Luming Li and Xu Yu and Chenglizhao Chen and Shanchen Pang}, doi = {10.1016/j.engappai.2025.112067}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112067}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SharpEdge: High-quality data-driven monocular depth estimation for enhanced boundary precision}, volume = {161}, year = {2025}, }
• (2025). A systematic review on vision-based gaze estimation: Advance in computer vision and deep learning. EAAI, 161, 112066. (www)

  @article{ , abstract = {Eye gaze estimation is all about figuring out where a person is looking by analyzing eye movements. This is usually done using cameras or eye-tracking devices, a key technique in computer vision. Gaze estimation has a wide range of uses—from helping people interact with computers more naturally (HCI), monitoring drivers, enhancing virtual reality experiences, and reading emotional cues in affective computing. Deep learning has improved how accurately and reliably these systems work in recent years. Models like Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Generative Adversarial Networks (GANs) have made it easier to detect and interpret eye movements in complex settings. This review closely examines how gaze estimation works, the datasets commonly used, and both traditional and deep learning-based methods. It also points out what's working well, where current methods fall short, and what researchers still need to solve. The goal is to offer helpful insights for anyone working to improve or apply gaze tracking technologies in real-world scenarios.}, archive = {J_EAAI}, author = {Sapna Singh Kshatri and Deepak Singh}, doi = {10.1016/j.engappai.2025.112066}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112066}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A systematic review on vision-based gaze estimation: Advance in computer vision and deep learning}, volume = {161}, year = {2025}, }
• (2025). A novel framework for improving railway driver performance based on emotional intelligence and job-driven factors: An artificial neural network method. EAAI, 161, 112065. (www)

  @article{ , abstract = {Efficient transportation systems critically rely on human factors to ensure safety, reliability, and service quality. This study introduces a novel framework for evaluating and improving the performance of the Tehran Urban and Suburban Railway Company (TRS) drivers by focusing on the role of emotional intelligence (EI) and job-driven (JD) factors. The core of the proposed method is a two-stage hybrid approach: first, an Artificial Neural Network (ANN) is used to model driver performance using EI and JD factors as inputs and outputs, respectively; then, Data Envelopment Analysis (DEA) validates the results and benchmarks performance levels. Data were collected from 146 TRS drivers through a standardized questionnaire. Sensitivity analysis and statistical tests evaluated the impact of EI and JD factors on driver performance. Additionally, a SWOT (Strengths, Weaknesses, Opportunities, Threats) analysis translated the findings into practical recommendations for system improvement. The findings reveal high job satisfaction and low job stress levels among TRS drivers, highlighting the significant contribution of EI and JD factors. By combining data-driven prediction, performance benchmarking, and strategic analysis, this study introduces a novel, human-centered framework for driver Performance Evaluation (PE)—one that fills methodological gaps in the literature and supports enhancing safety, efficiency, and passenger satisfaction in railway systems.}, archive = {J_EAAI}, author = {Narges Hajloo and Behnaz Salimi and Mahdi Hamid and Masoud Rabbani}, doi = {10.1016/j.engappai.2025.112065}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112065}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel framework for improving railway driver performance based on emotional intelligence and job-driven factors: An artificial neural network method}, volume = {161}, year = {2025}, }
• (2025). An efficient fusion-based deep learning framework for land use and land cover image clustering. EAAI, 161, 112061. (www)

  @article{ , abstract = {Land use and land cover (LULC) analysis is vital for understanding spatial dynamics and informing environmental management, urban planning, and sustainable development. Traditional approaches, such as manual surveys and conventional image clustering methods, often face limitations in scalability and adaptability. This paper presents a novel deep learning framework that combines the Vision Transformer (ViT) and Variational Autoencoder (VAE) to extract complementary feature representations for LULC image clustering. The ViT tokenizes image patches to capture high-level semantic features, while the VAE models latent structures to integrate contextual and structural information. To further improve clustering performance, the framework incorporates Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction followed by k -means++ clustering, enabling a scalable and robust solution for diverse datasets. Experiments on multiple datasets, including the Urban Atlas LULC 2018 dataset and recent LULC maps of Japan and Vietnam, demonstrate the framework’s superior ability to capture complex LULC patterns compared to traditional methods. The datasets and source code will be made publicly available at https://github.com/ClarkDinh/LULCMiner . This framework has broad applications across geospatial and remote sensing engineering, civil and environmental engineering, agricultural planning, transportation, and urban development.}, archive = {J_EAAI}, author = {Tai Dinh and Dat Tran and Zdena Dobešová and Huynh Van Hong and Daniil Lisik and Rameesh Khan}, doi = {10.1016/j.engappai.2025.112061}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112061}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient fusion-based deep learning framework for land use and land cover image clustering}, volume = {161}, year = {2025}, }
• (2025). Optimizing multi-objectives urban siting of hydrogen refueling dispenser using fuzzy NSGA-II: A case study in fez, morocco. EAAI, 161, 112060. (www)

  @article{ , abstract = {To support the growing need for sustainable urban transport, it is essential to reduce the environmental impact of vehicles; this study focuses on optimizing hydromobility through the strategic integration of hydrogen dispensers (HDs) into existing locations, rather than deploying standalone hydrogen refueling stations (HRSs). This approach minimizes costs while enhancing accessibility, addressing the current limitations in hydrogen vehicle (HV) adoption. A novel method combining the non-dominated sorting genetic algorithm II (NSGA-II) with fuzzy logic and Pareto front analysis is proposed to identify optimal HD locations using a real case study of Fez city in Morocco. The results provide actionable strategies for implementing HDs in a way that balances efficiency, accessibility, and budget constraints. The study demonstrates how intelligent planning can support the transition to cleaner energy solutions in urban mobility.}, archive = {J_EAAI}, author = {Soukayna Abibou and Dounia El Bourakadi and Ali Yahyaouy and Hamid Gualous}, doi = {10.1016/j.engappai.2025.112060}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112060}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing multi-objectives urban siting of hydrogen refueling dispenser using fuzzy NSGA-II: A case study in fez, morocco}, volume = {161}, year = {2025}, }
• (2025). Synthetic data enhancement using diffusion models for improved miscanthus identification and bioenergy extraction. EAAI, 161, 112059. (www)

  @article{ , abstract = {Miscanthus, a high-yielding perennial grass pivotal for bioenergy production, requires precise species identification to optimize bioenergy extraction. However, limited annotated spectral datasets hinder robust classification model development. To address this challenge, the paper proposes a novel diffusion probabilistic model tailored for near-infrared spectral synthesis. Unlike conventional generative approaches, the proposed model integrates a bidirectional gated recurrent unit-based temporal encoder and one dimension convolutional neural networks within a diffusion framework, augmented by a spectral attention module to prioritize critical absorption bands. This architecture uniquely addresses the sequential dependencies and subtle biochemical variations inherent in near-infrared spectral, enabling high-fidelity generation of diverse synthetic data. The diffusion process is optimized through a hybrid loss function combining variational lower bound training with mean squared error for pixel-level fidelity and maximum mean discrepancy for distributional alignment. Evaluated on 517 near-infrared spectral samples across three Miscanthus species, the proposed model outperforms traditional variational autoencoders , generative adversarial networks, and standard diffusion models in terms of sample authenticity and diversity. Incorporating synthetic data enhanced the accuracy, precision, and recall of downstream classifiers by 10%–15%, with the convolutional neural networks attaining 87% accuracy using hybrid real-synthetic training data. Remarkably, even with 50% synthetic data substitution, classification accuracy remained robust at 75%, demonstrating the model’s efficacy in mitigating data scarcity and advancing precision agriculture for bioenergy optimization.}, archive = {J_EAAI}, author = {Xinyue Wang and Xiangdong Chen and Jun Jiang and Ronggao Gong and Biao Wang}, doi = {10.1016/j.engappai.2025.112059}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112059}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Synthetic data enhancement using diffusion models for improved miscanthus identification and bioenergy extraction}, volume = {161}, year = {2025}, }
• (2025). Multi-order feature learning with global–local attention for real-time X-ray security inspection. EAAI, 161, 112058. (www)

  @article{ , abstract = {X-ray security inspection in high-throughput environments such as airports faces fundamental challenges stemming from the intrinsic complexity of baggage scanning. These challenges include feature entanglement from overlapping objects, visual distortions due to fixed imaging angles, inconsistent pseudo-color representations caused by varying X-ray absorption, and the difficulty of detecting prohibited items across multiple scales. To tackle these challenges, we introduce Multi-Order Gated Network (MoGNet), a lightweight transformer-based architecture that integrates three core innovations. First, a Multi-order Gated Aggregation Block designed for efficient multi-order feature extraction. Second, a Global–Local Self-Attention mechanism that enhances differentiation between foreground and background elements. Finally, a DynamicFusion module for adaptive integration of multi-scale features. Comprehensive evaluations on five challenging datasets establish our proposed framework, MoGNet, as the new state-of-the-art (SOTA). The model demonstrates superior performance, achieving mean Average Precision (mAP) scores at a 50% Intersection over Union (IoU) threshold (mAP 50 ) of 75.4%, 91.9%, 91.2%, 96.6%, and 80.0%, respectively. This high accuracy is maintained while operating at an efficient 64.1 Frames Per Second (FPS). These comprehensive experimental results demonstrate its remarkable capability to optimize the balance between computational efficiency and detection accuracy, establishing it as a preferable solution for real-time contraband detection in practical security screening.}, archive = {J_EAAI}, author = {Ling Guo and Yangbin Xu and Shouhong Chen}, doi = {10.1016/j.engappai.2025.112058}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112058}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-order feature learning with global–local attention for real-time X-ray security inspection}, volume = {161}, year = {2025}, }
• (2025). A stress-dependent strength neural network model for predicting the true triaxial strength of rocks. EAAI, 161, 112057. (www)

  @article{ , abstract = {Understanding the true triaxial strength of rocks is essential for safe underground engineering, yet existing empirical and data-driven models often fail to capture the nonlinear effects of the intermediate principal stress σ 2 . This study proposes a stress-dependent strength neural network (SDSNN) that integrates physically informed constraints, including monotonicity and boundary conditions, as well as an exponential adaptive weighting strategy to balance data and constraint losses. Cohesion and internal friction angle are used as input features, replacing conventional reliance on uniaxial strength. Compared with a purely data-driven neural network (DDNN) and a constraint-aware variant without adaptive weighting (SDSNN#0), SDSNN achieves significantly better predictive accuracy and robustness across test sets, representative rock types, and five-fold cross-validation. In particular, it maintains consistent strength trends and improved stability across six representative rocks with diverse mechanical properties. Importantly, the model maintains high performance even when trained only on low-to-mid σ 2 data and tested on high σ 2 conditions—demonstrating strong generalization under extrapolation. This capability has received relatively limited attention in data-driven models and is particularly valuable in practical scenarios where high- σ 2 test data are limited or difficult to obtain. Furthermore, ablation analysis demonstrates that removing physical constraints leads to a notable decrease in model accuracy, underscoring the importance of incorporating strength variation characteristics into the model. SDSNN shows particular advantage under boundary stress conditions and when facing noisy or sparse datasets, indicating its potential to serve as a robust and interpretable tool for true triaxial strength prediction in geotechnical applications.}, archive = {J_EAAI}, author = {Tianzhi Yao and Yunpeng Gao and Jianhai Zhang and Ru Zhang and Li Qian and Qijun Hu and Xianliang Wang and Feng Jiang}, doi = {10.1016/j.engappai.2025.112057}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112057}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A stress-dependent strength neural network model for predicting the true triaxial strength of rocks}, volume = {161}, year = {2025}, }
• (2025). Multi-source contrastive cluster center method for cross-domain bearing fault identification. EAAI, 161, 112056. (www)

  @article{ , abstract = {Because of the complexity and time-varying attributes of the exterior surroundings, rolling bearings commonly operate under variable working conditions at any time. Therefore, there is a multi-source domain adaptation problem that involves multiple source domains and a target domain. In this circumstance, identifying faults directly in a multi-source domain through a model built in a single source domain will also lead to limited model generalization ability, i.e., due to the incompleteness of the training sample size caused by the singularity of the domain quantity limited access to fault knowledge in a single source domain, the established model may be prone to over-fitting, thus whose generalization ability has been decreased under complex and variable working conditions to a certain extent. Hence, this paper has proposed a Multi-source Contrastive Cluster Center (MS3C) Method for addressing the aforementioned issues. Experimental findings on two datasets have suggested that MS3C has not only considered the domain shifts of the same classes between different source domains and the target domain but also adaptively aligned the feature distributions of the same classses in different source domains, therefore, MS3C has a higher identification rate, a better clustering and classification performance and a superior convergence.}, archive = {J_EAAI}, author = {Pengfei Chen and Lizhen Wu and Rongzhen Zhao and Kongyuan Wei and Yuqiao Zheng and Linfeng Deng and Yongfei Zhang and Mingkuan Shi and Zhuo Chen}, doi = {10.1016/j.engappai.2025.112056}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112056}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-source contrastive cluster center method for cross-domain bearing fault identification}, volume = {161}, year = {2025}, }
• (2025). An enhanced encoder–decoder network for temporomandibular joint segmentation in magnetic resonance images. EAAI, 161, 112054. (www)

  @article{ , abstract = {Accurate segmentation of the condyle, articular disc, and articular eminence from magnetic resonance imaging (MRI) is essential for the diagnosis and treatment planning of temporomandibular joint (TMJ) disorders. However, current deep learning methods for TMJ segmentation face significant challenges: (1) anatomical structures of the TMJ exhibit complex inter-patient variability; (2) structural boundaries are often ambiguous; and (3) large inter-slice spacing in MRI makes it difficult to fully utilize spatial context, limiting the effectiveness of both three-dimensional and two-dimensional approaches. To address these challenges, we propose an edge-enhanced TransUNet-based encoder–decoder framework that effectively integrates local edge features and global contextual information to obtain accurate and robust TMJ segmentation. Specifically, a cross-slice attention transformer (CAT) is introduced to capture inter-slice dependencies, addressing the challenge of large slice spacing and enhancing the utilization of contextual information across slices. Moreover, a feature enhancement module (FEM) is designed to explicitly fuse edge information, facilitating accurate localization where boundaries are blurred or indistinct. In addition, an attention gate (AG) mechanism adaptively highlights salient anatomical structures and suppresses irrelevant background, improving the model’s overall focus and resistance to noise. Evaluated on the private TMJ MRI dataset, our method achieves Dice similarity coefficients (DSC) of 0.922, 0.834, and 0.837 for the condyle, articular disc, and articular eminence, respectively, outperforming baseline and comparison methods. Further validation on a public knee MRI dataset demonstrates the good generalizability of the proposed method. These results demonstrate robust and precise TMJ segmentation, supporting reliable automated analysis in clinical settings.}, archive = {J_EAAI}, author = {Yilin Hu and Yunan Zhang and Wei Tang and Jixiang Guo}, doi = {10.1016/j.engappai.2025.112054}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112054}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An enhanced encoder–decoder network for temporomandibular joint segmentation in magnetic resonance images}, volume = {161}, year = {2025}, }
• (2025). Three-way decision method integrating probabilistic linguistic term sets and prospect theory in large-scale scenarios. EAAI, 161, 112053. (www)

  @article{ , abstract = {Probability linguistic term sets (PLTSs), owing to their ability to effectively represent the uncertainty and complexity of decision information, have been widely adopted in large-scale multiattribute decision-making (MADM) scenarios. However, existing decision-making methods in the PLTS environment typically suffer from information loss in similarity computation, subjectivity in weight determination, and inadequate consideration of decision-makers’ behavioral psychology, which limits their applicability in real-world complex situations. To address these issues, this paper proposes an innovative hybrid three-way decision (TWD) method that integrates prospect theory with multiobjective optimization on the basis of the ratio analysis (MOORA) approach. Compared with existing studies, the main contributions of this paper include (1) an objective weight determination mechanism based on the information entropy of PLTSs, which improves the scientific validity and computational efficiency of weight allocation compared with subjective or complex traditional methods (2) a novel direct similarity degree based on the probability linguistic Jensen–Shannon (PLJS) divergence, which effectively overcomes the information loss caused by indirect distance measures and achieves accurate similarity characterization between PLTSs (3) a new conditional probability determination strategy utilizing θ -level similarity classes that enhances the dynamic adaptability and accuracy of conditional probability estimation and (4) the integration of prospect theory and the MOORA method within the TWD framework, enabling hierarchical classification of alternatives while fully reflecting decision-makers’ psychological and behavioral characteristics, thereby facilitating the classification and ranking of alternatives. Furthermore, through comparative analysis with large-scale real-world air quality assessment data and mainstream PLTS-based decision-making methods, the effectiveness and rationality of the proposed method are comprehensively verified, demonstrating its superiority, robustness, and practical application value.}, archive = {J_EAAI}, author = {Zhaxi Pahua and Haidong Zhang and Yizhu Cairang and Yanping He}, doi = {10.1016/j.engappai.2025.112053}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112053}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-way decision method integrating probabilistic linguistic term sets and prospect theory in large-scale scenarios}, volume = {161}, year = {2025}, }
• (2025). Adaptive feature-level fusion of manifold and deep learning for robust multi-view face recognition. EAAI, 161, 112052. (www)

  @article{ , abstract = {Robust face recognition across varying poses, lighting conditions, and viewpoints remains a significant challenge in engineering applications such as surveillance, authentication, and human–computer interaction. This paper presents Dynamic Manifold–Deep Learning Fusion (DM-DLF), an artificial intelligence (AI)-based framework that integrates manifold learning with deep neural networks using an adaptive feature-level weighting strategy. By combining global semantic features and local geometric structures, the proposed AI model enhances identity recognition under real-world multi-view scenarios. Experimental results show that DM-DLF surpasses recent transformer-based and graph-based models, achieving up to 94.00 % classification accuracy. The method also reduces prediction error and improves training efficiency. These findings confirm DM-DLF as a powerful AI solution for multi-view face recognition in engineering systems where labeled data is limited and view diversity is high.}, archive = {J_EAAI}, author = {Faraein Aeini}, doi = {10.1016/j.engappai.2025.112052}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112052}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive feature-level fusion of manifold and deep learning for robust multi-view face recognition}, volume = {161}, year = {2025}, }
• (2025). Enhancing long-term load forecasting with convolutional informer-based hybrid model. EAAI, 161, 112051. (www)

  @article{ , abstract = {Long-term load forecasting (LTLF) is essential for energy management but challenged by the complexity of non-stationary time series. The Informer model struggles to capture localized peak–valley patterns, while Variational Mode Decomposition (VMD) faces issues with feature complexity. This study proposes a hybrid framework integrating VMD, Informer, and a Convolutional Long Short-Term Memory (CNN-LSTM) module for accurate LTLF. VMD decomposes non-stationary load data into multi-scale intrinsic mode functions, refined through spectral and autocorrelation analyses to ensure robust feature extraction. The Informer employs sparse self-attention for efficient long-sequence modeling, with CNN-LSTM enhancing the decoder to capture localized temporal dynamics. Experiments on non-stationary load time series across multiple prediction horizons demonstrate that the proposed framework significantly improves forecasting accuracy and robustness compared to baseline models, including Informer and its derivatives. By excelling in complex load pattern prediction, the framework supports efficient grid scheduling and resource optimization in energy systems.}, archive = {J_EAAI}, author = {Bin Sun and Xudong Chen and Tao Shen and Liyao Ma}, doi = {10.1016/j.engappai.2025.112051}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112051}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing long-term load forecasting with convolutional informer-based hybrid model}, volume = {161}, year = {2025}, }
• (2025). Fault diagnosis of high-voltage three-phase asynchronous motors using residual neural networks and bidirectional gated recurrent units. EAAI, 161, 112049. (www)

  @article{ , abstract = {High-voltage three-phase asynchronous motors are extensively used in industrial production; however, rotor bar breakage faults are characterized by strong concealment and severe consequences, making early and accurate diagnosis difficult with traditional methods. To address this issue, this paper proposes a deep learning model based on a Residual Neural Network (ResNet) and a Bidirectional Gated Recurrent Unit (Bi-GRU) for motor fault diagnosis. The proposed model integrates the spatial feature extraction advantages of ResNet with the temporal modeling capabilities of Bi-GRU, simultaneously capturing spatiotemporal information from voltage and current signals. The model was trained and tested on a large-scale dataset comprising 324,000 data points under nine distinct operating conditions. Experimental results demonstrate that the proposed method achieves over 99.83 % in key metrics such as accuracy, precision, recall, and F1-score, significantly outperforming traditional approaches and other comparative models. Additionally, it maintains stable performance under varying operating conditions, demonstrating strong robustness and generalization capability. Further ablation experiments also validated the effectiveness of each module within the proposed model. This study indicates that the application of deep learning in industrial motor fault diagnosis has promising prospects and practical value.}, archive = {J_EAAI}, author = {Huihui Yang and Yuxin Wu}, doi = {10.1016/j.engappai.2025.112049}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112049}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis of high-voltage three-phase asynchronous motors using residual neural networks and bidirectional gated recurrent units}, volume = {161}, year = {2025}, }
• (2025). Zero-shot image translation via query compensation and style enhancement. EAAI, 161, 112048. (www)

  @article{ , abstract = {In zero-shot image translation methods based on diffusion models, two prevalent challenges are the loss of identity information and insufficient stylization. To address these issues, we propose a zero-shot image translation method. Specifically, we use the edit-friendly noise space for image inversion. By this way, the identity of the input image is largely preserved. Additionally, we introduce an identity compensation mechanism by injecting the source query vectors into the denoising process. Furthermore, to tackle the model’s insufficient stylization ability, we propose cross-attention style modulation (CSM) to transfer style information from the reference to the input image. Finally, to further enhance the style effect of the translated image, we design initial latent self-enhancement (ILS), style supplementation (SS), and style alignment (SA) strategies. Our zero-shot image translation method does not require training samples, optimization, or fine-tuning. We achieve this by manipulating the self-attention features of a pre-trained diffusion model in a manner analogous to cross-attention—by replacing the key and value vectors of the input image with those of the reference image during the denoising process. Extensive in-domain and cross-domain translation experiments demonstrate the effectiveness of our method across a wide range of object categories and show strong robustness to variations in object shape, size, posture and instance between the input and reference image. The proposed method achieves competitive performance in identity preservation, as measured by the mean Intersection over Union (mIoU), and attains the best style transfer in terms of Single Image Fréchet Inception Distance (SIFID). In addition, qualitative comparisons demonstrate that our approach significantly outperforms the baselines.}, archive = {J_EAAI}, author = {Heng Zhang and Yi-Jun Yang and Wei Zeng}, doi = {10.1016/j.engappai.2025.112048}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112048}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Zero-shot image translation via query compensation and style enhancement}, volume = {161}, year = {2025}, }
• (2025). Camouflaged object detection with boundary localization in complex backgrounds. EAAI, 161, 112047. (www)

  @article{ , abstract = {The primary challenge of Camouflaged Object Detection (COD) lies in the high similarity between the target and the complex background, making it difficult for the human eye to distinguish them. Based on the phenomenon that human attention shifts between the target and the background when observing objects, we propose a network model named MENet. This model adopts a three-stage decoupled architecture of “localization-interaction-fusion.” In the localization stage, we utilize an attention mechanism-based backbone network (Pyramid Vision Transformer V2, abbreviated as PVT-V2) to generate multi-level features, which can initially locate the target area. In the interaction stage, we design a Contour-Aware Edge Module (CAEM) and an Area Decoder (AD) to capture the target edges and background information, respectively, thereby achieving precise localization of the target boundary and reducing interference from background noise. Furthermore, we developed a Boundary Guidance Module (BGM) that effectively injects boundary cues and relevant background information separately into the multi-level features, enhancing the model’s ability to detect target edges in complex backgrounds. In the fusion stage, we design two Feature Fusion Modules (FFM and KFFM) to effectively merge multi-level features with precise boundaries and de-noised features, thereby enhancing the prediction performance of camouflaged objects. Extensive experiments on three challenging benchmark datasets demonstrate that our MENet outperforms many existing state-of-the-art methods. Our method leverages artificial intelligence (AI) techniques to improve the accuracy of camouflaged object and pest detection in complex visual environments. Our code is publicly available at: https://github.com/yang19950966666/MENet .}, archive = {J_EAAI}, author = {Guangjian Zhang and Zhengming Yang and Yong Wang and Yuliang Chen and Duoqian Miao}, doi = {10.1016/j.engappai.2025.112047}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112047}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Camouflaged object detection with boundary localization in complex backgrounds}, volume = {161}, year = {2025}, }
• (2025). A grid-based boundary sharpening clustering algorithm. EAAI, 161, 112045. (www)

  @article{ , abstract = {To address the clustering problem for arbitrary shapes, in this paper, we propose a Grid-based Boundary Sharpening clustering algorithm called as “GBSharp”. This method is grounded in morphology and relies on two fundamental morphological operations: dilation and erosion. The main innovations of the proposed algorithm lie in two aspects. Firstly, we further introduce the concepts of inward dilation and bridge erosion based on the basic morphological operations to reduce the impact of the chain effect. Secondly, a unique indexing structure is designed specifically for non-empty cells in high dimensional space. In addition, to tackle the complex conditional judgments encountered in high-dimensional scenarios, we further utilize the inversion method for bridge-erosion operation. Experiments conducted on synthetic datasets and real-world datasets further validate the effectiveness and efficiency of the proposed algorithm.}, archive = {J_EAAI}, author = {Lin Ma and Qijing Yan and Mengxia Lv and Tiefeng Ma and Mingchang Cheng}, doi = {10.1016/j.engappai.2025.112045}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112045}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A grid-based boundary sharpening clustering algorithm}, volume = {161}, year = {2025}, }
• (2025). Advances in physics-informed neural networks for solving complex partial differential equations and their engineering applications: A systematic review. EAAI, 161, 112044. (www)

  @article{ , abstract = {Most physical and engineering problems can be described by partial differential equations (PDEs), which are typically solved using numerical methods such as the finite difference method and the finite element method. However, conventional numerical discretization approaches face significant challenges in terms of computational efficiency and convergence speed when dealing with complex nonlinear PDEs, such as high-dimensional nonlinear PDEs, stiff PDEs, PDEs with complex boundary conditions or irregular geometries, and multi-scale PDEs. Recently, physics-informed neural networks (PINNs) have emerged as a transformative methodology for solving complex PDEs by integrating physical laws intrinsically into deep learning architectures. While PINNs effectively overcome mesh dependency and dimensionality constraints inherent in traditional numerical methods, they still encounter persistent challenges related to training convergence and generalization robustness. This paper aims to present a comprehensive review of the state-of-the-art developments in PINNs for solving complex PDE problems. The core ideas, network architectures, and generic implementation frameworks, along with associated open-source Python libraries, are first introduced in detail. Furthermore, a systematic taxonomy of optimization techniques is provided, covering hyperparameter selection, adaptive sampling strategies, physics-constrained loss formulations, hybrid differentiation approaches, and architectural innovations. Subsequently, various coping strategies and research advancements of PINNs in addressing complex nonlinear PDE problems are thoroughly discussed. Real-world engineering applications are then reviewed across multiple domains, including cosmology and quantum mechanics, materials science and manufacturing, fluid mechanics, energy systems, biological and environmental sciences, and power and information technologies. Finally, this paper discusses the current challenges and limitations of PINNs in solving complex PDEs and outlines potential directions for future research. By addressing the current limitations and pursuing targeted improvements in architectures, training, interpretability and generalization, PINNs can become a powerful tool in engineering and scientific applications.}, archive = {J_EAAI}, author = {Jiangtao Guo and Hao Zhu and Yujie Yang and Chenrui Guo}, doi = {10.1016/j.engappai.2025.112044}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112044}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advances in physics-informed neural networks for solving complex partial differential equations and their engineering applications: A systematic review}, volume = {161}, year = {2025}, }
• (2025). DPMF-net: A dual-path perceptive multi-stage fusion network for skin lesion segmentation. EAAI, 161, 112043. (www)

  @article{ , abstract = {Accurate segmentation of skin lesions in dermoscopic images is crucial for skin cancer detection and treatment. Despite progress in deep learning-based methods, challenges remain due to diverse skin lesion shapes, colors, and blurred boundaries. We propose a novel Dual-path Perceptive Multi-stage Fusion Network (DPMF-Net) for skin lesion segmentation. DPMF-Net integrates multiple feature refinement modules. It aims to gradually optimize lesion representations by leveraging the dual-path framework to perceive high-level contextual information. The Spatial Frequency Dual-path Cascaded Perception Module (SFDCP) synergizes spatial and frequency domains to model long-range dependencies and suppress noise, enhancing perception of low-contrast lesions. Subsequent to the SFDCP, the Spatial Channel Dual-path Parallel Perception Module (SCDPP) employs entropy-driven attention and multi-granularity convolutions in skip connections to dynamically select informative channels and extract lesion details across spatial scales. To verify the efficacy of our proposed DPMF-Net, extensive experimental assessments are carried out across four challenging datasets. The outcomes of these quantitative and qualitative experiments confirm that our approach significantly outperforms current state-of-the-art methods in terms of all evaluation metrics.}, archive = {J_EAAI}, author = {Yuling Huang and Yaoyao Ma and Jing Wang and Chao Xu and Zhiwei Fan and Di Wu}, doi = {10.1016/j.engappai.2025.112043}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112043}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DPMF-net: A dual-path perceptive multi-stage fusion network for skin lesion segmentation}, volume = {161}, year = {2025}, }
• (2025). Gradient surgery: A necessity for robust test-time adaptation for detecting casting defects. EAAI, 161, 112039. (www)

  @article{ , abstract = {Casting defects pose a significant challenge in the manufacturing industry, leading to material waste, production inefficiencies, and compromised product quality. While deep learning models have shown promise in automating defect detection, their effectiveness is often constrained by domain shifts and variability in real-world data distributions. In this work, we propose Bayesian Test-Time Adaptation (BTTA), a novel framework designed to enhance the robustness and adaptability of machine learning models in such dynamic environments. Unlike traditional Test-Time Adaptation (TTA) methods, our approach employs gradient-guided diversification with Stein Variational Gradient Descent (SVGD) to explore diverse optimization paths. Experimental results on benchmark datasets, including CIFAR-10-C , Casting Defects , and GDXray , demonstrate significant performance improvements across key metrics. Notably, the framework achieves an average accuracy improvement of 2-3% under severe corruption levels and excels in cross-domain generalization, highlighting its ability to handle diverse and unseen defect categories. This dynamic adaptability not only addresses the limitations of static models but also offers a practical and cost-effective solution for real-time defect detection in industrial settings. Our study underscores the potential of BTTA to transform quality assurance processes, ensuring reliable performance across varying operational conditions without the need for extensive retraining or large annotated datasets. The codebase for BTTA is available on: https://github.com/afsharshamsi/GradSurgery .}, archive = {J_EAAI}, author = {Afshar Shamsi and Rejisa Becirovic and Hamid Alinejad-Rokny and Arash Mohammadi and Ahmadreza Argha}, doi = {10.1016/j.engappai.2025.112039}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112039}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Gradient surgery: A necessity for robust test-time adaptation for detecting casting defects}, volume = {161}, year = {2025}, }
• (2025). Enhancing real-time detection transformer for small floating targets. EAAI, 161, 112038. (www)

  @article{ , abstract = {Addressing the challenge of detecting small floating target in complex water environments, where it is difficult to balance real-time performance and end-to-end capabilities, this study introduces a specialized model called the Enhancing Real-Time Detection Transformer (ERT-DETR). This model integrates a Dynamic Feature Pyramid Network (DFPNet) with a Micro-Attention Module (MAM) to achieve refined feature extraction and enhanced object recognition mechanisms. Additionally, it incorporates an Inner Intersection over Union (Inner-IoU) auxiliary bounding box loss function, which accelerates model convergence and improves bounding box accuracy. Experiments conducted on the Floating Object in Water Image Dataset (FloW-Img) demonstrate that the ERT-DETR model achieves a 92.6% Average Precision (AP) and 118.84 Frames Per Second (FPS), outperforming the baseline Real-Time Detection Transformer (RT-DETR) by 4.3% and 19.8%, respectively. The ERT-DETR model’s precision and real-time performance in dynamic water surface environments surpass those of the most advanced You Only Look Once (YOLO) and Detection Transformer (DETR) series detectors. This model is significant for enhancing small floating target detection capabilities in complex water environments and can be extended to applications in marine management, environmental monitoring, and vessel tracking.}, archive = {J_EAAI}, author = {Guobing Xie and Xinran Wu and Jiefeng Shi and Yixin Su and Binghua Shi}, doi = {10.1016/j.engappai.2025.112038}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112038}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing real-time detection transformer for small floating targets}, volume = {161}, year = {2025}, }
• (2025). Non-contact weight intelligent estimation based on yak skeleton localization. EAAI, 161, 112036. (www)

  @article{ , abstract = {Weight estimation is a vital method for monitoring the growth and health of yaks, however, traditional techniques-such as relying on herders’ experience or using weighbridge-are labor-intensive, time-consuming and pose safety risks. Currently, many studies have shown that yak body size can be an effective indicator of weight. With the advancement of computer vision, non-contact weight estimation has become increasingly feasible for livestock. Yet, studies focusing on yak weight estimation, particularly in high-altitude plateau regions, remain limited. Herein, we propose a novel weight estimation approach based on deep learning and binocular vision technology to address this gap. The method involves four main steps: (1) yak image acquisition, (2) skeletal key point localization, (3) body size calculation (4) weight estimation using Gaussian process regression. To enhance practicality and mobility, we also developed two edge-intelligent devices: an intelligent inspection vehicle and a handheld detection unit, enabling convenient and non-invasive weight estimation. Our models are trained and tested on a yak dataset collected by our team on the Tibetan Plateau. Experimental results demonstrate the effectiveness of our approach, achieving an Mean Absolute Percentage Error (MAPE) of 0.12 percent, a Mean Absolute Error (MAE) of 25.4 kilograms (kg) and a Coefficient of determination ( R 2 ) value of 0.72. It not only provides a new technical solution for the yak industry but also provides innovative insights for advancing intelligent animal husbandry. The code and data can be accessed at https://github.com/FeiWang-swun/YakWeight .}, archive = {J_EAAI}, author = {Fei Wang and Xinghua Zou and Zhijiang Chen and Qi Tang and Tianshuo Li and Shuiying Wang and Lijun Yang and Dongming Tang}, doi = {10.1016/j.engappai.2025.112036}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112036}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Non-contact weight intelligent estimation based on yak skeleton localization}, volume = {161}, year = {2025}, }
• (2025). Comparison of the performances of artificial intelligence bots using continuous intuitionistic fuzzy evaluation based on distance from average solution method. EAAI, 161, 112033. (www)

  @article{ , abstract = {The rapid evolution of artificial intelligence (AI) has introduced novel opportunities and challenges in various fields. In this study, we present a pioneering approach known as Continuous Intuitionistic Fuzzy (CINFU) Evaluation based on Distance from Average Solution (EDAS), an innovative extension of the EDAS method tailored to Continuous Intuitionistic Fuzzy Sets. This methodology is designed to compare the performance of AI tools. The capabilities of AI bots have been examined through their success rates in various tasks and uncertainty levels in decision-making processes. The study aims to evaluate the effectiveness of different models in decision-making processes by analyzing the performances of AI bots such as Chat Generative Pre-trained Transformer (ChatGPT), Bard, and Claude based on both objective measurements and fuzzy evaluation criteria. The comparison focuses on key performance criteria such as Bots Triggered, User Engagement, Message Click-Through Rate, Chat Handoff, User Retention, Bounce Rate & Dwell Time, Leads Captured, and Customer Satisfaction Score. Ultimately, the validity and robustness of the approach have been tested with sensitivity analysis.}, archive = {J_EAAI}, author = {Nurşah Alkan and Umut Aydın and Akın Menekşe and Cengiz Kahraman}, doi = {10.1016/j.engappai.2025.112033}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112033}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comparison of the performances of artificial intelligence bots using continuous intuitionistic fuzzy evaluation based on distance from average solution method}, volume = {161}, year = {2025}, }
• (2025). A global linear attention incorporated video transformer for robust sintering condition recognition. EAAI, 161, 112032. (www)

  @article{ , abstract = {Robust and accurate sintering condition recognition is a fundamental yet critical issue in the design of image-based intelligent combustion control systems. However, owing to the weak texture and fast changing characteristics of flame videos, capturing the condition indicator using existing gradient-based methods is challenging. To address this issue, we propose a global linear attention incorporated video transformer model for sintering condition recognition. First, to reduce the prediction error and uncertainty, the spatial-temporal features are extracted to describe the dynamic characteristics of the flame video streams based on the video shifted window (Swin) Transformer architecture. Next, to address the problem that the local attention strategy used in the Video Swin Transformer is insufficient for global flame feature extraction, we propose a Video Linear Attention block that obtains the global attention as a supplement. Extensive experiments conducted on a real-world rotary kiln sintering dataset demonstrate the effectiveness of our approach, achieving an overall accuracy of 97.76% and an F1-score of 95.30%. Compared to the Video Swin Transformer model, these results represent improvements of 2.00% in accuracy and 4.96% in F1-score, respectively. This research is particularly significant in the context of real-time identification of combustion process conditions, optimization of control parameters, and realization of more stable and efficient combustion process control.}, archive = {J_EAAI}, author = {Leyuan Wu and Junlin Wu and Dingxiang Wang and Qiang Fu}, doi = {10.1016/j.engappai.2025.112032}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112032}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A global linear attention incorporated video transformer for robust sintering condition recognition}, volume = {161}, year = {2025}, }
• (2025). Markerless gait analysis for parkinson’s disease diagnosis: A study on machine learning integration and features explainability. EAAI, 161, 112031. (www)

  @article{ , abstract = {Parkinson’s disease is a common neurodegenerative disorder with progressive loss of dopaminergic and other subcortical neurons that significantly impair motor functions, necessitating accurate diagnostic and monitoring techniques. Researchers commonly use gait analysis to estimate gait parameters to classify diseases and their progression. Traditional gait analysis, while effective, typically requires expensive, specialized equipment and lacks the flexibility for widespread clinical use. Furthermore, many existing studies on markeless approaches do not use public datasets, giving comparable and benchmarking results. Additionally, these methods often do not incorporate explainability, leaving a gap in understanding which specific gait features most significantly impact diagnostic outcomes. To address this issue, we propose a novel system to support diagnosis and analysis of the gait of Parkinson’s disease. The two-step approach involves pose extraction with vision-based pose estimation and classification employing machine learning techniques based on the subject’s temporal, kinematic, and symmetric characteristics. The explainability technique used in this study shows the important role of features such as the knee angle, cadence, and double support duration in influencing diagnostic outcomes. The system shows promising results for non-invasive, low-cost Parkinson’s disease diagnostics, reaching an accuracy of 0.95, a sensitivity of 0.90, a specificity of 0.97, a Matthew’s Correlation Coefficient (MCC) of 0.88, an Area Under the Curve (AUC) of 0.99.}, archive = {J_EAAI}, author = {Cesare Davide Pace and Alessandro Marco De Nunzio and Claudio De Stefano and Francesco Fontanella and Mario Molinara}, doi = {10.1016/j.engappai.2025.112031}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112031}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Markerless gait analysis for parkinson’s disease diagnosis: A study on machine learning integration and features explainability}, volume = {161}, year = {2025}, }
• (2025). Offshore wind power multi-step forecasting based on multi-scale attention mechanism fusion network. EAAI, 161, 112026. (www)

  @article{ , abstract = {The nonlinear, unstable, and multi-scale characteristics of offshore wind power present significant challenges for grid scheduling and power forecasting. To address this issue, this paper proposes a multi-step forecasting hybrid model for offshore wind power based on the sequence-to-sequence (Seq2Seq) architecture, which is named Multi-scale, Denoising, Attention Fusion Network (mDAFNet). The model integrates a multi-level learnable wavelet decomposition network, frequency-domain interpolation denoising technique, and dual-attention mechanism, and improves forecasting performance through joint high- and low-frequency dual-branch predictions. The process begins by improving data quality through outlier cleaning and feature selection based on grey correlation analysis and mutual information. Subsequently, a multi-level learnable wavelet decomposition network is introduced to extract high- and low-frequency subsequences from the frequency domain and optimize parameters dynamically through backpropagation. To address the noise impact on high-frequency subsequences, frequency domain interpolation denoising technique is applied for noise reduction. Then, a sequence-to-sequence model based on the dual-attention mechanism is used to forecast the denoised high-frequency subsequences, while a Long Short-Term Memory (LSTM) neural network predicts the low-frequency subsequences. Finally, the dual-branch forecasting network enables joint modeling of high- and low-frequency subsequences. We conducted experiments on two offshore wind power datasets. The results indicate that, compared to baseline models, the proposed mDAFNet reduces Mean Absolute Error, Root Mean Square Error, and Mean Absolute Percentage Error by an average of 33.94 %, 36.62 %, and 39.84 %, respectively, while significantly improving forecasting accuracy and stability. This study provides a new approach and perspective for offshore wind power forecasting.}, archive = {J_EAAI}, author = {Yizhuang Xiong and Wenbo Wang}, doi = {10.1016/j.engappai.2025.112026}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112026}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Offshore wind power multi-step forecasting based on multi-scale attention mechanism fusion network}, volume = {161}, year = {2025}, }
• (2025). Weighted multi-granularity fuzzy probabilistic rough set based on semi-overlapping function and its application in three-way decision. EAAI, 161, 112023. (www)

  @article{ , abstract = {This paper introduces a novel model of fuzzy probabilistic rough sets and develops an interesting three-way decision (TWD) applicable to multi-attribute group decision-making and classification problems. First, we simultaneously consider the cut-levels of fuzzy relations and decision-making fuzzy sets, investigating a new fuzzy probabilistic rough set model. This model employs semi-overlapping functions as the range of the membership functions of the fuzzy set, enhancing interpretability since the semi-overlapping functions can act as the truth tables of fuzzy logic. Second, by assigning different weights to various granular fuzzy relations, we propose a weighted multi-granularity fuzzy probabilistic rough set model. It is proven that the proposed (weighted multi-granularity) fuzzy probabilistic rough set model encompasses many existing models as special cases, thereby providing a very broad framework for rough set theory. Third, by integrating the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) method with Bayesian theory, we develop a method for computing the threshold of fuzzy probabilistic rough sets and construct a new TWD. Sorting and classification rules and algorithms for TWD are provided. Finally, we perform parameter analysis and comparative experiments on nine public datasets to verify the effectiveness and rationality of our TWD. The experimental results demonstrate that our approach exhibits superior fault tolerance and noise reduction capabilities. Specifically, the introduction of semi-overlapping functions enhances our TWD model’s performance, with A c c u r a c y improving the most on the Heart dataset, reaching 9.37%, and the F 1 score increasing the most on the Heart and Plrx datasets, reaching 5.3%. In summary, this paper offers a novel solution for applying artificial intelligence in decision-making scenarios.}, archive = {J_EAAI}, author = {Xinru Li and Lingqiang Li and Chengzhao Jia}, doi = {10.1016/j.engappai.2025.112023}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112023}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weighted multi-granularity fuzzy probabilistic rough set based on semi-overlapping function and its application in three-way decision}, volume = {161}, year = {2025}, }
• (2025). Feature-driven double deep Q-network with iterated greedy for intelligent scheduling optimization in reentrant hybrid flow shops. EAAI, 161, 112012. (www)

  @article{ , abstract = {The reentrant hybrid flow shop scheduling problem (RHFSP) poses significant challenges due to its complex structure and limited research on model accuracy, solution space exploration, and adaptive strategy selection. To address these gaps, this paper proposes a novel feature-driven double deep Q-network with iterated greedy (FD3QNIG) algorithm. First, a five-dimensional mixed-integer linear programming (MILP) model is developed to significantly improve modeling accuracy. Second, a feature-driven initialization strategy (FDNEH) enhances the quality of initial solutions, while an adaptive historical information-driven destruction and reconstruction strategy (AHDDR) effectively balances exploration and exploitation during the search process. Third, multi-scale local search strategies are employed, including the first application of comprehensive exploration-driven mandatory operations local search (CED_MOLS) to RHFSP, substantially deepening the search capability. Fourth, an adaptive strategy selection mechanism based on double deep Q-Network (DDQN_ASS) dynamically guides the algorithm toward more efficient decision-making. Extensive experiments on 285 benchmark instances demonstrate that FD3QNIG achieves a 53 %–94 % improvement in average relative percentage increase ( A R P I ) over state-of-the-art methods, confirming its effectiveness and robustness.}, archive = {J_EAAI}, author = {Chexiang Li and Yuyan Han and Yuting Wang and Yiping Liu and Biao Zhang and Leilei Meng}, doi = {10.1016/j.engappai.2025.112012}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112012}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature-driven double deep Q-network with iterated greedy for intelligent scheduling optimization in reentrant hybrid flow shops}, volume = {161}, year = {2025}, }
• (2025). A semi-supervised framework for generating multi-dimensional taxonomies from asset maintenance documents. EAAI, 161, 112010. (www)

  @article{ , abstract = {The operation and maintenance of buildings generate large volumes of unstructured textual data, such as inspection reports and service requests. These records contain valuable insights that can support fault detection, cost tracking, and resource planning. However, existing classification approaches often rely on static, expert-defined labels that fail to reflect the complexity of real-world maintenance operations. This paper introduces a hybrid framework that combines sentence embedding, clustering, topic modeling, and network modularization to uncover recurring patterns in maintenance text. The extracted patterns are then reviewed and refined by facility management experts to develop a multi-dimensional taxonomy model tailored to operational needs. The methodology is applied to a case study involving over 30,000 work orders. The results demonstrate how the proposed system captures fine-grained details such as system type, failure mode, and required trade expertise. A proof-of-concept software tool, developed in collaboration with facility managers, showcases the practical value of the taxonomy in enabling data-driven decision-making, such as identifying cost drivers and recurring issues. Additionally, the resulting taxonomy models serve as effective prompts for zero-shot text classification, enabling large language models to classify new maintenance records without requiring retraining or labeled data. This approach provides a scalable and adaptable foundation for text classification systems in asset management.}, archive = {J_EAAI}, author = {Soroush Sobhkhiz and Tamer El-Diraby}, doi = {10.1016/j.engappai.2025.112010}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112010}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A semi-supervised framework for generating multi-dimensional taxonomies from asset maintenance documents}, volume = {161}, year = {2025}, }
• (2025). A comprehensive review of facial beauty prediction using deep learning techniques. EAAI, 161, 112009. (www)

  @article{ , abstract = {Facial beauty prediction (FBP) is an emerging area of artificial intelligence (AI) that focuses on the development of models that analyze facial features to assess beauty, based on human perception. This task is particularly challenging due to the subjective nature of beauty and limited resources. Deep learning methods have proven their exceptional ability to capture complex features and are therefore well suited for FBP tasks. This paper reviews recent advances in FBP, focusing on deep learning techniques and benchmark datasets. A proposed taxonomy organizes the main methods according to their design and applications, and comparative analyzes highlight trends and the performance of different models. Finally, the study outlines future research directions to drive progress in this evolving field.}, archive = {J_EAAI}, author = {Djamel Eddine Boukhari and Fadi Dornaika and Ali Chemsa and Abdelmalik Taleb-Ahmed}, doi = {10.1016/j.engappai.2025.112009}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112009}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A comprehensive review of facial beauty prediction using deep learning techniques}, volume = {161}, year = {2025}, }
• (2025). Maximum relevant minimum redundant multi-label feature selection using ant colony optimization. EAAI, 161, 112007. (www)

  @article{ , abstract = {Multi-label learning tasks involve instances that may belong to multiple categories simultaneously, making feature selection particularly challenging in high-dimensional feature spaces. Existing multi-label feature selection methods often suffer from limitations such as high computational complexity, inadequate handling of feature redundancy, and insufficient modelling of label dependencies. To overcome these challenges, we propose a novel framework called Maximum Relevant Minimum Redundant Multi-Label Feature Selection (MR2MLFS), which integrates a two-layer graph representation with a modified Ant Colony Optimization (ACO) strategy. The first graph layer clusters correlated features using Louvain community detection, while the second constructs a meta-graph to model inter-cluster relationships. ACO then explores this structure, favouring the selection of highly relevant and non-redundant features. To reduce computational overhead, we introduce an information-theoretic metric that estimates both feature-label relevance and feature-feature redundancy, eliminating the need for repeated classifier training during the search. We evaluated the proposed method on ten benchmark multi-label datasets using several multi-label classifiers. Experimental results show that the proposed method outperforms six state-of-the-art methods across multiple evaluation metrics, achieving an average relative improvement of 5–12 % while reducing feature dimensionality by up to 80 %. These results confirm the method's robustness, efficiency, and effectiveness in multi-label feature selection.}, archive = {J_EAAI}, author = {Mohammad Hatami and Parham Moradi and Sadegh Sulaimany and Mahdi Jalili}, doi = {10.1016/j.engappai.2025.112007}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112007}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Maximum relevant minimum redundant multi-label feature selection using ant colony optimization}, volume = {161}, year = {2025}, }
• (2025). A two-stage model for unified sentence- and document-level biomedical event extraction. EAAI, 161, 112001. (www)

  @article{ , abstract = {Biomedical event extraction, a cornerstone of information extraction, has increasingly attracted attention within the biomedical research community. Moreover, it is a highly complex task, which not only deals with many sub-tasks but also involves nested events. Currently, the research on biomedical event extraction, whether pipelined model or joint method, needs to be processed for each sub-task. The process of processing each sub-task one by one lead to the degradation of event extraction performance. In addition, most studies focus on extracting sentence-level events and ignore cross-sentence event information. To solve these problems, we simplify the process of event extraction, reduce the processing steps, and combine the two sub-tasks of relation extraction and argument combination as one sub-task. In addition, we consider document-level event extraction, which not only extracts cross-sentence events but also considers broader context information. Experimental results indicate that our novel approach outperforms prior studies. Additionally, the document-level event extraction model attains the top performance on the BioNLP’11 test data and achieves near-leading performance on the BioNLP’13 test data.}, archive = {J_EAAI}, author = {Fangfang Su and Yue Zhang and Pengfei Jiao and Zhidong Zhao and Bobo Li and Fei Li and Donghong Ji}, doi = {10.1016/j.engappai.2025.112001}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {112001}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A two-stage model for unified sentence- and document-level biomedical event extraction}, volume = {161}, year = {2025}, }
• (2025). A position-aware sets based weakly supervised framework for whole-slide subtype classification. EAAI, 161, 111998. (www)

  @article{ , abstract = {Identifying cancer subtypes is essential for personalized treatment and accurate prognosis due to the varying sensitivities of subtypes to therapies. However, in cancer subtype classification tasks, normal slides are usually scarce or absent as negative samples, while subtype whole-slide images (WSIs) often contain extensive unannotated normal tissue regions. These regions introduce significant noise during feature fusion and subtype classification, leading to degraded performance of existing weakly supervised methods In this paper, we propose the Position-aware Sets based Weakly Supervised learning framework (PSWS), designed for cancer subtype classification using WSIs, with a two-stage structure to enhance model efficiency. Specifically, it first presents a novel patch organization approach, distinct from the bag concept of traditional Multiple Instance Learning (MIL), called position-aware sets, as basic units for learning. Then, PSWS automatically selects subtype-specific features based on enhanced histological features and mutual-patch relations, mitigating the negative impact of unannotated negative regions. In the experiments, the superior performance of PSWS over representative MILs is validated through subtype classification tasks on both public datasets and our internally constructed dataset. Furthermore, class probabilities of position-aware sets and attention region visualizations demonstrate its post-hoc interpretability, assisting pathologists in locating suspicious areas.}, archive = {J_EAAI}, author = {Jiuman Song and Bo Yu and Xiaomin Liu and Lele Cong and Zilong Zhou and Xianling Cong and Hongyan Sun and Shuchao Pang and Hechang Chen}, doi = {10.1016/j.engappai.2025.111998}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111998}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A position-aware sets based weakly supervised framework for whole-slide subtype classification}, volume = {161}, year = {2025}, }
• (2025). Action understanding in low-light and pitch-dark conditions: A comprehensive survey. EAAI, 161, 111996. (www)

  @article{ , abstract = {Action understanding in low-light and dark environments (AULLD) is a critical and technically demanding challenge at the intersection of computer vision and engineering applications. This task focuses on interpreting human actions using motion cues under visually constrained conditions, such as poor illumination or complete darkness. Since 2010, machine learning and, more notably, deep learning have significantly advanced AULLD by enabling systems to extract and learn discriminative features from complex, low-visibility data. These techniques have resulted in state-of-the-art solutions, with recent methods achieving accuracies on various benchmark datasets of up to 97.10%, and a 73.20% top-1 accuracy on an infrared dataset. These solutions have had applications across domains such as healthcare monitoring, surveillance and security, human–computer interaction, and social computing. In this survey, we provide the first comprehensive overview of the progress of the research and topical developments in AULLD. We cover a broad range of elements, including the datasets, evaluation protocols, methods, challenges, and emerging research directions. To organize and evaluate the literature in a structured, integrated way, we introduce a novel taxonomic framework, grounded in four core dimensions: body representation, temporal representation, feature representation, and neural architectures. Employing this taxonomy, we perform an analysis of AULLD approaches, exploring their methodological foundations, system design considerations, performance outcomes, and inherent trade-offs. The survey concludes with an in-depth discussion of the existing challenges confronting AULLD, highlighting several promising research directions that hold potential for advancing the field and setting the stage for future innovations and improvements in AULLD.}, archive = {J_EAAI}, author = {Muhammad Munsif and Samee Ullah Khan and Noman Khan and Altaf Hussain and Sung Wook Baik}, doi = {10.1016/j.engappai.2025.111996}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111996}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Action understanding in low-light and pitch-dark conditions: A comprehensive survey}, volume = {161}, year = {2025}, }
• (2025). Transfer learning framework integrating attention mechanism and domain adaptation for low earth orbit satellite network traffic prediction. EAAI, 161, 111992. (www)

  @article{ , abstract = {Traffic prediction is a crucial prerequisite for planning and even network security in Low Earth Orbit (LEO) satellite networks (LSNs). This paper designed a transfer learning framework for LSN traffic prediction that leverages attention mechanism and domain adaptation. Firstly, by integrating five-dimensional data including global population distribution, local time coefficient, the Internet penetration rate of the country, daily data volume of a single Internet user, and global aeronautical traffic demand, a traffic model that could characterize the traffic situation of the area covered by LEO satellite within a specific time range was constructed. Considering the problem of insufficient online traffic data, knowledge was transferred from terrestrial network traffic (source domain) to satellite network traffic (target domain) by incorporating the Domain-Adversarial Neural Network (DANN) method to tackle the data distribution discrepancies between the source and target domains. Finally, by combining DANN with the attention mechanism, the domain-invariant features of the source domain and the target domain were extracted to predict satellite network traffic accurately. Experimental results show that compared to baseline models, the error of the proposed framework in terms of root mean square error measurement is reduced by 9.57% to 33.47% and 18.85% to 38.99% in the two simulated LSN traffic scenarios. Moreover, this framework has low computational complexity among other transfer learning models, which can lay a foundation for subsequent satellite traffic planning and network security.}, archive = {J_EAAI}, author = {Yan Zhang and Yong Wang and Qingsong Zhao and Yadi Zhai and Zhi Lin and Luda Zhao and Yihua Hu}, doi = {10.1016/j.engappai.2025.111992}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111992}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transfer learning framework integrating attention mechanism and domain adaptation for low earth orbit satellite network traffic prediction}, volume = {161}, year = {2025}, }
• (2025). Skeleton-based multi-person action recognition towards real-world violence detection. EAAI, 161, 111987. (www)

  @article{ , abstract = {Human action recognition is crucial in intelligent systems like robotics, healthcare, and surveillance, gaining significant attention with the rise of AI in computer vision. This work presents the Multi-Skeleton Action Recognizer (MSAR), a comprehensive multi-stage process specifically designed to enhance the recognition and classification of violent activities involving multiple individuals. The approach aims to improve accuracy and robustness in real violence detection and classification leveraging skeletal data. The proposed MSAR pipeline consists of five main stages: Human detection, Human tracking, Skeleton extraction, Data pre-processing and Padding, and Action classification. Experimental results demonstrate the effectiveness of our multi-stage process, showing significant improvements in multi-person action recognition in violence detection. In the human detection stage, our integration of multiple detection models provides a comprehensive visual comparison of model performance, offering empirical insights into accuracy, computational efficiency, and processing speed. The fusion Bidirectional Long Short-Term Memory - Gated Recurrent Unit (BiLSTM-GRU) architecture designed for action classification integrating Deep Bidirectional Long Short-Term Memory (BiLSTM) and Gated Recurrent Unit (GRU) components, demonstrated maximum accuracy of 96.46% and an average accuracy of 91.18%. The proposed architecture is explored to potentially enhance the model’s ability to capture temporal dependencies in action sequences. Additionally, our proposed multi-stage pipeline for action recognition exemplifies a modular, systematic approach, where each stage functions as an independent module is expanded, optimized, replaced without affecting others. This modularity enhances the flexibility, adaptability, scalability of the pipeline, providing a solid foundation for future advancements across various tasks, application domains.}, archive = {J_EAAI}, author = {Minh-Trieu Truong and Van-Dung Hoang}, doi = {10.1016/j.engappai.2025.111987}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111987}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Skeleton-based multi-person action recognition towards real-world violence detection}, volume = {161}, year = {2025}, }
• (2025). Malware attack and defense game in fractional-order internet of underwater things: Model-based and model-free approaches. EAAI, 161, 111970. (www)

  @article{ , abstract = {With the rapid development of the Internet of Underwater Things (IoUT), its cybersecurity faces increasingly severe challenges, as malware attacks have become a common and highly destructive threat. However, most existing studies rely on integer-order models or static defense mechanisms, which fail to capture the memory and hereditary properties of underwater malware propagation and lack adaptability in dynamic adversarial environments. To address this issue, this paper constructs a fractional-order propagation model and introduces differential game theory to investigate the dynamic interactions between malware and system defense, aiming to derive the optimal dynamic strategies and the corresponding Nash equilibrium for both parties. Furthermore, two innovative artificial intelligence strategy learning approaches are proposed: a model-based method and a model-free method, designed to approximate the optimal control strategies in the game. Extensive comparative experiments and simulation results demonstrate the effectiveness of the proposed methods in strategy learning. We further discuss the practical applications of these methods in IoUT as well as their scalability to other domains.}, archive = {J_EAAI}, author = {Guiyun Liu and Zulong Peng and Tingting Tan and Xiaojing Zhong and Zhongwei Liang}, doi = {10.1016/j.engappai.2025.111970}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111970}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Malware attack and defense game in fractional-order internet of underwater things: Model-based and model-free approaches}, volume = {161}, year = {2025}, }
• (2025). Semi-supervised graph contrastive learning for emotion recognition based on electroencephalogram signals. EAAI, 161, 111969. (www)

  @article{ , abstract = {Emotion recognition plays a crucial role in affective computing, aiming to identify and interpret human emotions. Since electroencephalography (EEG) measures brain electrical activity associated with emotional processing, its use in emotion recognition has garnered significant attention. However, due to the complexity of EEG, capturing the low-dimensional manifold structure from a high-dimensional EEG signal remains a challenge. Moreover, even though EEG labeling is expensive and requires specialization, the development of emotion recognition models has predominantly relied on labeled EEG datasets. To address these issues, we propose a semi-supervised graph contrastive learning (SGCL) model for EEG-based emotion recognition, leveraging the abundance of unlabeled EEG data to generate clearer representations. The proposed SGCL extracts two frequency-domain features from the EEG signal, differential entropy (DE) and power spectral density (PSD), which are integrated through a symmetric similarity network fusion (SSNF) with graph contrastive learning (GCL) to improve the generalization and representation capability of the graph convolutional network (GCN) to complex EEG data in transductive learning tasks. Extensive experimental results on three public datasets reveal that the proposed SGCL consistently outperforms state-of-the-art models. Notably, on the Shanghai Jiao Tong University EEG Emotion dataset (SEED), SGCL achieves an average accuracy of 99.99% using only 3.5% of the data as labeled input, setting a new benchmark in both effectiveness and efficiency. Further analyses confirm our model learns highly discriminative representations and offers insightful explanations, demonstrating its potential for robust emotion recognition.}, archive = {J_EAAI}, author = {Dae Hyeon Kim and Young-Seok Choi}, doi = {10.1016/j.engappai.2025.111969}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111969}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised graph contrastive learning for emotion recognition based on electroencephalogram signals}, volume = {161}, year = {2025}, }
• (2025). Explainable automated wild-orchid identification combining deep neural networks and bayesian networks. EAAI, 161, 111961. (www)

  @article{ , abstract = {Deep learning has been shown repeatedly to be a successful method of obtaining accurate classifiers. This also applies to orchid identification from digital photographs. However, deep neural networks possess the major weakness of lack of explainability, missing the ability to explain the reasons behind a decision. Nevertheless, most current research regarding automated orchid identification applies this blackbox approach. By contrast, in this paper we propose a new method for trustworthy automated orchid identification combining two complementary methods: deep neural networks and feature-based Bayesian networks, where the Bayesian network is also utilized for providing an explanation of the generated solutions. We use other deep neural networks to extract flower characteristics, the features, from the images which are subsequently fed into the Bayesian network as uncertain evidence. When combining the deep neural network and the Bayesian network as an ensemble classifier, both reaching the same conclusion, an accuracy of 89.4% is achieved, the most trustworthy outcome. With a human-in-the-loop ensemble classifier, validation results are even better, yielding an accuracy of 98.1%. Our approach also exploits the taxonomic knowledge represented in the Bayesian network to provide an explanation of the solutions for every case, reinforcing further trust in the method. The result is an explainable user-in-the-loop ensemble classifier. Providing explainability can help build user trust in a system and may play a major role when it is used as a learning aid for new orchid enthusiasts. Finally, the proposed method may be also of value in many fields other than plant determination.}, archive = {J_EAAI}, author = {Diah Harnoni Apriyanti and Luuk J. Spreeuwers and Peter J.F. Lucas}, doi = {10.1016/j.engappai.2025.111961}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111961}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable automated wild-orchid identification combining deep neural networks and bayesian networks}, volume = {161}, year = {2025}, }
• (2025). A safe multi-agent reinforcement learning algorithm using constraint update projection approach. EAAI, 161, 111929. (www)

  @article{ , abstract = {Traditional reinforcement learning has a major limitation, which is that they optimize agent’s policy purely for maximizing rewards. It completely ignore safety considerations. However, in certain critical engineering fields, ensuring safety is of utmost importance, otherwise it can cause incalculable losses. Therefore, this paper proposes a safe Multi-Agent Constrained Update Projection(MACUP) algorithm, which can safely control agents to complete tasks. We solve this problem from the perspective of policy constraint optimization. Firstly, we derive the new bounds of multi-agent policy performance difference based on a tighter general policy performance difference. It contains generalized advantage estimates, and we utilize these bounds as surrogate functions concerning the objective and constraints. Secondly, to address the coordination issue among multiple agents, we employ a multi-agent sequential policy update framework. Finally, we use a projection method to optimize policies, which has low computational complexity and does not require convex approximation of the surrogate function for solving. It can help us reduce errors. Finally, we have validated our algorithm in two different multi-agent safety environments, and the results show that it is able to satisfy safety constraints while achieving higher rewards.}, archive = {J_EAAI}, author = {Yang Liu and Xiang Feng and Huiqun Yu}, doi = {10.1016/j.engappai.2025.111929}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111929}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A safe multi-agent reinforcement learning algorithm using constraint update projection approach}, volume = {161}, year = {2025}, }
• (2025). Multi-criteria group decision-making method using spherical fuzzy Z-numbers for smart technology revolution in municipal waste management. EAAI, 161, 111928. (www)

  @article{ , abstract = {Due to the new phenomenon of globalization and the fast growing cities, there is drastic pressure on the conventional methods of waste management hence the need for innovative management wastage that is proactive to the theme of environmental and economic challenge of the modern world. This paper aims at investigating the following research question: How has the introduction of smart technologies impacted waste management in the context of a mid-sized city that is in the cross-road of generating more wastes while at the same time, concerns over ecological attainability. They present a new type of Spherical Fuzzy Z-Number Sets in organizational environments dealing with multi criteria group decision making where it possess higher order of uncertainty than conventional fuzzy sets. For handling the multi facility nature of multi criteria group decision making in waste management, we propose the so-called Additive Ratio Assessment method when the attribute weights are unknown. To ensure that the criteria weights are determined objectively in this research, the CRITIC (CRiteria Importance Through Intercriteria Correlation) technique is used. The first part of the study provides a theoretical framework of spherical fuzzy Z-numbers concerning accuracy, scoring functions, and operations. Then we introduce this framework to actual multi criteria group decision making cases in municipal waste management to show that how Spherical Fuzzy Z-Number can facilitate decision-making processes by dealing with the vagueness and fuzziness of stakeholder preferences. The TODIM (Tomada de Decisao Iterativa Multicriterio) technique is used in this to validate and compare for efficiency of the proposed additive ratio assessment method. Besides, this research will not only enhance the theoretical aspect of fuzzy decision making but also present a feasible and effective framework for handling unsound and random decision making problems especially within the context of urban waste management.}, archive = {J_EAAI}, author = {Shahzaib Ashraf and Muhammad Naeem and Chiranjibe Jana and Maria Akram and Gerhard-Wilhelm Weber}, doi = {10.1016/j.engappai.2025.111928}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111928}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-criteria group decision-making method using spherical fuzzy Z-numbers for smart technology revolution in municipal waste management}, volume = {161}, year = {2025}, }
• (2025). From image processing to artificial intelligence-driven tools: A comprehensive survey on the evolution of feature extraction methods in paintings. EAAI, 161, 111794. (www)

  @article{ , abstract = {This study explores the computational process of converting qualitative elements of paintings, such as shape, color, texture, and line, into quantitative numerical values, which aid in identifying and extracting features from painting images. These identified features are then used to classify paintings based on artist, art style, genre, and art movements. This review employs a systematic literature review methodology, examining approximately 80 research papers to track trends in this field from 2015 to 2025. With the increasing presence of paintings in online media, museums, and galleries, artificial intelligence (AI) plays a significant role in interpreting these subjective elements. This review examines various image processing techniques in conjunction with AI implementations to extract the local and global features of a painting. These methods are combined with AI models, such as deep learning and computer vision algorithms, to improve feature extraction and provide a more thorough and accurate paintings analysis.}, archive = {J_EAAI}, author = {Rekha Sharma and Rishi Gupta and Aditya Sinha}, doi = {10.1016/j.engappai.2025.111794}, journal = {Engineering Applications of Artificial Intelligence}, month = {12}, pages = {111794}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {From image processing to artificial intelligence-driven tools: A comprehensive survey on the evolution of feature extraction methods in paintings}, volume = {161}, year = {2025}, }
• (2025). Deep residual network guided by channel multi-scale attention mechanism for predicting free-field sound absorption performance. EAAI, 160, 112095. (www)

  @article{ , abstract = {Accurate prediction of free-field sound absorption performance is crucial for the design and application of acoustic materials. However, traditional free-field methods typically assume that the sample is infinitely large, which can lead to significant inaccuracies in the low- and mid-frequency ranges due to finite-size effects. Moreover, the adoption of spherical wave sound field models suffers from the lack of explicit analytical solutions, resulting in a solution process that relies on numerical iterations sensitive to initial values. To address this issue, this paper develops a nonlinear inverse solver based on a channel multi-scale deep residual network (CMDRN), which directly infers the free-field sound absorption performance of materials from measured spatial acoustic field data without relying on prior knowledge of material parameters and iterative inversion methods. Firstly, a physically informed dataset that explicitly accounts for finite-size effects is constructed by integrating the pressure–particle velocity (PU) method with the boundary element method (BEM). Subsequently, CMDRN is trained as an alternative solver for nonlinear inversion, enabling efficient and accurate prediction of material sound absorption from spatial sound fields. By integrating the strengths of deep residual networks and the channel multi-scale attention mechanism, CMDRN enhances the model's ability to capture key spatial features. Finally, through comparisons with other methods, the accuracy of CMDRN in predicting sound absorption performance is validated, and an experimental case is presented to further highlight the potential of the proposed method in real-world applications.}, archive = {J_EAAI}, author = {Zu-Jie Yang and Yong-Bin Zhang and Chang-Jun Zheng and Liang Xu}, doi = {10.1016/j.engappai.2025.112095}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112095}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep residual network guided by channel multi-scale attention mechanism for predicting free-field sound absorption performance}, volume = {160}, year = {2025}, }
• (2025). BridgeFusionNet: A hybrid convolutional-transformer architecture for road surface crack. EAAI, 160, 112085. (www)

  @article{ , abstract = {Road surface crack detection is essential for the reliability and preventive maintenance of transportation infrastructure. Due to the small and narrow nature of cracks, as well as the presence of background interferences such as cluttered textures, variations in lighting, and structural complexity, crack detection poses significant challenges. Therefore, we propose a novel BridgeFusionNet to segment road surface cracks. We first employ two backbone networks, a Convolutional Neural Network (CNN) and a Transformer, as encoders for extracting multi-scale features. Then, the Feature Bridging Fusion Model (FBFM) is developed to iteratively aggregate the features extracted by each layer of the CNN and the Transformer to refine the object progressively. FBFM adopts Channel Attention and Coordinate Attention to enhance feature and location information, thus improving long-distance CNN modeling ability to reduce the discrepancy between CNN and Transformer features. We compared the proposed model with state-of-the-art segmentation models on standard datasets. The results show that our network can segment pavement cracks in severe disturbances with 1.9907 % and 1.5018 % higher Intersection over Union (IoU) performance than the comparative state-of-the-art, demonstrating the superiority of the proposed network under complex segmentation scenes.}, archive = {J_EAAI}, author = {Jia Zheng and Chuan Tang and Yuanxi Sun}, doi = {10.1016/j.engappai.2025.112085}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112085}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {BridgeFusionNet: A hybrid convolutional-transformer architecture for road surface crack}, volume = {160}, year = {2025}, }
• (2025). Imitation-constrained evolutionary learning for multi-locomotion skill of bipedal wheeled robots in unstructured terrains. EAAI, 160, 112079. (www)

  @article{ , abstract = {Bipedal wheeled robots present significant locomotion challenges in unstructured terrains due to inherent dynamic instability and under-actuation characteristics. While reinforcement learning offers promising solutions for mobility enhancement, achieving multi-skill generalization and avoiding local optima in policy optimization remains problematic. This paper proposes an imitation-constrained evolutionary reinforcement learning framework that reformulates multi-locomotion skill acquisition as an evolutionary process. The methodology is characterized by: (1) Task-specific sub-policies with simplified reward functions ensuring local optimality, (2) Expert-student architecture where optimized sub-policies serve as expert models, and (3) Dynamic imitation-constrained Reinforcement Learning strategy enabling progressive skill inheritance during new task learning. The evolutionary mechanism guarantees global optimality of multi-locomotion skill learning through a single agent. Experimental validation demonstrates 192 % cumulative reward improvement compared to end-to-end learning paradigms. Through dynamic parameter randomization, the control strategy was successfully implemented on physical prototypes, achieving zero-shot sim-to-real transfer with multimodal adaptation across muddy rock, grassland, and discrete obstacle terrains.}, archive = {J_EAAI}, author = {Mingquan Zhu and Tie Zhang and Yanbiao Zou}, doi = {10.1016/j.engappai.2025.112079}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112079}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Imitation-constrained evolutionary learning for multi-locomotion skill of bipedal wheeled robots in unstructured terrains}, volume = {160}, year = {2025}, }
• (2025). Design of an intelligent grading system for chinese water chestnuts utilizing advanced artificial intelligence methods. EAAI, 160, 112070. (www)

  @article{ , abstract = {This study aims to overcome limitations in Chinese water chestnut production and processing by developing an innovative grading system that integrated advanced artificial intelligence and engineering design. The artificial intelligence component features an improved convolutional neural network architecture, specifically an enhanced version of the MobileNetV2 architecture combined with a Squeeze-and-Excitation module, which is referred to as MobileSeNet. Compared to the baseline MobileNetV2 model, MobileSeNet achieved superior performance not only in grading accuracy (97.25 %) but also in convergence stability and model generalization, benefiting from enhanced gradient flow and feature recalibration. The engineering component involved a purpose-built hardware platform that evaluated Chinese water chestnuts based on external visual traits using advanced image acquisition, preprocessing, segmentation, edge detection, and ellipse fitting for diameter estimation. A user-friendly host computer interface was incorporated to support real-time monitoring and user interaction. The system introduced in this study offered a low-cost and scalable solution, making it well-suited for implementation in real agricultural scenarios. Overall, this research advances global smart agriculture by delivering a robust and efficient grading solution underpinned by artificial intelligence, with strong adaptability to diverse agricultural products.}, archive = {J_EAAI}, author = {Yinping Zhang and Joon Huang Chuah and Anis Salwa Mohd Khairuddin and Dongyang Chen and Xuewei Zhao and Junwei Huang and Chenyang Xia and Jingjing Li and Wenlong He}, doi = {10.1016/j.engappai.2025.112070}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112070}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Design of an intelligent grading system for chinese water chestnuts utilizing advanced artificial intelligence methods}, volume = {160}, year = {2025}, }
• (2025). Probabilistic mixture model driven interpretable modeling, clustering, and predicting for physical system data. EAAI, 160, 112069. (www)

  @article{ , abstract = {The modeling, clustering, and prediction of measurement data are important processes for converting sensing information into digital representation for physical systems, who are also the premise of sensing results serving engineering practice. Probabilistic machine learning can efficiently complete the above processes and provide interpretability, which is the continuous pursuit for artificial intelligence used for physical systems. Focusing on time series data and discrete data as the main object of research, this paper proposes a set of parametric and interpretable modeling, clustering, and prediction methods of physical system input and output data driven by Gaussian mixture model (GMM). Two original algorithms involving the adaptive determination of the number of GMM components assisted by frequency histogram, as well as the time series prediction of system output driven by multiple bivariate GMMs for physical systems of multiple-input single-output. Real measurement data of transportation and environment from the two bridges is used to verify the effectiveness. The results demonstrated that the root mean squared error between the bin heights of frequency histogram and the estimated GMM function can help the adaptive determination of the optimal number of GMM components for discrete data and time series data; the proposed GMM driven predicting algorithm for the time series prediction of output data of multiple-input single-output system can approach the predictive performance of deep neural networks (for experimental dataset, the correlation coefficient of prediction and measurement for them is 0.9542 versus 0.9686) while providing statistical parameters with physical meaning.}, archive = {J_EAAI}, author = {Hanwei Zhao and Xiaonan Zhang and Youliang Ding and Tong Guo and Aiqun Li and Chee-Kiong Soh}, doi = {10.1016/j.engappai.2025.112069}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112069}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Probabilistic mixture model driven interpretable modeling, clustering, and predicting for physical system data}, volume = {160}, year = {2025}, }
• (2025). Physics-constrained neuroforecasting of time-dependent local scour depths at bridges under unsteady flows. EAAI, 160, 112064. (www)

  @article{ , abstract = {The field of intelligent scour prediction encounters significant challenges in practical applications, particularly in integrating physical mechanisms into data-driven models while addressing the computational intensity of time-series prediction. This study proposes a novel lightweight architecture of transformer with adaptive temporal resolution, specifically designed for local scour prediction under unsteady flow conditions. The architecture links scour physics and machine learning through two coordinated modules: (1) a hydrodynamic feature extractor incorporating pier geometry, sediment transport thresholds, and clear-water scour hydraulics; (2) a logic-enhanced Transformer architecture with temporal attention mechanisms; Benchmark testing on standard computers demonstrated a 5.06 times computational efficiency gain compared to conventional machine learning approaches in engineering case scenarios. The proposed architecture significantly reduces training data dependency through a scour fluid mechanics informed mask that intrinsically embeds time-varying hydraulic parameters and scour parameters, while maintaining interpretability via a physics-consistent scour attention mechanism. This advancement makes it possible to develop a deployable solution for real-time bridge scouring monitoring systems and enhances the operational readiness of flood resilience infrastructure.}, archive = {J_EAAI}, author = {Zhanchen Li and Dawei Guan and Yee-Meng Chiew and Jian-Hao Hong and Jinhai Zheng}, doi = {10.1016/j.engappai.2025.112064}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112064}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-constrained neuroforecasting of time-dependent local scour depths at bridges under unsteady flows}, volume = {160}, year = {2025}, }
• (2025). Automatic measurement of surface parameters of composite braided fabrics based on modified U-shaped network. EAAI, 160, 112063. (www)

  @article{ , abstract = {Braided composite materials have broad application prospects in advanced fields such as aerospace and rail transportation. Studies have shown that surface parameters of braided fabrics—such as braiding angles and pitch lengths—significantly influence their performance. However, measuring these parameters during mass production presents challenges related to automation, efficiency, and precision. Traditional image processing methods often suffer from limited accuracy, while deep learning approaches are difficult to implement, and their generalization ability remains unverified. In this paper, a measurement method based on a modified U-shaped network (denoted as U-Net) is proposed. First, a data-augmented image dataset of fabric surfaces with yarn edge annotations is constructed. Then, an enhanced U-Net model incorporating Dropout and an encoder based on the Visual Geometry Group's 16-layer (denoted as VGG16) network is applied, along with a weighted mixed loss function combining Focal Loss and Dice Loss to improve detection of weak yarn edge regions. Morphological operations and distortion rectification are employed for post-processing and noise reduction. Experimental results show that the proposed method improves measurement accuracy by up to 1.84 % across various fabric samples, demonstrating excellent stability and generalization performance.}, archive = {J_EAAI}, author = {Yicen Gao and Jiale Liu and Zhongde Shan and Zitong Guo and Zheng Sun and Xiangyu Zhu and Jiaqi Sun}, doi = {10.1016/j.engappai.2025.112063}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112063}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic measurement of surface parameters of composite braided fabrics based on modified U-shaped network}, volume = {160}, year = {2025}, }
• (2025). Dynamic neighborhood point cloud registration and optimized path planning for manipulator grasping in complex environments. EAAI, 160, 112062. (www)

  @article{ , abstract = {In industrial automation, real-time object recognition and grasping by manipulators are critical. Although deep learning has excelled in image-based tasks, it often demands significant computational resources, large annotated datasets, and suffers from limited interpretability. Conversely, traditional methods can outperform under constrained data or hardware conditions. Conventional point cloud template registration is computationally intensive and vulnerable to noise and occlusion. To address these issues, we introduce a refined Intrinsic Shape Signatures (ISS) feature extraction scheme that employs a multi-scale dynamic neighborhood search to adaptively adjust local geometric constraints around keypoints. This mechanism analyzes point density in real time across varying radii and iteratively updates neighborhood size to sustain robustness in complex scenes. Experiments show that, compared to the standard Iterative Closest Point (ICP) and an ISS + ICP baseline, our method increases registration accuracy by 6.25 % and 3.34 %, while accelerating processing speed by 21.35 % and 10.18 %, respectively. Furthermore, to overcome the poor convergence and limited obstacle avoidance of Rapidly-Exploring Random Tree Connect (RRT-Connect) in high-dimensional spaces, we propose a composite-optimized RRT-Connect approach. It first generates multiple sampling trees via node resampling, then biases sampling toward the goal to improve path guidance, and finally applies gradient-based pruning to remove redundant nodes and smooth trajectories. Simulation results demonstrate a 28.90 % improvement in planning efficiency and a 29.16 % reduction in path length. Tests on a pick-and-place platform confirm that our optimized algorithms consistently outperform existing methods in both simple and complex scenarios, validating their effectiveness and robustness.}, archive = {J_EAAI}, author = {Chao Sun and Zhi‐Ming Hu and Lin‐Yi Peng and Cheng‐Ji Qin and Chao Zhang and Jia‐Hang Liang and Jian‐Jun Ding}, doi = {10.1016/j.engappai.2025.112062}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112062}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic neighborhood point cloud registration and optimized path planning for manipulator grasping in complex environments}, volume = {160}, year = {2025}, }
• (2025). Self-distilled masked autoencoders for medical images. EAAI, 160, 112055. (www)

  @article{ , abstract = {Self-supervised learning (SSL) is a transformative paradigm in deep learning, reducing reliance on extensive labeled datasets. As a prominent SSL framework, the masked autoencoders (MAE) leverages masked image modeling to learn robust representations. However, the shallow layers of the MAE encoder struggle to capture sufficient contextual information, which may limit the model's overall performance. To address this limitation, we introduce self-distilled masked autoencoders (SD-MAE), a novel framework that embeds a self-distillation mechanism into the MAE encoder to harness transformer architectures’ inherent structural priors. Inspired by DeepCluster , a self-supervised clustering-based approach that uses alternating clustering and refinement strategy, SD-MAE iteratively refines shallow-layer representations by aligning them with pseudo-labels from deeper layers. Specifically, it first generates pseudo-labels from deep-layer class token features with effectively encoded structural priors, and then propagates this knowledge to refine the shallow-layer features via Kullback-Leibler (KL) divergence minimization. We evaluated the efficacy of SD-MAE on two challenging medical image classification tasks: multi-labeled pediatric thoracic disease classification using chest X-rays and multi-class classification using optical coherence tomography (OCT) images. Experimental results demonstrated that SD-MAE consistently outperformed baseline models, including residual network-34 (ResNet-34), vision transformer-small (ViT-S), and the standard MAE, across various evaluation metrics. Notably, SD-MAE achieved an area under the curve (AUC) of 0.757 for pediatric chest X-ray classification and an AUC of 0.995 for OCT classification, accompanied by significant improvements in sensitivity, precision, and F1-score.}, archive = {J_EAAI}, author = {Taeyoung Yoon and Daesung Kang}, doi = {10.1016/j.engappai.2025.112055}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112055}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-distilled masked autoencoders for medical images}, volume = {160}, year = {2025}, }
• (2025). Multi-scale gradient guided retinex for images enhancement under transformer oil. EAAI, 160, 112050. (www)

  @article{ , abstract = {As an important carrier of transformer internal information, images under transformer oil play a vital role in identifying and locating transformer internal faults. However, these images suffer from color distortion, blurred details, and low brightness, which directly affect the accuracy of fault identification and localization. To address this issue, we construct the first large-scale dataset under transformer oil, consisting of 410 images that encompass diverse internal structures and varying degrees of oil degradation. Based on this dataset, we propose a multi-scale gradient guided Retinex algorithm for image enhancement in oil-immersed environments. Firstly, to alleviate selective attenuation of light, we propose a gridded hybrid multi-channel compensation algorithm. Then, to address color distortion, we propose a standard deviation constrained Gray World algorithm. Finally, based on the information distribution characteristics of Retinex reflection components at different scales, we propose a multi-scale dynamic guided filtering and gradient adaptive detail enhancement algorithms. Extensive experiments demonstrate that the enhanced results exhibit vivid color and sharp structure, significantly improving the effectiveness of fault detection under transformer oil. The dataset is available at https://drive.google.com/file .}, archive = {J_EAAI}, author = {Hu Qiang and Yuzhong Zhong and Songyi Dian}, doi = {10.1016/j.engappai.2025.112050}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112050}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale gradient guided retinex for images enhancement under transformer oil}, volume = {160}, year = {2025}, }
• (2025). Scaleformer: A hierarchical transformer network utilizing multi-scale spatial-temporal features for electricity consumption prediction in cement raw material grinding system. EAAI, 160, 112046. (www)

  @article{ , abstract = {Accurate electricity consumption prediction for cement raw material grinding system is crucial for energy saving and emission reduction. However, due to nonlinearity, strong coupling, and multiple operating conditions, along with the insufficient utilization of multi-scale spatial-temporal features by most existing models, leads to unsatisfactory prediction accuracy. To solve this issue, this paper proposes Scaleformer, a hierarchical Transformer network that utilizes multi-scale attention mechanism to extract multi-scale spatial-temporal features. To extract multi-scale temporal features, Local Correlation Information Embedding first enhances the correlation representation between consecutive time steps. Then, Temporal Multi-scale Attention utilizes the fine-grained vectors from Local Correlation Information Embedding and the coarse-grained vectors from Time Segment Fusion to extract deep hidden temporal features at different scales. To extract multi-scale spatial features, Spatial Multi-scale Window Attention extracts spatial features at different scales by dynamically adjusting the size of the attention window. Finally, Prediction Layer fuses feature information from different scales to predict electricity consumption. To validate the superior performance of the proposed model, evaluation metrics including mean absolute error (MAE), mean squared error (MSE), mean absolute percentage error (MAPE), coefficient of determination (R 2 ), variance, and Diebold-Mariano (DM) statistic are employed to compare it with seven comparison models. The results show that the proposed model significantly outperforms all comparison models in terms of accuracy, effectiveness, significance, stability, and robustness.}, archive = {J_EAAI}, author = {Gang Liu and Bocheng Cao and Xiaochen Hao and Jiaan Huang and Huan Guo}, doi = {10.1016/j.engappai.2025.112046}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112046}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Scaleformer: A hierarchical transformer network utilizing multi-scale spatial-temporal features for electricity consumption prediction in cement raw material grinding system}, volume = {160}, year = {2025}, }
• (2025). A friction temperature model for dynamic operating bearing based on CNN and CNNLSTM. EAAI, 160, 112042. (www)

  @article{ , abstract = {Bearing temperature is a critical parameter affecting operational longevity and indicating potential failure. Operating parameters influence the temperature rise and its distribution in the rolling bearings, such as speed, load, external conditions, grease content, and clearance. Traditionally, the SKF model has been utilized to estimate bearing temperature. This study developed a high-speed bearing test rig to experimentally evaluate two different bearings under various conditions, considering the aforementioned influencing factors. Convolutional Neural Networks (CNN) and a hybrid Convolutional Neural Network combined with Long Short-Term Memory (CNNLSTM) models were employed to enhance the accuracy of bearing temperature prediction. A comparative analysis of the performance of these models against the SKF model revealed that both CNN and CNNLSTM models outperformed the traditional SKF approach in terms of accuracy, stability, and robustness, with the CNNLSTM model demonstrating superior predictive performance. For bearing 1 under 9 steady thermal cases, the coefficients of determination (R 2 ) for the SKF, Back Propagation (BP), Genetic Algorithm-Back Propagation (GABP), CNN, and CNNLSTM models are 0.6452, 0.856, 0.9051, 0.6558, and 0.917, respectively. The Root Mean Square Error (RMSE) values are 7.0034, 4.4608, 3.63, 6.3814, and 3.6211, respectively, and the Mean Absolute Percentage Error (MAPE) values are 0.0936, 0.088, 0.0579, 0.0779, and 0.0547, respectively. Bearing 2 yields a similar result, with a slight difference. Furthermore, the study investigated the effects of various operational factors on bearing temperature. In particular, an increase in both axial and radial loads results in higher bearing temperatures. This indicates a nonlinear relationship between temperature increase and load magnitude. As the rotational speed increases, the bearing temperature increases linearly with it. The proposed algorithmic models for bearing temperature estimation eliminate the need for individual modeling of bearing dynamics, thermodynamics, and lubrication mechanisms while significantly improving performance metrics. This research offers substantial potential for advancing bearing condition monitoring and temperature regulation systems, contributing to more effective and efficient predictive maintenance strategies.}, archive = {J_EAAI}, author = {Changcheng Deng and Jianfeng Lu and Linchao An and Zhiqiang Gao and Xueli Cheng and Changliu Tian}, doi = {10.1016/j.engappai.2025.112042}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112042}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A friction temperature model for dynamic operating bearing based on CNN and CNNLSTM}, volume = {160}, year = {2025}, }
• (2025). Numerical analysis of heat and mass transfer in off-centered stagnation point casson fluid flow over a rotating disc with thermophoretic particle deposition and artificial neural network-based optimization. EAAI, 160, 112041. (www)

  @article{ , abstract = {Thermophoretic particle deposition (TPD) plays a crucial role in environmental engineering, semiconductor manufacturing, targeted drug delivery systems, and diagnosis. On the other hand, Artificial Neural Networks (ANN) are widely applied in predicting and optimizing complex systems where the analytical models are very difficult to formulate. In the field of fluid dynamics, the ANN helps in forecasting skin friction, rate of heat, and mass transfer without repeated trials, enabling design optimization and real-time control. Because of the above-stated application, the present work investigated the stagnation point flow of a Casson fluid via an off-centered rotating disc with the effect of TPD. The stable, incompressible, three-dimensional Casson fluid flow with the influence of thermal radiation (T-R) and inclined magnetic fields was also taken into consideration. By using appropriate similarity variables, partial differential equations get converted into ordinary differential equations. These equations and the corresponding boundary conditions (B-Cs) are then solved using the Runge-Kutta-Fehlberg-45 (RKF-45) scheme and the Shooting technique. The numerical results on the effect of several nondimensional characteristics across different profiles are presented, together with a detailed discussion of important engineering coefficients. Further, artificial neural network was trained using Leaky ReLu activation function and Adam optimizer on the RKF -45 generated data to examine the heat and mass distribution characteristics. The ANN served as a surrogate model with performance validated with regression plots, Mean Square error and R 2 metrices. The results show that increasing thermophoretic parameter will reduces the concentration and ANN model accurately captures these trends with efficient predictive fidelity.}, archive = {J_EAAI}, author = {K. Vinutha and J.K. Madhukesh and Nagaraj Patil and Amal Abdulrahman}, doi = {10.1016/j.engappai.2025.112041}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112041}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Numerical analysis of heat and mass transfer in off-centered stagnation point casson fluid flow over a rotating disc with thermophoretic particle deposition and artificial neural network-based optimization}, volume = {160}, year = {2025}, }
• (2025). Explainable artificial intelligence based microscopic peripheral blood cell image classification by exploiting quadradic convex optimization. EAAI, 160, 112040. (www)

  @article{ , abstract = {Deep learning models have shown remarkable efficacy in classifying microscopic peripheral blood cell images. However, for clinical applications, it is essential to provide transparency in their evaluation. To improve explainability, the Local Interpretable Model-agnostic Explanations (LIME) in Explainable Artificial Intelligence (XAI) methods effectively translate positive outcomes into practical usage. Benchmark data was collected from Kaggle and pre-processed using Min-Max normalization. Feature extraction was performed using a compound scaling method in neural networks, followed by classification using quadratic convex optimization. Our experimental results indicate that the proposed model achieves an impressive test accuracy of 99.678 % for classifying peripheral blood cell images. The correlation between computational learning and expert insights was observed, emphasizing the impact of human knowledge on enhancing computational outcomes. Explainable AI techniques will empower medical practitioners to identify morphological abnormalities in blood cells and prioritize critical findings, aiding in severity assessments and improving patient care.}, archive = {J_EAAI}, author = {Pradeepa Sampath and N. Sasikaladevi and Mukesh Prasanna and Saketh pallempati and Vimal Shanmuganathan and Seifidine Kadry}, doi = {10.1016/j.engappai.2025.112040}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112040}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable artificial intelligence based microscopic peripheral blood cell image classification by exploiting quadradic convex optimization}, volume = {160}, year = {2025}, }
• (2025). An adaptive reinforcement learning approach with trait-awareness for heterogeneous multi-robot cooperative pursuit. EAAI, 160, 112037. (www)

  @article{ , abstract = {Multi-robot cooperative pursuit is a critical task that has been extensively studied in domains such as military, security and rescue operations. However, most of the existing methods focus on homogeneous assumptions with limited research exploring heterogeneous teams. Moreover, they fail to generalize learned policies to new team compositions. In this paper, H eterogeneous team compositions with diverse capabilities are considered and we propose an A daptive T rait- A ware framework (HATA) to address these challenges. First, we propose an adaptive trait encoding module to capture team heterogeneity and produce expressive representations for heterogeneous robots. Second, a trait-aware behavior module is designed to generate different collaborative pursuit actions based on the distinct traits of heterogeneous robots. Consequently, composite teams can leverage the strengths of different traits to achieve efficient coordination. In addition, we introduce a novel training paradigm with a regularization term to enhance robustness to unseen heterogeneous configurations. Our method facilitates comprehensive planning of coalition cooperation without the need for domain knowledge or expert experience. Finally, we conducted experiments in a pursuit scenario with obstacles, and the final results show that our method achieves a 95.1% success rate while using only 35.2% of the maximum number of steps to complete the task. Additionally, we deploy the proposed method on a physical multi-drone platform to validate its real-world applicability.}, archive = {J_EAAI}, author = {Heteng Zhang and Yunjie Jia and Yong Song and Bao Pang and Xianfeng Yuan and Rui Song and Simon X. Yang}, doi = {10.1016/j.engappai.2025.112037}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112037}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive reinforcement learning approach with trait-awareness for heterogeneous multi-robot cooperative pursuit}, volume = {160}, year = {2025}, }
• (2025). Automated machine learning for residential property valuation. EAAI, 160, 112035. (www)

  @article{ , abstract = {Many studies have demonstrated that machine learning (ML) models surpass other approaches in residential property valuation. However, developing well-performing ML models requires the deep involvement of domain experts and data scientists, posing challenges for the scale of artificial intelligence (AI) implementation and application. Automated ML (AutoML) implements the end-to-end process of repetitive ML tasks without much human assistance. Although existing AutoML techniques have yielded promising results, no domain-specific AutoML framework is designed for residential property valuation. Most AutoML frameworks are domain-agnostic, overlooking domain knowledge, which limits their model performance and the practical applications of AI in real-world scenarios. This study proposes a domain-specific AutoML framework for residential property valuation (AutoML4RPV), which incorporates both domain-specific and domain-agnostic function modules to streamline AI implementation. Three real housing transaction datasets (New York, London, and Singapore) and a public valuation dataset are used for experimental validation by comparing AutoML4RPV with benchmark ML models and openly available AutoML frameworks. Integrating domain-specific modules, AutoML4RPV achieves the best performance with raw data input, outperforming the second-ranked domain-agnostic AutoML framework by 17.6 %, 18.2 %, and 9.80 % for the New York, London, and Singapore datasets, respectively. The domain-agnostic modules of AutoML4RPV are highly competitive and achieve the best model performance for the public valuation dataset. The data and codes of AutoML4RPV are openly available at https://github.com/ Linhkust/AutoML4RPV.}, archive = {J_EAAI}, author = {Lin Deng and Xueqing Zhang}, doi = {10.1016/j.engappai.2025.112035}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112035}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated machine learning for residential property valuation}, volume = {160}, year = {2025}, }
• (2025). Physics-informed neural networks for three-dimensional cerebrovascular hemodynamic prediction: A point cloud preprocessing strategy based on limited data. EAAI, 160, 112034. (www)

  @article{ , abstract = {Hemodynamic parameters are crucial for diagnosis and treatment of cerebrovascular diseases, yet real-time, high-resolution, and accurate acquisitions remain challenging due to limitations of current medical imaging and computational techniques. Artificial intelligence (AI)-based hemodynamic prediction also faces data scarcity in clinical settings due to ethical consideration. To address this issue, we aim to enhance learning performance under limited patient datasets with a tailored point cloud preprocessing strategy and a designed neural network architecture. The preprocessing strategy ensures point spatial homogeneity by applying resampling techniques including voxelization and distance-weighted interpolation, optimizing the data derived from computational fluid dynamics (CFD) for AI model training. The physics-informed neural networks (PINNs) module transits the network from a purely data-driven to a semi-data-driven framework, decreasing the data dependency. Four controlled trials were conducted using a limited CFD dataset of 51 patients, with 11 cases reserved for testing, to evaluate the performance of different model combinations, with and without the integration of a PINNs module and point cloud preprocessing. The combined approach showed superior performance in predicting spatially anisotropic hemodynamic fields—including velocity components and pressure—by efficiently mapping spatial coordinates to three-dimensional hemodynamic variables. This method achieved strong visual agreement with CFD simulations while reducing prediction time to 1 s. On the test set, it achieved normalized mean absolute errors of 7.79 ± 2.14 % for velocity and 6.63 ± 2.80 % for pressure, comparable to previously reported results based on large synthetic datasets. These results demonstrate the method's efficiency, accuracy, and real-time potential for clinical hemodynamic modeling in data-limited scenarios.}, archive = {J_EAAI}, author = {Jing Liao and Gaoyang Li and Keito Yanagisawa and Shin-Ichiro Sugiyama and Makoto Ohta and Hitomi Anzai}, doi = {10.1016/j.engappai.2025.112034}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112034}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed neural networks for three-dimensional cerebrovascular hemodynamic prediction: A point cloud preprocessing strategy based on limited data}, volume = {160}, year = {2025}, }
• (2025). Label-guided graph learning network via two-stage cross-modal fusion for multi-label skin disease diagnosis. EAAI, 160, 112025. (www)

  @article{ , abstract = {The accurate diagnosis of skin diseases relies on combining cortical lesion morphological features from clinical and dermoscopic data with deep subcutaneous lesion characteristics from ultrasound data. However, current multi-modal diagnostic methods mainly emphasize clinical and dermoscopic data analysis, lacking a thorough exploration of subcutaneous tissue features provided by skin ultrasound. Additionally, existing research often overlooks the analysis of relationships between skin labels and categories across different diagnostic tasks, constraining model performance as the number of target tasks grows. This paper proposes a label-guided graph learning network (LGL_Net) based on two-stage cross-modal fusion to achieve accurate diagnosis of skin diseases. Specifically, this paper first establishes a two-stage cross-modal fusion unit (TC_Fusion) to enable the fusion and transmission of morphological features from clinical data and deep lesion features from ultrasound data. Subsequently, a label-guided graph learning unit (LGL_Unit) is constructed to explore the correlations between multi-label data of skin lesions by building a graph convolutional network (GCN) at the label and category levels, thereby improving the accuracy of various skin disease diagnostic tasks. Extensive experiments were conducted on a public dataset (including clinical and dermoscopic data) and a private dataset (including clinical and ultrasound data). The experimental results demonstrate that the proposed method achieves optimal performance on tasks like pathological diagnosis (PG), benign or malignant diagnosis (BM), and 7-Point Checklist scoring (7 PC), offering a new approach for skin disease diagnosis. Our code is available at: https://github.com/Zhaocheng1/LGL-Net .}, archive = {J_EAAI}, author = {Cheng Zhao and Chunlun Xiao and Feifei Jin and Anqi Zhu and Zhuo Xiang and Yiyao Liu and Lehang Guo and Tianfu Wang and Baiying Lei}, doi = {10.1016/j.engappai.2025.112025}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112025}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Label-guided graph learning network via two-stage cross-modal fusion for multi-label skin disease diagnosis}, volume = {160}, year = {2025}, }
• (2025). Enhanced pediatric pneumonia auxiliary diagnosis: Integrating optical fiber vibration sensing with machine learning. EAAI, 160, 112024. (www)

  @article{ , abstract = {Childhood pneumonia remains a primary cause of death in children under five. Early detection is arduous due to its inconspicuous symptoms, and the existing radiological diagnostic techniques carry the risk of radiation - induced harm to young patients. To overcome these limitations, a groundbreaking study has proposed an innovative non - invasive diagnostic approach that integrates fiber optic vibration sensing technology with machine learning algorithms for pediatric pneumonia diagnosis. A novel fiber optic sensor is engineered to precisely capture respiratory vibration signals (RVS). These signals are then processed and analyzed using a Stacked - Grid Search Ensemble Learning Model (SGELM). In the experiment, respiratory vibration signals were gathered from 1649 pediatric patients aged between 3 and 14 who suffered from respiratory diseases. Through data balancing techniques, the dataset was expanded to 2184 samples. This dataset was partitioned into training, testing, and validation subsets. The developed system exhibited remarkable performance on the test dataset. It achieved high levels of accuracy, sensitivity, and specificity. Notably, it was also capable of classifying different pneumonia pathological types and statuses. This innovative method not only mitigates the radiation - related risks associated with traditional diagnostic methods but also holds great promise in revolutionizing the diagnosis of pediatric respiratory diseases. It could potentially improve the early - diagnosis rate and contribute to better treatment outcomes for children with pneumonia, thus playing a significant role in enhancing child health globally.}, archive = {J_EAAI}, author = {Pengfei Cao and Jiawei Xu and Yifan Zhao and Qian Ni and Yuxia Li and Hansen Chen and Ming Song and Jiqiang Shang and Mengqiang Yu and Xia Ding and Zhanhua Ma and Li Mao and Wenxia Tian and Xiaofeng Zhang and Mengyun Liang and Hao Wen and Jie Cao and Bin Hu}, doi = {10.1016/j.engappai.2025.112024}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112024}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced pediatric pneumonia auxiliary diagnosis: Integrating optical fiber vibration sensing with machine learning}, volume = {160}, year = {2025}, }
• (2025). Dynamic lane-changing trajectory planning for autonomous vehicles: A novel approach using gravitational search algorithm with blend crossover. EAAI, 160, 112022. (www)

  @article{ , abstract = {With the rapid developments in autonomous driving technology, the safe and efficient operation of autonomous vehicles is ensured using a highly potent critical component called dynamic lane-changing trajectory planning. Likewise, we present a novel approach using the Gravitational Search Algorithm with Blend Crossover (GSA-Blend) to tackle the challenges related to real-time trajectory optimization in dynamic and complex driving scenarios. Both the principles of gravitational search and blend crossover techniques are leveraged in the proposed GSA-Blend algorithm to dynamically plan lane-changing trajectories. Using GSA-Blend, the comprehensive cost function is optimized to enhance passenger comfort, safety, and efficiency, aiming to achieve a balanced trade-off among these key aspects of autonomous driving. Furthermore, a new blend crossover operator introduced, which has tailored lane changing trajectory planning to improve the algorithm's performance under complex scenarios. The effectiveness of GSA-Blend is validated through conducting extensive simulations in diverse driving scenarios, including hazardous, complex and simple conditions. The algorithms performance was evaluated against existing methods, considering lane-changing efficiency, safety, and comfort metrics. The obtained results have clearly shown that GSA-Blend consistently outperformed conventional methods in terms of its adaptability and robustness across various real-world scenarios. This research advances autonomous driving through introducing an innovative trajectory planning solution that enhances real-time optimization, ensuring safer and efficient lane changes. Ultimately, reliable and safer transportation systems can be achieved with the improvement of autonomous vehicle abilities using the promising potentials offered by the GSA-Blend algorithm.}, archive = {J_EAAI}, author = {M. Nithya and Vinu Sundararaj}, doi = {10.1016/j.engappai.2025.112022}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112022}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic lane-changing trajectory planning for autonomous vehicles: A novel approach using gravitational search algorithm with blend crossover}, volume = {160}, year = {2025}, }
• (2025). Utilizing modularity to construct a trading strategies network for enhancing the diversification of group trading strategy portfolio by grouping genetic algorithm. EAAI, 160, 112021. (www)

  @article{ , abstract = {Portfolio diversification is a perennial problem in finance. Identifying assets that offer uncorrelated returns is important for minimizing risk. To solve the problem, this paper proposes the Q-Group Trading Strategy Portfolio Optimization (QGTSPO) approach to increase the diversity of trading strategies for generating a QGTSP. QGTSPO extends the previous approach by modeling relationships between trading strategies using both correlation and mutual information, integrating a threshold component to determine optimal connectivity in the adjacency matrix. In other words, QGTSPO conceptualizes trading strategies as nodes in a graph and utilizes the grouping genetic algorithm to group similar nodes using modularity as a fitness measure. It first applies technical indicators and multiple financial datasets to generate candidate trading strategies with differentiated risk factors. The selected trading strategies are then used to generate chromosomes to initialize the population. Every possible QGTSP is encoded by the threshold, the take-profit and stop-loss, the grouping, and the weight parts. Next, the fitness value of a chromosome is evaluated by the Sharpe ratio, the modularity of the grouping structure, and the group balance. Extensive experiments on financial datasets from the New York Stock Exchange and Taiwan Stock Exchange demonstrate that QGTSPO achieves superior performance compared to the previous approach in terms of volatility reduction, crisis management, and consistency risk-adjusted returns, making it ideal in high-volatility markets and for investors with stringent risk management goals.}, archive = {J_EAAI}, author = {Kudakwashe Chideme and Chun-Hao Chen}, doi = {10.1016/j.engappai.2025.112021}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112021}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Utilizing modularity to construct a trading strategies network for enhancing the diversification of group trading strategy portfolio by grouping genetic algorithm}, volume = {160}, year = {2025}, }
• (2025). A heuristic with clustering-routing approach for the multi-depot multiple traveling salesman problems in different fields. EAAI, 160, 112013. (www)

  @article{ , abstract = {A heuristic approach involving clustering and routing operations is proposed for solving the multi-depot multiple traveling salesman problems(MDMTSPs). A modified form of K -means cluster operation is used to cluster the nodes under balance constraint. The optimal routes through the nodes in different clusters are searched separately. A set of perturbation rules on the routes are proposed for this purpose using a novel select and interchange operation on two different routes. Rules are selected randomly according to their performance using Roulette wheel selection process. K -opt operation is also used for the possible movement of the stagnant routes. The algorithm is named K-means clustering with multiple search strategies(KMCMSS). The performance of the KMCMSS is tested against a set of benchmark test instances from the traveling salesmen problem library. To solve the problem with cost matrix in fuzzy field, at first a fuzzy objective goal is set and then its credibility concerning the fuzzy objective is optimized using KMCMSS for a real-life decision. The efficiency of the algorithm is compared against a set of state-of-the-art algorithms for the MDMTSPs using statistical tools and test instances involving up to 783 nodes, 3-10 salesmen from it is library, and its superiority compared to others is strongly established. Moreover, the experimental results obtained by KMCMSS is significantly better than that of previous studies. KMCMSS provides better results as the closer nodes are entered in the same group and optimal schedule of each salesman is done separately, i.e., due to a significant reduction in the search volume of scheduling.}, archive = {J_EAAI}, author = {Prasanta Dutta and Indadul Khan and Krishnendu Basuli and Manas Kumar Maiti}, doi = {10.1016/j.engappai.2025.112013}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112013}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A heuristic with clustering-routing approach for the multi-depot multiple traveling salesman problems in different fields}, volume = {160}, year = {2025}, }
• (2025). A semantic-aware transfer learning framework via natural language processing to predict the cross-material bond strength of composite bars with limited data. EAAI, 160, 112011. (www)

  @article{ , abstract = {This study develops a novel transfer learning (TL) framework integrating natural language processing (NLP) and deep learning (DL) to predict the bond strength of fiber-reinforced polymer (FRP) composite bars in ultra-high-performance seawater sea-sand concrete (UHPSSC) using limited experimental data. The framework incorporates an NLP data processor to handle textual variables, a gated transformer to enhance parameter representations, and DL models for regression. Through leveraging the characteristics of seawater sea-sand concrete (SWSSC) and ultra-high-performance concrete (UHPC), it enables effective TL from large FRP–SWSSC/UHPC datasets to smaller FRP–UHPSSC datasets. Three pre-trained models were developed using FRP–SWSSC, FRP–UHPC, and combined datasets, respectively, and then fine-tuned on FRP–UHPSSC data to create three TL models. Compared with traditional machine learning (ML) models such as XGBoost and random forest (RF) trained directly on FRP–UHPSSC data, the proposed TL models achieved significantly higher predictive accuracy. Specifically, the optimal-performing TL model fine-tuned using combined FRP–SWSSC/UHPC data, achieved an R 2 of 0.914 and root mean square error (RMSE) of 0.75 on the test set, corresponding to a 13 % increase in R 2 and a 47.6 % reduction in RMSE relative to the optimal ML model. Furthermore, NLP-based textual representation improved the TL model's test R 2 from 0.876 to 0.914, highlighting the effectiveness in handling tabular data. The TL model also demonstrates semantic recognition capabilities, addressing variations in parameter expressions and dimensionality, enhancing robustness across scenarios. Overall, the proposed NLP-based TL framework effectively integrates heterogeneous material knowledge and substantially enhances predictive performance and generalization, offering a promising approach for data-limited engineering applications.}, archive = {J_EAAI}, author = {Pei-Fu Zhang and Daxu Zhang and Qi Zhao and Xuan Zhao and Yiliyaer Tuerxunmaimaiti}, doi = {10.1016/j.engappai.2025.112011}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112011}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A semantic-aware transfer learning framework via natural language processing to predict the cross-material bond strength of composite bars with limited data}, volume = {160}, year = {2025}, }
• (2025). Infant cry classification via deep learning based infant cry networks using discrete stockwell transform. EAAI, 160, 112008. (www)

  @article{ , abstract = {Background Accurate classification of infant cry signals is critical for understanding and addressing the underlying causes of distress in infants. However, the variability and complexity of cry signals present significant challenges, necessitating advanced methods for effective analysis. Issue Existing techniques for infant cry classification often lack the robustness to effectively handle the complex time-frequency features of cry signals, resulting in limited accuracy and reliability. This paper addresses the problem of automatic cry classification for infant cry analysis. Method The main goal is to automatically classify different types of infant cries from recorded cry signals. The proposed approach consists of three main stages: computation of Discrete Stockwell Transform (DST) to the infant cry signals, generation of DST based overlapped trajectory/contour images (DSTOIs)/DST based contour (DSTC) images, and design of deep learning-based Infant cry network variants ( IncryNet v1 , IncryNet v2, and IncryNet v3 ) for automatic classification. The DST based images provide time-frequency representation of the signals which are useful for distinguishing different categories of infant cries. These images are fed as input to all three deep learning network models, which could automatically extract significant features and classify into various cry types. The proposed models were evaluated on the Neo-Cry and Donate-a-Cry Corpus datasets. Results The experimental outcomes reveal that the application of the DST-based trajectory/contour images and IncryNet models enhanced the accuracy by an average of 10 % compared with other frequency/wavelet-based features and traditional machine learning models. Experiments were conducted and IncryNet v3 with DSTC attained the highest accuracy of 99.90 % and 99.96 %, with precision of 99.39 % and 98.99 %, recall of 98.50 % and 98.19 %, specificity of 99.50 % and 99.96 %, and F-Score of 98.40 % and 98.89 % using the donate-a- cry and neo-cry datasets respectively.}, archive = {J_EAAI}, author = {T. Jayasree and S. Blessy}, doi = {10.1016/j.engappai.2025.112008}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112008}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Infant cry classification via deep learning based infant cry networks using discrete stockwell transform}, volume = {160}, year = {2025}, }
• (2025). Scenario-based improved bayesian network model for construction safety assessment. EAAI, 160, 112006. (www)

  @article{ , abstract = {Assessing the safety of construction projects amidst potential accidents is crucial, especially with the rapid growth of modern infrastructure, which increases the risk of incidents. Existing safety assessment methods often fall short due to uncertainties and a lack of reliable analyses. To address this, a new method based on Scenario Theory (ST) and an improved Bayesian Network (BN) is proposed. This approach uses multi-source data to create a construction accident ontology and safety assessment paradigm. It refines the traditional BN algorithm to establish a construction accident BN and develops task and capability assessment functions. The method's effectiveness is validated through simulations and a real construction project in Tianjin, China. It improves safety by assessing engineering projects' emergency management capabilities in the face of accidents, benefiting construction firms and government agencies.}, archive = {J_EAAI}, author = {Zelong Lin and Dewei Kong and Wei Li and W.M. Edmund Loh and C.J. Wong and Zhijian Sun and Wei He}, doi = {10.1016/j.engappai.2025.112006}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112006}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Scenario-based improved bayesian network model for construction safety assessment}, volume = {160}, year = {2025}, }
• (2025). Using domain-specific keyword features to enhance deep learning-based pressure vessel inspection problem identification. EAAI, 160, 112005. (www)

  @article{ , abstract = {Pressure vessels are critical components in high-risk industries, such as petrochemicals, and require regular inspections to document problems. Accurate identification of these problems is critical. However, prevalent deep learning models often lack sufficient domain-specific knowledge, resulting in limited performance. Additionally, inspection records frequently contain unlabeled text, further complicating the identification process. To address these challenges, explicit domain-specific keyword features are leveraged to enhance deep learning models. A Keyword-based method employing both keywords and exclusion keywords is presented to assign preliminary labels and generate explicit domain-specific keyword features. Exclusion keywords filter out incorrect labels caused by negation or the sharing of common keywords. Two optional integration techniques are proposed to explore whether these explicit features can enhance representative deep learning models. One technique involves substituting a portion of the word embedding vector with keyword features, while the other technique involves fusing keyword features with hidden layer output in an attention module. These techniques are designed to provide clear and specific signals to the models, thereby enhancing their ability to identify domain-specific multiple labels. Evaluations conducted on real-world datasets demonstrate that the enhanced deep learning models significantly outperform their original counterparts on most evaluation metrics. The F1 scores of the optimal enhanced models improved to approximately 0.98 With the macro F1 score increasing by 16% to 0.979. These results highlight the effectiveness of explicit domain-specific keyword features in enhancing deep learning models. These findings contribute to knowledge-enhanced deep learning research and its application in domains requiring domain-specific semantic understanding.}, archive = {J_EAAI}, author = {Yuyuan Zhang and Xiaolan Wang and Feiyan Liu and Yibin Jin and Zechen Liu and Kun Shi}, doi = {10.1016/j.engappai.2025.112005}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112005}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Using domain-specific keyword features to enhance deep learning-based pressure vessel inspection problem identification}, volume = {160}, year = {2025}, }
• (2025). Artificial neural network model based on kolmogorov-arnold representation theorem and retention mechanism for real-time aircraft flight phases classification. EAAI, 160, 112004. (www)

  @article{ , abstract = {In modern aviation, accurate classification of flight phases is crucial for both operational safety and efficiency, as well as for reliable air traffic modeling and prediction. This paper introduces KARMA (Kolmogorov-Arnold Retention Memory Aware), a novel artificial neural network that uniquely integrates the Kolmogorov-Arnold representation theorem with a multi-scale retention mechanism for real-time flight phase classification. KARMA incorporates Kolmogorov-Arnold Network (KAN) layers for flexible function approximation, and employs convolutional feature extraction to enhance temporal data processing. The model also features a memory aggregation block that maintains context across sequential predictions. Benchmarking against state-of-the-art models—including Transformers, Retentive Networks, Extended Long-Short Term Memory, and State Space Models—demonstrates that KARMA achieves superior accuracy. The KARMA model achieved an accuracy of 0.93, a precision of 0.93 and a recall of 0.96. Additionally, the research introduces a custom, high-quality dataset collected with the developed device, enabling real-time, expert-labeled sensor data. KARMA's compact design ensures suitability for embedded avionics systems, setting a new standard for flight phase classification.}, archive = {J_EAAI}, author = {Paweł Tomiło and Jan Laskowski and Agnieszka Laskowska}, doi = {10.1016/j.engappai.2025.112004}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112004}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial neural network model based on kolmogorov-arnold representation theorem and retention mechanism for real-time aircraft flight phases classification}, volume = {160}, year = {2025}, }
• (2025). Integrating neural networks and spark for cloud security: A unified dataset approach to network intrusion detection. EAAI, 160, 112003. (www)

  @article{ , abstract = {The heterogeneous network traffic from Internet of Things (IoT) devices and sophisticated computing infrastructure have enabled attackers to employ cutting-edge technology to exploit the advanced cloud infrastructure. Hence, the focus of research has drifted toward developing high-performance Network Intrusion Detection Systems (NIDS) to ensure the integrity of cloud systems and defend against cyber-attacks. However, the unavailability of a universal dataset encapsulating multifarious attack scenarios, the absence of thorough validation techniques, and an inadequate discourse on the deployment architecture collectively present an impediment. The research offers a five-stage framework for constructing a high-performance cloud-based NIDS. The proposed framework involves preparing a unified dataset covering various attack scenarios and adapting multiple feature selection techniques. Further, the experiment performs a comprehensive validation by employing diverse deep learning techniques (Artificial Neural Networks, Long Short-Term Memory, and Auto Encoders) to assess the efficacy of each feature set derived from the feature selection techniques. Finally, the paper deploys the model that yields the highest accuracy onto the Spark cluster for real-time volumetric network traffic. The multi-dimensional feature selection drops the number of employed features by 77 %. Experiment results indicate a high accuracy of 99.34 percent for Artificial Neural Networks implemented on top of a Random Forest feature set for binary classification and 99.76 percent for Analysis of Variance (ANOVA) features for multi-class classification. Whereas Long Short Term Memory model showed an accuracy of 99.62 % and Auto Encoder has accuracy of 99.50 %. The article presents a thorough performance analysis of the Spark-based design on the Docker cluster at Amazon Web Services.}, archive = {J_EAAI}, author = {Jyoti Verma and Abhinav Bhandari and Manish Snehi and Pratham Snehi}, doi = {10.1016/j.engappai.2025.112003}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112003}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating neural networks and spark for cloud security: A unified dataset approach to network intrusion detection}, volume = {160}, year = {2025}, }
• (2025). Integrated dual-level dependency analysis framework for multi-task judicial decision prognosis in murder cases. EAAI, 160, 112002. (www)

  @article{ , abstract = {Judicial decision prognosis involves predicting judicial outcomes based on the case facts and evidence, a critical yet challenging task in the legal domain. Despite its importance, there is a scarcity of multi-task datasets and frameworks specifically tailored to the Indian judicial system. To address this gap, this work introduces the Indian Judicial Multi-task (IJMT) dataset, developed to resolve murder cases by incorporating facts, evidence, and exceptions. Furthermore, this work proposes a multi-task decision-making framework that integrates a lexical-vector dependency analysis engine designed using deep learning techniques and applies it to the IJMT dataset. To address the issue of label imbalance prevalent in legal datasets, a novel context-aware threshold modulation strategy is developed that enables controlled adjustment and dynamic fine-tuning of thresholds. The framework’s superior performance on the IJMT dataset compared to similar baselines leveraging dependency analysis between tasks validates its robustness and the dataset’s applicability for real-world legal experiments. By bridging the gap in multi-task datasets for the Indian legal context and presenting a robust decision-making framework, this work provides a significant contribution to advance research in legal judgment prognosis.}, archive = {J_EAAI}, author = {Prameela Madambakam and Himangshu Sarma}, doi = {10.1016/j.engappai.2025.112002}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112002}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrated dual-level dependency analysis framework for multi-task judicial decision prognosis in murder cases}, volume = {160}, year = {2025}, }
• (2025). A data-driven method approach for prediction of coal seam gas content combining feature selection and machine learning. EAAI, 160, 112000. (www)

  @article{ , abstract = {Coal seam gas content (CSGC) is a key factor for mine safety and methane resource utilization. To address these limitations, an integrated feature selection and machine learning approach was developed for improved gas content prediction. In this study, a total of 27 sets of experimental data covering geologic structure, coal quality characteristics, and coal occurrence conditions were collected from the left flank area of Taoyuan Coal Mine in Huaibei, China. Missing values in the dataset were handled using the chain interpolation method. Subsequently, the Genetic Algorithm with Ant Colony Optimization (GA-ACO) was introduced and compared with several traditional feature selection algorithms, including All Subset Regression (ASR), Forward Selection Regression (FSR), Backward Selection Regression (BSR), Random Forest Algorithm (RFA), and Lasso regression to obtain the main controlling factors of gas content. The Particle Swarm Optimization Support Vector Regression (PSO-SVR) to build a regression prediction model. Various regression models such as Multivariate Linear Regression (MLR), Multi-Layer Perceptron (MLP), Gradient Boosting Regression Tree (GBRT) and SVR were then used to construct regression prediction models. The dataset was randomly divided into training and validation sets in an 8:2 ratio, and prediction performance was evaluated using metrices. The results showed that the use of feature selection and machine learning effectively predicted the gas content of coal seams. The PSO-SVR model based on ASR achieved the best prediction performance with an R 2 of 0.99 on the training set and 0.978 on the test set, whit a maximum error of 0.077. Finally, the constructed model was used to predict regional gas content and a regional gas content distribution map was plotted, providing a visual representation for analyzing gas content distribution in coal seams. This study demonstrates the potential of combining feature selection with machine learning for regression prediction in coal bed geology, offering a feasible and promising approach for research in this field.}, archive = {J_EAAI}, author = {Sheng Su and Liang Wang and Songwei Wu and Yuechen Zhao and Chenghao Wang and Longyong Shu}, doi = {10.1016/j.engappai.2025.112000}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {112000}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A data-driven method approach for prediction of coal seam gas content combining feature selection and machine learning}, volume = {160}, year = {2025}, }
• (2025). An efficient and lightweight adaptive network for three-dimensional medical image segmentation. EAAI, 160, 111999. (www)

  @article{ , abstract = {The Vision Transformer (ViT) demonstrates exceptional performance in three-dimensional medical image segmentation by introducing a non-local self-attention mechanism, thereby providing a larger receptive field. However, the increase in model capacity and dataset scale, along with the lack of image-specific biases and scaling behaviors, affects the accuracy and detail of segmentation results, particularly when dealing with large-scale three-dimensional images that require capturing local information. In this study, we re-evaluate and improve the architecture of Convolutional Neural Networks (CNNs) by constructing a lightweight three-dimensional convolutional network (ConvNet++). We revisit volumetric convolution through the concept of large-kernel deep convolution modules and implement independent linear depth convolution scaling for each channel feature. This improvement not only contributes to the architecture of deep learning models in the field of artificial intelligence but also provides a more efficient solution for medical image segmentation. We design a feature fusion method to better utilize multi-scale spatial information, avoiding simple addition or concatenation, thereby alleviating issues related to excessive model parameters and ineffective feature fusion. Additionally, we introduce a Lightweight Hybrid Attention (LHA) module that employs a self-attention mechanism with a shared key-query scheme to efficiently encode spatial and channel information. The LHA module facilitates effective communication between spatial and channel branches, providing complementary features while reducing the overall number of parameters. This design represents an innovation in adaptive modules within the field of artificial intelligence, enabling better capture of subtle features in medical images. We conduct extensive evaluations on four challenging benchmark datasets, and the results indicate that our contributions are highly effective in terms of parameter efficiency and accuracy. In the Brain tumor segmentation benchmark test, our model achieves a Dice coefficient of 85.37%, with a model complexity of only 27.39 million parameters, highlighting its lightweight and efficient characteristics. A comprehensive comparison of the model’s inference time is conducted on the Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography dataset. The results indicate that 3D ConvNet++ performs relatively well in inference efficiency, with our model’s speed showing an approximately 3-fold improvement compared to 3D UX-NET and MedNeXt-B-K5. This achievement not only showcases the potential of artificial intelligence in the field of medical image segmentation but also provides an efficient solution for practical applications in medical engineering, assisting physicians in more accurately diagnosing diseases and planning treatments. Code: https://github.com/6018203135/ConvNet .}, archive = {J_EAAI}, author = {Dayu Tan and Manman Shi and Yansen Su and Xin Peng and Chunhou Zheng and Kaixun He and Weimin Zhong}, doi = {10.1016/j.engappai.2025.111999}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111999}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient and lightweight adaptive network for three-dimensional medical image segmentation}, volume = {160}, year = {2025}, }
• (2025). PromptDNet: A weakly supervised prompt framework for single image dehazing via dual-level depth cues. EAAI, 160, 111997. (www)

  @article{ , abstract = {Image dehazing is aimed at reconstructing high-fidelity visual information to support reliable spatial representations for downstream computer vision tasks. However, the scarcity of high-quality paired clear/hazy images in real-world scenarios constrains supervised dehazing approaches to heavily rely on large-scale synthetic datasets, inevitably leading to domain shift and suboptimal performance in practical applications. Furthermore, the distribution of haze exhibits strong spatial dependency, which is often undermodeled by existing methods owing to their reliance on purely data-driven, unimodal learning paradigms. To address these limitations, a novel unpaired weakly supervised framework, named the Dual-level Depth Prompt D ehazing Net work (PromptDNet), is introduced. A dual-level depth cue prompting strategy is introduced to enhance spatial understanding. At the image level, the depth map is jointly fed with the original hazy image to guide the model in capturing global features across varying depth regions. At the feature level, a meta-depth prompt attention network is developed to extract high-dimensional semantic representations from depth information, thereby injecting spatial context into the learning process and refining the network’s spatial reasoning and fine-grained feature discrimination. Furthermore, the dark channel prior is embedded into the framework to generate pseudo-paired samples, enabling the model to exploit physical priors under weak supervision. The dehazing performance is further enhanced through adversarial training and contrastive learning. Extensive experiments have demonstrated that PromptDNet performs well across synthetic and real-world benchmarks. The proposed approach offers an efficient, scalable solution for image dehazing, validated in diverse and challenging environments. The source code is available at https://github.com/JRw1998/PromptDNet .}, archive = {J_EAAI}, author = {Jiarui Wang and Shilong Wang and Yuting Pang and Jianlei Liu}, doi = {10.1016/j.engappai.2025.111997}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111997}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {PromptDNet: A weakly supervised prompt framework for single image dehazing via dual-level depth cues}, volume = {160}, year = {2025}, }
• (2025). Generating intersection pre-crash trajectories for autonomous driving safety testing using transformer time-series generative adversarial networks. EAAI, 160, 111995. (www)

  @article{ , abstract = {Ensuring the safety of autonomous vehicles in complex urban intersections remains a critical challenge, as intersections account for a significant share of urban crashes. To evaluate autonomous vehicles’ safety performance under these complex conditions, scenario-based testing has become essential. However, existing scenario generation methods often struggle to construct high-risk intersection scenarios with realistic dynamic interactions and long-term temporal dependencies needed for robust safety validation. To address this gap, this paper proposes a novel framework using Transformer Time-Series Generative Adversarial Networks to generate realistic pre-crash trajectories for autonomous vehicle safety testing at intersections. By learning from reconstructed real-world crashes in the China In-depth Mobility Safety Study-Traffic Accident dataset, the proposed generative model captures long-range dependencies and produces physically plausible, high-risk scenarios. These scenarios are integrated into counterfactual virtual simulations with the Baidu Apollo system, enabling systematic safety performance evaluation under challenging conditions. Experimental results show that over 90% of the generated scenarios pose crash risks and offer broader coverage and diversity than baseline cases. Unlike conventional approaches that remain at the logical scenario level, our method generates realistic, dynamic pre-crash trajectories. This better captures complex interactions and long-term dependencies, broadening risk exposure and improving the reliability of autonomous vehicle safety evaluation. As a result, it provides a scalable solution for testing under complex, high-risk intersection conditions.}, archive = {J_EAAI}, author = {Xichang Liu and Helai Huang and Jiang Bian and Rui Zhou and Zhiyuan Wei and Hanchu Zhou}, doi = {10.1016/j.engappai.2025.111995}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111995}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generating intersection pre-crash trajectories for autonomous driving safety testing using transformer time-series generative adversarial networks}, volume = {160}, year = {2025}, }
• (2025). MambaRSIS: Context-aware multi-scale feature aggregation with selective state space model for remote sensing instance segmentation. EAAI, 160, 111993. (www)

  @article{ , abstract = {Remote sensing instance segmentation aims to detect and assign pixel-level labels to each instance in remote sensing images, which holds critical engineering significance for both civil and military applications. While existing domain-specific methods have made progress, they still struggle with three persistent challenges: ineffective context modeling in cluttered backgrounds, information loss during multi-scale feature fusion, and blurred boundaries for densely clustered small objects. To address these limitations, we propose a novel remote sensing instance segmentation framework with three artificial intelligence (AI) methodological innovations, which comprises: a Context Perception Module (CPM) for context modeling, a Context Guided Multi-Scale Feature Aggregation (CGFA) method for multi-scale feature fusion, and a Multi-Path Region Proposal Extractor (MPRPE) with boundary-refined segmentation. The CPM leverages the selective state space model (Mamba) to capture long-range contextual information, effectively addressing the issue of cluttered backgrounds in remote sensing images. The CGFA replaces standard feature pyramid network architecture which is limited by direct summation or concatenation, preserving fine-grained spatial details with context guidance. The MPRPE and boundary-aware segmentation head mitigate the challenges of missed detection of small objects and blurred edge predictions, which arise from the clustered distribution of small objects and semantic ambiguity. Extensive experiments on the challenging iSAID and NWPU VHR-10 datasets validate the proposed method’s consistent improvements across metrics while demonstrating its practical engineering impact on remote sensing interpretation systems.}, archive = {J_EAAI}, author = {Liyuan Pan and Xu Cao and Huanxin Zou and Hao Chen and Shitian He and Yuqing Zhang and Xuanming Liu and Jiangshan Li and Wanyu Chen}, doi = {10.1016/j.engappai.2025.111993}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111993}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MambaRSIS: Context-aware multi-scale feature aggregation with selective state space model for remote sensing instance segmentation}, volume = {160}, year = {2025}, }
• (2025). Boosting industrial anomaly detection performance using generated artificial fault data. EAAI, 160, 111991. (www)

  @article{ , abstract = {Data-driven anomaly detection aims to learn a decision boundary, enveloping the normal region, and separating normal data from abnormal data. However, industrial data are fairly complex due to varying feedstock and unclear transfer processes and chemical reactions. This means the decision boundary will be very complex and even intractable. In addition, process variables are high-dimensional in modern industrial processes, which strengthens the difficulty of boundary extraction. Generally, the boundary should exactly exceed the outermost samples for precisely drawing normal regions. However, what we have in most situations is just normal data contaminated by unknown noises. Hence, conventional solutions that use statistical analysis to define a normal region result in a not-so-accurate decision boundary where missing alarms occur frequently. In addition to the conventional solution based entirely on historical data, i.e., passive fault detection (PAD), an alternative detection method, active fault detection (AAD), can circumvent the above problem by stimulating system performance through the intervention of auxiliary signal. While it results in disruption of the normal operation conditions for the process, its method to enhance output performance through additional signals inspires us. In this paper, we resort to the ability of deep neural networks to fit nonlinear data and perform dimension reduction. A fault data generation strategy is proposed and the artificially generated fault data are used to regulate the model training. The new virtual fault data aids in suppressing the decision boundary closest to the outermost periphery. We propose the principles of data generation and form a network structure, implementing information fusion of genuine normal samples and virtual fault samples. Two cases demonstrate the efficiency of the proposed method.}, archive = {J_EAAI}, author = {Kai Wang and Jiayi Zhang and Yishun Liu and Jie Han and Xiaofeng Yuan and Le Zhou}, doi = {10.1016/j.engappai.2025.111991}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111991}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Boosting industrial anomaly detection performance using generated artificial fault data}, volume = {160}, year = {2025}, }
• (2025). Toward accurate and efficient weather prediction using a dual-gated spatiotemporal attention network. EAAI, 160, 111990. (www)

  @article{ , abstract = {Accurate weather forecasting plays a vital role in safeguarding human activities, mitigating the risks of extreme climate events, and supporting environmental policy and disaster preparedness. However, existing data-driven approaches often struggle to effectively model the complex spatiotemporal dynamics and multivariate dependencies inherent in meteorological systems, limiting their reliability and scalability. To address these challenges, we propose a novel dual-gated spatiotemporal attention network (DSANet) for multivariate weather prediction. DSANet integrates a convolutional self-attention hybrid module to jointly capture local and global spatial features, and a dual-gated channel-time module to model temporal patterns and inter-variable relationships. A wavelet-guided composite loss function is introduced to enhance prediction accuracy in fluctuating and dynamic weather regions. Extensive experiments on both global and regional datasets demonstrate that DSANet outperforms baseline models in terms of accuracy, with a mean absolute error of 1.78 K in 3-day lead-time global temperature forecasting. In addition, DSANet exhibits strong generalization and fast inference, making it well-suited for real-time and off-site forecasting. By significantly improving the accuracy, efficiency, and transferability of multivariate weather forecasting, DSANet provides a scalable and effective tool for next-generation climate intelligence and decision-making support systems.}, archive = {J_EAAI}, author = {Zongwei Zhang and Lianlei Lin and Sheng Gao and Junkai Wang and Hanqing Zhao}, doi = {10.1016/j.engappai.2025.111990}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111990}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Toward accurate and efficient weather prediction using a dual-gated spatiotemporal attention network}, volume = {160}, year = {2025}, }
• (2025). Respiratory parameter estimation using pharyngeal phonetics and machine learning: Breaking free from spirometry. EAAI, 160, 111989. (www)

  @article{ , abstract = {The capacity to screen for respiratory diseases is vital for clinical diagnostics. Forced Expiratory Volume in 1 relevant second and Forced Vital Capacity are the two most common parameters or measures of respiratory health. Although respiratory health may be assessed using spirometry or other traditional forms of diagnostics, spirometry has variances in device availability, patient compliance during testing, and complexity of the testing procedure. This study found a novel, non-invasive means of estimating pulmonary function based on the voiced pharyngeal sound of "He" using analysis of the voiced sound. The study explored a model for estimating Forced Expiratory Volume in 1 s and Forced Vital Capacity values based on the outputs from traditional spirometry and features extracted from voice signals. There were a total of 21 features that were extracted from the voiced segments of the pharyngeal sound. All machine learning models of Forced Expiratory Volume in 1 s and Forced Vital Capacity used three machine learning algorithms. Data was collected from 18 male participants, aged 33–49 years old, from June 2022 to August 2022, which resulted in 56 recordings. Among the models that were evaluated in comparison to the linear, neural network, and quadratic models, the quadratic model performed the worst, while the neural network performed the best. The neural network model that used three features estimated Forced Expiratory Volume in 1 s with a mean error of 0.24 % while the two-feature neural network model estimated Forced Vital Capacity with a mean error of 0.58 %.}, archive = {J_EAAI}, author = {Ata Jahangir Moshayedi and Abolfazl Moradian Aghda and S. Ali Eftekhari and Mehran Emadi Andani and Soheil Salahshour}, doi = {10.1016/j.engappai.2025.111989}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111989}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Respiratory parameter estimation using pharyngeal phonetics and machine learning: Breaking free from spirometry}, volume = {160}, year = {2025}, }
• (2025). Visual-tactile fusion learning for material recognition based on channel switching and dual cross-attention. EAAI, 160, 111986. (www)

  @article{ , abstract = {Visual-tactile multimodal object recognition has attracted increasing attention, as information from different modalities can complement each other and enhance recognition performance. However, the inherent heterogeneity between vision and touch presents a key challenge for effective fusion, limiting accuracy and robustness. To address this, we propose a Multimodal Channel-Switching and Dual Cross-Attention Fusion (MCSDCF) method for integrating multisource data. The channel-switching module adaptively weights and exchanges information across modalities, enabling more discriminative and complementary feature representations. To further capture high-level semantic correlations and heterogeneous cues, we introduce a dual crossattention fusion structure that combines intra-modal self-attention with cross-modal mutual attention, reinforcing the quality of fused representations. Extensive experiments on three public benchmark datasets demonstrate the effectiveness of our MCSDCF framework, with multimodal fusion consistently outperforming unimodal baselines. Ablation studies further confirm the individual contributions of the proposed modules to the overall recognition performance.}, archive = {J_EAAI}, author = {Song Li and Wei Sun and QiaoKang Liang and Jian Sun and Hui Yang and YuDong Yang}, doi = {10.1016/j.engappai.2025.111986}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111986}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Visual-tactile fusion learning for material recognition based on channel switching and dual cross-attention}, volume = {160}, year = {2025}, }
• (2025). A comb concatenation diffusion model for hyperspectral image super-resolution. EAAI, 160, 111985. (www)

  @article{ , abstract = {One primary challenge of hyperspectral image super-resolution is addressing the high dimensionality and complexity of both the spectral and spatial domains. The diffusion model incorporates conditional knowledge and utilizes a step-by-step denoising method, thereby skillfully coping with the high dimensionality and complexity of hyperspectral images. However, a major difficulty encountered by diffusion-based hyperspectral image super-resolution techniques is the accurate use of conditional knowledge to guide and limit the denoising process. Additionally, the lack of sufficient conditional knowledge results in inadequate information for generating detailed high-resolution images. To address these issues, we develop a comb concatenation strategy specifically designed for diffusion models. This strategy strengthens the accuracy of conditional knowledge in guiding and constraining the noisy image through the comb concatenation of conditional knowledge and noisy images. Additionally, we propose a comb concatenation diffusion model for hyperspectral image super-resolution, consisting of three components. The first component is the comb concatenation block, which accurately utilizes conditional knowledge to guide the denoising process. The second component is the conditional encoding block, responsible for generating rich conditional knowledge. Finally, we develop a noise prediction block tailored for the hyperspectral image super-resolution diffusion framework. This block effectively manages accurate and comprehensive conditional knowledge representations. The three complementary components work together to maintain both spatial resolution and spectral fidelity. Extensive experimental results on the Houston, Chikusei, and Qingdao university of science and technology-1 datasets demonstrate that our framework outperforms state-of-the-art methods in both quantitative evaluations and visual quality across various scenarios. We release our source code at https://gitee.com/YinghaoXU/ISEDM for public evaluations.}, archive = {J_EAAI}, author = {Yinghao Xu and Hao Wang and Xin Sun and Qianlong Xie and Wenwen Zhang and Peng Ren and Fei Zhou and Susanto Rahardja}, doi = {10.1016/j.engappai.2025.111985}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111985}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A comb concatenation diffusion model for hyperspectral image super-resolution}, volume = {160}, year = {2025}, }
• (2025). Distribution-aware network with context and entity attention for scene graph generation. EAAI, 160, 111984. (www)

  @article{ , abstract = {Scene Graph Generation (SGG) aims to detect objects and infer their pairwise relationships in images, forming a structured semantic graph. Despite recent advances, existing methods struggle with effectively capturing rich contextual dependencies and suffer from biased relation prediction due to long-tail data distributions. To address these challenges, we propose a novel SGG framework, named DANCE, which integrates three key modules: the Context-Augmented Message Passing (CAMP) module, the Distribution-aware Dynamic Weighted Loss (DDW-Loss), and the Dual-Entity Attention Enhancement (DEAE) mechanism. Specifically, the CAMP module leverages a Gated Recurrent Unit (GRU) based architecture to reason over the scene graph, capturing entity interactions through sequential context encoding. To mitigate noise from random graph connections, we incorporate multi-head attention into GRU and fuse its output with initial visual features via residual connections. This design enhances context propagation while maintaining stability. Furthermore, the DDW-Loss function dynamically adjusts the loss weights of relation categories based on their frequency distribution, thus improving the learning of semantically meaningful but infrequent relations. Finally, the DEAE module employs parallel multi-head attention over subject and object features, enabling the model to extract fine-grained semantic dependencies and generate more discriminative relational embeddings. Experimental results on three popular datasets demonstrate that our method significantly improves the performance of scene graph generation tasks and outperforms existing methods. In addition, the proposed method exhibits strong adaptability and robustness in real-world scenarios characterized by complex scenes, noisy inputs, and varying image resolutions, highlighting its superior generalization capability for practical deployment.}, archive = {J_EAAI}, author = {Tongling Pan and Lulu Wang and Ruoyu Zhang and Zhengtao Yu and Yingna Li}, doi = {10.1016/j.engappai.2025.111984}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111984}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Distribution-aware network with context and entity attention for scene graph generation}, volume = {160}, year = {2025}, }
• (2025). Test-time data augmentation: Improving predictions of recurrent neural network models of composites. EAAI, 160, 111983. (www)

  @article{ , abstract = {Short fiber reinforced composites (SFRCs) are widely used in variety of industries for their high stiffness- and strength-to-weight ratios, yet accurate elasto-plastic modeling via high-fidelity simulations remains computationally costly. Recurrent Neural Networks (RNNs) have emerged as an interesting alternative to conventional material modeling approaches, particularly for non-linear path-dependent materials. This study aims to address and mitigate inaccuracies induced by neural networks in predicting path-dependent plastic deformations of SFRCs. In this paper, we have developed a Test-Time data Augmentation (TTA), which, to the best of the authors’ knowledge, is a method previously untested in the context of RNNs. The method is based on augmenting the input test data using random rotations and subsequently rotating back the predicted output signal. By aggregating the back-rotated predictions, a more accurate prediction compared to individual predictions is obtained. Using the method on a micro-mechanical full-field dataset, reduces mean relative error by 19% and yields a standard-deviation-based uncertainty measure correlated with an absolute error of r=0.73. TTA incurs only 1.85 s per sample which is more than 1600 times faster than full-field simulations. Our analysis also demonstrates improved shape consistency between the predictions and the target pseudo time-signals. The TTA approach requires no retraining and is reproducible with different randomly generated data augmentations, establishing a promising and robust framework for optimizing predictions of deep learning models. We also believe there are broader implications of the proposed method for other artificial neural network architectures (such as transformers) and various data-driven predictive modeling applications.}, archive = {J_EAAI}, author = {Petter Uvdal and Mohsen Mirkhalaf}, doi = {10.1016/j.engappai.2025.111983}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111983}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Test-time data augmentation: Improving predictions of recurrent neural network models of composites}, volume = {160}, year = {2025}, }
• (2025). Seismic data reconstruction via an adaptive feature fusion network. EAAI, 160, 111982. (www)

  @article{ , abstract = {Seismic data reconstruction is a crucial step in seismic data processing. Traditional methods and deep learning approaches have both been widely used in this field. However, they ignored the interactive learning of inter-channel information, especially in the case of high missing rate where feature extraction became more difficult. To address this issue, we propose an adaptive feature fusion network for the reconstruction of both random and consecutive missing seismic data. The information interaction block is designed into this model to improve the efficiency and adaptability of feature selection. It adaptively emphasizes important feature channels and enables inter-channel information exchange learning. To enhance the ability to capture global and local details, a cross-dimensional feature fusion module is designed at the bottleneck, integrating information from both the channel and spatial dimensions. Additionally, the strategy loss is designed to enable the network to learn the correlations among missing parts of the seismic traces, thereby boosting the reconstruction performance of our model. Compared with other state-of-the-art seismic data reconstruction methods, the proposed algorithm achieves improvements in both qualitative and quantitative evaluations: the reconstruction quality has improved by 20% on both synthetic and field datasets with random missing data. The reconstruction quality has improved by 30% on both synthetic and field datasets with consecutive missing data. At the end of the paper, we conducted ablation experiments, hyperparameter analysis and discussion.}, archive = {J_EAAI}, author = {Yuting Mu and Changpeng Wang and Xin Geng and Chunxia Zhang and Jiangshe Zhang and Junxiong Jia}, doi = {10.1016/j.engappai.2025.111982}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111982}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Seismic data reconstruction via an adaptive feature fusion network}, volume = {160}, year = {2025}, }
• (2025). Density peak clustering based on nearest neighbors. EAAI, 160, 111981. (www)

  @article{ , abstract = {As a promising clustering approach, the density peak clustering (DPC) algorithm is sensitive to density kernels and involved parameters, and its non-central data allocation method is likely to result in error propagation. In order to solve these two problems in dealing with real-world data, in this paper we present a new density peak clustering algorithm by making full use of the concept of k -nearest neighbors. In density estimation, we present a Gaussian kernel constrained by k -nearest neighbors and determine the cutoff distance based on local data distribution adaptively. In non-central data allocation, we propose a method to avoid error propagation by using the k -nearest neighbors of allocated data and unallocated data alternately. Starting from the cluster centers, we search the k -nearest neighbors of allocated data points, and include the qualified nearest neighbors into clusters. For each unallocated data point, we determine its belonging based on the allocated data points in its k -nearest neighbors. After that, we do a re-allocation of some allocated data points based on the allocated data points in their k -nearest neighbors, to improve the clustering accuracy further. In experiments on 50 synthetic and real datasets, our algorithm generates better results than 14 DPC-based and 6 non-DPC based algorithms. By making full use of k -nearest neighbors in all the major steps and exploiting the power of Gaussian distribution in real-world data processing, our algorithm is shown to be promising in dealing with real-world data clustering.}, archive = {J_EAAI}, author = {Houshen Lin and Jian Hou and Huaqiang Yuan}, doi = {10.1016/j.engappai.2025.111981}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111981}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Density peak clustering based on nearest neighbors}, volume = {160}, year = {2025}, }
• (2025). Adaptive adversarial pattern contrast algorithm for black-box model and domain attack. EAAI, 160, 111980. (www)

  @article{ , abstract = {The transferability of adversarial attacks in deep neural networks (DNNs) is a significant challenge, especially for achieving effective attacks across models and data domains. Unfortunately, existing attack approaches primarily focus on cross-model transferability, often overlooking the potential for black-box attacks across diverse data domains. This paper proposes the A daptive adversarial P att e rn C ontrast (APEC) algorithm, designed to achieve cross-model and domain adversarial attacks with high transferability. Firstly, APEC generates transferable adversarial examples by leveraging spatial characteristics such as regional homogeneity, repetition, and density, thereby increasing classifier misclassification rates. Secondly, a key innovation in APEC is the similarity contrast loss inspired by contrastive learning. It guides the model to learn discriminative adversarial features by aligning adversarial examples with adversarial patterns and distancing them from clean examples. Importantly, this optimization is performed label-free, enhancing APEC’s practicality in real-world black-box scenarios. Additionally, we introduce a Gaussian low-pass filter in APEC to generate adversarial perturbation patterns adaptively. This operation suppresses high-frequency information while preserving the low-frequency characteristics of natural examples, enhancing APEC’s attack capabilities. The APEC algorithm shows relative improvement across models and data domains compared to state-of-the-art transferability attacks. Our code is available at https://github.com/cs-igps/APEC-TransferAttack .}, archive = {J_EAAI}, author = {Weidong Wang and Yi Wang and Zhi Li and Long Zheng and Li Zhang}, doi = {10.1016/j.engappai.2025.111980}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111980}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive adversarial pattern contrast algorithm for black-box model and domain attack}, volume = {160}, year = {2025}, }
• (2025). Rethinking low-light knowledge for pedestrian detection in nighttime conditions. EAAI, 160, 111979. (www)

  @article{ , abstract = {In single-spectral nighttime pedestrian detection, existing approaches follow the convention of daytime pedestrian detection by employing models pre-trained on classification tasks as feature backbones. However, these pretraining backbones have not been exposed to low-light samples, lacking the specialized low-light knowledge necessary for effective performance in nighttime scenarios. In this paper, we present the Low-Light Knowledge Network (LLKNet). LLKNet comprises two primary components: (1) the Wavelet Multi-Level Extractor (WMLE), which enables the extraction of hierarchical features specifically tailored under unfavorable illumination, and (2) the Knowledge Scheduling Module (KSM) that ensures the adaptive integration of semantic hierarchical features in suboptimal lighting environments. Experiments demonstrate that LLKNet achieves state-of-the-art results on nighttime pedestrian detection datasets. Furthermore, LLKNet also exhibits strong performance for daytime pedestrian detection.}, archive = {J_EAAI}, author = {Ziyang Chen and He Yao and Wenting Li and Wei Long and Yongjun Zhang}, doi = {10.1016/j.engappai.2025.111979}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111979}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rethinking low-light knowledge for pedestrian detection in nighttime conditions}, volume = {160}, year = {2025}, }
• (2025). Incremental self-supervised learning based on transformer for anomaly detection and localization. EAAI, 160, 111978. (www)

  @article{ , abstract = {Anomaly detection and localization within image data pose significant challenges in the field of machine learning due to the unsupervised nature of the task, which requires the simultaneous identification of both global and local anomalies without labeled data. To address this challenge, patch-level self-supervised learning methods have traditionally been employed to detect both overall and local anomalies. However, these methods tend to prioritize information from patches, often neglecting critical image details. To overcome this limitation, this paper introduces a novel self-supervised method for anomaly detection and localization. The proposed method leverages the Vision Transformer as its core network and integrates a Vision Transformer-based Masked Autoencoder along with a unique pixel-level self-supervised learning approach. During training, we employ progressive data augmentation techniques, referred to as incremental self-supervised learning. We validate the effectiveness of our method through experiments on real-world industrial anomaly detection datasets, comparing its performance with that of relevant methods. Our code is available at https://github.com/xjspin/ISSTAD .}, archive = {J_EAAI}, author = {Wenping Jin and Fei Guo and Qi Wu and Li Zhu}, doi = {10.1016/j.engappai.2025.111978}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111978}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Incremental self-supervised learning based on transformer for anomaly detection and localization}, volume = {160}, year = {2025}, }
• (2025). Toward accurate detection of small objects in rail construction: A deep learning perspective. EAAI, 160, 111977. (www)

  @article{ , abstract = {Object detection is a fundamental task in computer vision that involves locating and recognizing all instances of objects within images or videos. Recent advances in deep learning have significantly enhanced object detection accuracy, particularly in complex environments such as rail construction. One of the pressing challenges in this domain is the accurate detection of small objects (e.g., clips, bolts, and nuts), which is critical for ensuring asset integrity, safety and the effective development of asset registers for maintenance. This paper addresses the following research question through a systematic review and meta-analysis of existing literature: Which deep learning algorithms are most effective in accurately detecting small objects (assets) during rail construction to support maintenance and management throughout a line's operational life? We examine the deep learning algorithms applied to object detection in rail infrastructure to answer this question and the key parameters and metrics used to benchmark their performance. We analyze the unique challenges of applying these architectures to rail construction environments. The conventional performance assessment of object detection models often provides an incomplete picture while relying on established metrics like accuracy, recall, precision, mean average precision, and intersection over union. Such evaluations tend to focus exclusively on final performance scores, overlooking the model's intrinsic complexity and the characteristics of the dataset it was trained on. Thus, a comprehensive analysis must extend beyond surface-level metrics to examine the model's architectural complexity and the dataset characteristics (i.e., quality, scale, and complexity). The contributions of this study are twofold as it: (1) provides new insights into the nature of deep learning architectures for small object detection in rail construction; and (2) identifies the limitations associated with applying small object detection frameworks to three-dimensional point cloud data are identified paving the way to make advances to accurately to detect small objects.}, archive = {J_EAAI}, author = {Hadi Mahamivanan and Jane Matthews and Peter E.D. Love and Farnad Nasirzadeh}, doi = {10.1016/j.engappai.2025.111977}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111977}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Toward accurate detection of small objects in rail construction: A deep learning perspective}, volume = {160}, year = {2025}, }
• (2025). Novel meta learning-artificial intelligence model for systematically detecting potential pan cancer-related targets. EAAI, 160, 111976. (www)

  @article{ , abstract = {Aims This study developed a novel artificial intelligence-driven computational framework to systematically detect potential PPOG (pan cancer-related targets). Methods Ribonucleic acid (RNA) transcriptome data and Deoxyribonucleic acid (DNA) methylation data of seven cancers were collected from the Cancer Genome Atlas database. A novel model “Few shot-Graph Convolutional Neural Network Model” (FSGCNM) integrating graph neural convolutional networks, meta learning, and newly developed algorithms was designed to systematically detect PPOG. The predictive performances of FSGCNM on cancer classifications with different numbers of support sets and query sets were compared, respectively. The performances of FSGCNM on systematically detecting PPOG were evaluated by comparing with Random Forest and Extreme Gradient Boosting. To further show the reliability of the model, differential expression analysis and Receiver Operating Characteristic Curve fitting analysis were further performed to demonstrate the pan cancer likelihoods of the top 10 PPOG. Findings Total of 84 potential PPOG was identified based on FSGCNM. Consequences The novel artificial intelligence method helps to systematically detect PPOG. The model and algorithm used in this study are the contribution in artificial intelligence, and the detection of pan cancer genes is the application in engineering.}, archive = {J_EAAI}, author = {Jiang Qi-yu and Zeng hui-yan}, doi = {10.1016/j.engappai.2025.111976}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111976}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Novel meta learning-artificial intelligence model for systematically detecting potential pan cancer-related targets}, volume = {160}, year = {2025}, }
• (2025). Federated learning-assisted intelligent yellow fever outspread prediction framework. EAAI, 160, 111975. (www)

  @article{ , abstract = {Deep learning has demonstrated strong potential in the detection of yellow fever. However, its effectiveness is often hindered by privacy concerns and limitations in data sharing across medical institutions. To address this challenge, this study proposes a novel Federated Learning-inspired framework for privacy-preserving, collaborative symptom analysis aimed at containing the spread of yellow fever. The proposed system integrates multimodal inputs specifically audio and visual data to detect key symptoms and utilizes a Dynamic Integration Approach for evaluating patient health severity. Comparative experiments were conducted against several state-of-the-art methods. Results show that the proposed model achieves superior performance, with an accuracy of 98.57%, precision of 99.85%, F-score of 96.53%, and recall of 94.72%. Additionally, the system exhibits enhanced delay latency (11.80 s) and operational stability (71%), indicating its robustness and real-time applicability in medical environments.}, archive = {J_EAAI}, author = {Munish Bhatia and Tariq Ahamad Ahanger and Abdulrahman Alabduljabbar and Abdullah Albanyan}, doi = {10.1016/j.engappai.2025.111975}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111975}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Federated learning-assisted intelligent yellow fever outspread prediction framework}, volume = {160}, year = {2025}, }
• (2025). An innovative framework for driving behavior analysis: Privacy-preserving, explainable, and adaptive artificial intelligence in cyber-physical systems. EAAI, 160, 111974. (www)

  @article{ , abstract = {Driving behavior analysis is pivotal for advancing intelligent transportation systems, a key domain of Cyber-Physical Systems (CPS). This study presents an innovative and cohesive framework that seamlessly integrates Personalized Federated Learning (PFL), Digital Twin (DT) technology, Crowdsensing, and Explainable Artificial Intelligence (XAI). By addressing critical challenges such as privacy protection, heterogeneous data management, and real-time adaptability, this framework enhances the synergy between Artificial Intelligence (AI) and CPS to deliver personalized, interpretable, and actionable feedback to drivers. Crowdsensing provides real-time driving and environmental data, which the Digital Twin uses to simulate various driving scenarios and generate localized datasets. These simulations support PFL by improving personalization and addressing data heterogeneity through dynamic clustering and Proximal Regularization ( μ ‖ w − w g ‖ 2 ), ensuring an effective balance between local and global models. A hybrid Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) model extracts spatiotemporal features to enhance prediction accuracy, while XAI, utilizing Local Interpretable Model-agnostic Explanations (LIME), provides transparent and actionable insights to users. Together, these components form a synergistic system that overcomes traditional limitations. Validated driving behavior data from Mendeley and Kaggle were used for evaluation, along with weather data collected from the open-meteo.com website. The results demonstrate that this framework outperforms existing approaches, achieving a 20 % improvement in prediction accuracy, a 15 % increase in F1-Score, a 50 % reduction in prediction time, and a 20 % decrease in communication costs. This comprehensive approach contributes to the evolution of trustworthy and ethical AI solutions within CPS, with applications in intelligent transportation and beyond.}, archive = {J_EAAI}, author = {Fatemeh Shabani and Rasool Esmaeilyfard and Alireza Nikseresht}, doi = {10.1016/j.engappai.2025.111974}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111974}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An innovative framework for driving behavior analysis: Privacy-preserving, explainable, and adaptive artificial intelligence in cyber-physical systems}, volume = {160}, year = {2025}, }
• (2025). A network intrusion detection system based on self-supervised learning of traffic differentiation in internet of things. EAAI, 160, 111973. (www)

  @article{ , abstract = {In light of the escalating threats within the Internet of Things (IoT) network environment, network intrusion detection has become a pivotal element of IoT network security, attracting substantial attention. Despite the remarkable performance of deep learning techniques in the field of intrusion detection, most current intrusion detection approaches are trained on labeled data. Due to the difficulty in obtaining a large amount of labeled data in real-world scenarios, many researchers have proposed methods utilizing unsupervised learning for network intrusion detection. However, there is still scope for enhancement in the efficacy of these methods. To achieve high accuracy detection with unlabeled data, we propose a novel intrusion and anomaly detection system based on Graph Neural Networks (GNNs), called SSTD, for Self-Supervised learning of Traffic Differentiation in IoT. In SSTD, we present a new training strategy accompanied by a cosine similarity-based loss function. This operation enhances the SSTD’s ability to discriminate abnormal traffic by constructing maximum differentiation negative examples and minimum differentiation positive examples to train the encoder. To demonstrate the effectiveness of SSTD, we adopt some common benchmark datasets for experimental validation. Compared to using raw features and other State-of-the-Art (SOTA) baselines, the detection performance of SSTD is significantly improved under different datasets and experiment scenarios.}, archive = {J_EAAI}, author = {Zijie Chen and Hailin Zou and Tao Hu and Xiaofen Fang and Jiexin Zheng and Jianqing Li and Yuanyuan Pan}, doi = {10.1016/j.engappai.2025.111973}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111973}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A network intrusion detection system based on self-supervised learning of traffic differentiation in internet of things}, volume = {160}, year = {2025}, }
• (2025). Quantum-enhanced deep lightweight object detection model for power tower base occlusion detection. EAAI, 160, 111972. (www)

  @article{ , abstract = {Power tower base is the support for transmission towers and transmission lines. However, the problems associated with tower base occlusion detection have not been studied. Therefore, this study integrates deep learning and quantum neural network techniques to propose a quantum-enhanced deep lightweight object detection model for power tower base occlusion detection. First, this study constructs a power tower base dataset (PTBD). Second, for the problem that the original backbone network cannot focus on features, this study proposes the cross-stage partial bottleneck with 2 convolutions for feature fusion bone, effectively extracting multi-scale features while maintaining light weight. Then, to address the problem of varying annotation quality of datasets, this study introduces the wise-intersection over union to assign gradient gains to anchor boxes with different qualities in the PTBD. Additionally, based on the knowledge of the features of the power tower base image and the auxiliary role of the priori box, this study removes the 20 × 20 detection head and improves the match between the priori bounding box and the real bounding box of the tower base by k-means++ clustering algorithm. Finally, the outputs of the proposed hybrid quantum neural network are utilized to correct the classification results through an averaging fusion strategy. The experimental results show that the mean average precision of the improved model is improved from 95.9 % to 97.3 %, and the detection speed is improved by 27 %–127.5 frames per second, while the memory size is 16.1 megabytes, which is only 1/4 of the original model.}, archive = {J_EAAI}, author = {Quan Lu and Yijin Lin and Linfei Yin}, doi = {10.1016/j.engappai.2025.111972}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111972}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Quantum-enhanced deep lightweight object detection model for power tower base occlusion detection}, volume = {160}, year = {2025}, }
• (2025). Multi-interest-aware adaptive self-attention network for session-based recommendation. EAAI, 160, 111971. (www)

  @article{ , abstract = {Session-based Recommendation (SBR) is to predict the next interactive item based on a group of anonymous users’ temporary behavior sequences. Despite the superior performance of existing methods for SBR, there are still several limitations: (i) Existing methods tend to predict the users’ newest interest but less consider users’ multiple interests, which is vital to obtain long-term preferences. (ii) Also, existing methods take all items in the sequences as wanted ones, ignoring the impact of noisy items, leading to suboptimal performance. To address these limitations, a Multi-interest-aware Adaptive Self-attention Network (MASN) for SBR is proposed, which explicitly considers users’ multiple interests while reducing misclicked items’ impact. Firstly, aiming at the diversity of user interests, multiple interest representations of users are learned from temporary sequences; Secondly, to reduce the impact of misclicked items, the importance score of each item is learned by an improved adaptive self-attention mechanism combined with the interest representation, which also reflects the current intention; Then, the overall preference of users is learned by Gated Recurrent Unit and soft-attention mechanisms; Finally, the intention and the preference are integrated to perform recommendation. The experimental results show that MASN outperforms nine representative baselines across four benchmark datasets.}, archive = {J_EAAI}, author = {Jian Feng and Haizheng Duan and Jiang Zheng}, doi = {10.1016/j.engappai.2025.111971}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111971}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-interest-aware adaptive self-attention network for session-based recommendation}, volume = {160}, year = {2025}, }
• (2025). Lightweight and high-performance object detector and tracking model for concrete surface defect detection. EAAI, 160, 111968. (www)

  @article{ , abstract = {The object detectors of the You Only Look Once (YOLO) series have been widely used in concrete surface defect detection. Achieving a balance among parameters, floating point operations (FLOPs), and performance is crucial for the detectors to realize real-time and efficient defect detection. While the basic YOLO framework guarantees real-time detection due to its high speed, maintaining high precision remains a challenge. This work designs a lightweight and high-performance damage detection network architecture based on the YOLO framework. A novel YOLO version 9 (YOLOv9) architecture with the Depthwise Convolution and the content guided mechanism (YOLOv9-DC) is developed by integrating the vanilla YOLOv9 with the improved submodules. The proposed method is validated both on custom and public concrete defect datasets. It achieves a mean Average Precision 50 (mAP 50 ) of 92.1 % and a mean Average Precision 50:95 (mAP 50:95 ) of 69.5 %, which is 0.9 % and 1.7 % higher than the prior state-of-the-art method, respectively. Additionally, it reduces parameters and FLOPs by 22.2 % and 27.3 %, respectively. For the requirements of practical engineering applications, the YOLOv9-DC-based trackers that combine the proposed model with state-of-the-art trackers are also developed to conduct multi-object tracking on videos with defect information collected on-site. Evaluation results demonstrate that the proposed method can not only detect incipient cracks but also accurately detect each crack when multiple cracks overlap, which meets the efficiency requirements of real-time multi-defect tracking.}, archive = {J_EAAI}, author = {Dandan Pang and Hanying Zhang and Yukang Cao and Yongqing Jiang}, doi = {10.1016/j.engappai.2025.111968}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111968}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight and high-performance object detector and tracking model for concrete surface defect detection}, volume = {160}, year = {2025}, }
• (2025). A prospect theory-based three-way consensus model for multicriteria and multi-agent conflict analysis. EAAI, 160, 111967. (www)

  @article{ , abstract = {The existing research on multicriteria and multi-agent (MC&MA) conflict analysis has gained increasing attention. However, it mainly focuses on inter-agent relationships and lacks resolution strategies. To this end, this paper proposes a three-way consensus model based on prospect theory (PT), aiming to provide feasible strategies for the MC&MA conflict analysis problem. Specifically, in Phase I , a PT-based three-way decision model (3WD) is introduced, which calculates the prospect value of each option by generalized interval-valued fuzzy rough sets (GIVFRSs). Secondly, a two-stage classification method is proposed to filter out a feasible option set with maximum prospect value. In Phase II , a consensus decision model based on a graph model is established from an optimization perspective, which revises target information by maximizing consensus degree and ensures that all agents reach consensus. Finally, a labor-management negotiation case study is used to demonstrate the practicality of the proposed model. Experimental analyzes with control variables show that the effects of parameter changes on the model are consistent with theoretical analysis, and the model is able to reach consensus in fewer iterations. Therefore, this study provides a theoretical foundation and practical guidelines for resolving the MC&MA conflict analysis problem, which has significant theoretical value and practical implications.}, archive = {J_EAAI}, author = {Jiajia Wang and Xiaonan Li}, doi = {10.1016/j.engappai.2025.111967}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111967}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A prospect theory-based three-way consensus model for multicriteria and multi-agent conflict analysis}, volume = {160}, year = {2025}, }
• (2025). Multi-objective optimization of thermomechanical-electrical performance in through-silicon via based three-dimensional chip using non-dominated sorting genetic algorithm-III. EAAI, 160, 111966. (www)

  @article{ , abstract = {Three-dimensional (3D) chips have emerged as a solution to the limitations posed by Moore's Law, offering enhanced integration density, reduced dimensions, and lower power consumption compared to conventional chips. These advantages have led to their widespread adoption in emerging fields such as artificial intelligence (AI), the Internet of Things (IoT), and autonomous vehicles. However, as application scenarios diversify and service environments become more demanding, there is an ongoing need to improve the comprehensive performance of 3D chips. To address these challenges, this study constructs a Through-Silicon Via (TSV) based 3D chip model with 25 links, 22 microstructures, 8 materials, and 39 design parameters. Drawing on multi-physics and electromagnetic transmission theories, three performance indexes are defined according to the structural characteristics of 3D chip, and then used for the establishment and error propagation analysis of the Gaussian-regression-based surrogate models. Subsequently, the non-dominated sorting genetic algorithm-III (NSGA-III) is employed to perform multi-objective optimization, generating a Pareto front for enhancing the thermomechanical-electrical performance of the chip. The optimal solution is then identified using the multi-factor comprehensive ranking method (MFCRM). Results demonstrate significant improvements over the initial design and relevant study, the optimization achieves a 13.50 % enhancement in thermal performance, a 12.00 % improvement in mechanical performance, and an 8.58 % increase in electrical performance. This work provides valuable insights for the AI-driven multi-objective design optimization of future high-performance electronic devices, offering a robust framework for balancing competing performance metrics in complex 3D chip systems.}, archive = {J_EAAI}, author = {Kaihong Hou and Zhengwei Fan and Yonggui Chen and Shufeng Zhang and Yashun Wang and Xun Chen}, doi = {10.1016/j.engappai.2025.111966}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111966}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective optimization of thermomechanical-electrical performance in through-silicon via based three-dimensional chip using non-dominated sorting genetic algorithm-III}, volume = {160}, year = {2025}, }
• (2025). Explainable data-driven formulation of chloride migration coefficient of eco-friendly concrete based on advanced automatic programming. EAAI, 160, 111965. (www)

  @article{ , abstract = {Data-driven approaches have demonstrated high efficiency in designing concrete materials with a large design space for desired material properties, but conventional data-driven approaches are inconvenient for practitioners and lack interpretability. This paper presents an explainable data-driven approach to formulate chloride migration coefficients of concrete using Artificial Bee Colony Expression Programming and Gene Expression Programming, aiming to overcome the limitations of black-box machine learning models while achieving high accuracy. The proposed approach integrates concrete domain knowledge with data-driven techniques including feature selection, anomaly detection, and hyperparameter selection. The approach, implemented for both conventional concrete and green concrete that contains fly ash, slag, and silica fume, achieves Pearson correlation coefficients exceeding 0.97 for both conventional concrete and green concrete on training and testing sets, and exhibits Scatter Indexes of 0.1 for conventional concrete and 0.2 for green concrete, demonstrating high accuracy while providing explicit mathematical formulas for engineer-friendly applications.}, archive = {J_EAAI}, author = {Seyed Amirhossein Moghaddas and Weina Meng and Yi Bao}, doi = {10.1016/j.engappai.2025.111965}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111965}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable data-driven formulation of chloride migration coefficient of eco-friendly concrete based on advanced automatic programming}, volume = {160}, year = {2025}, }
• (2025). Discriminative feature learning using class-aware and selective transfer adversarial network for partial cross-domain fault diagnosis of gearboxes. EAAI, 160, 111964. (www)

  @article{ , abstract = {Recently, partial domain adaptation (PDA) fault diagnosis methods have gained significant attention, as they assume that the target domain (TD) data class labels are only a subset of the source domain (SD) data class labels, making them more closely aligned with actual industrial scenarios. Most existing PDA methods adopt reweighting strategies to suppress the influence of outlier classes in the SD. However, these methods still face two major challenges: involving all SD samples in network training causes premature domain discrepancy, hindering the improvement of model transfer performance in the later stage of training; the relationship between target class samples is ignored, which makes the network classifier insufficient in extracting TD discriminable features. To solve these challenges, this study proposes a class-aware and selective transfer adversarial network (CSTAN) for partial cross-domain fault diagnosis of gearboxes. Firstly, the source data selector (SDS) in the CSTAN network is proposed to automatically filter out the source outlier data to avoid negative transfer; Secondly, the residual multi-channel attention mechanism feature extractor network is constructed to extract the features from the source data filtered by SDS and the target data, and the entropy-enhanced domain adversarial metric is adopted to promote positive transfer; Thirdly, to improve the model's ability to extract discriminable features of the TD, a target class-aware classification mechanism based on pseudo-label self-training technology is designed; Finally, the superiority and effectiveness of the CSTAN method are verified on the datasets of drivetrain diagnostics simulator test bench and wind turbine gearboxes.}, archive = {J_EAAI}, author = {Peng Zhu and Baoping Tang and Lei Deng and Qikang Li and Zihao Li}, doi = {10.1016/j.engappai.2025.111964}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111964}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Discriminative feature learning using class-aware and selective transfer adversarial network for partial cross-domain fault diagnosis of gearboxes}, volume = {160}, year = {2025}, }
• (2025). Multi-component controllable diversified augmentation of industrial images based on feature disentanglement. EAAI, 160, 111963. (www)

  @article{ , abstract = {Industrial defect segmentation based on deep learning heavily relies on diverse training data to achieve superior performance. However, it is challenging to collect diversified defects on production lines. Therefore, to improve the industrial image diversity from multiple feature dimensions, we propose a spatially-varying augmentation method based on feature disentanglement. Firstly, the Disentanglement AutoEncoder (DisAE) is proposed to extract multi-dimensional operable features of industrial images and generate new samples with accurate masks. The encoder disentangles images into spatially-variable defect features and normal background features, modeling the spatially local mapping between images and features. The decoder employs category labels and strength factors to modulate defect features, thus disentangling defect textures and structures. Additionally, adversarial loss and random sampling are introduced to enhance the generated quality and diversity. Secondly, based on the disentangled features from DisAE, we design four types of spatially-varying feature transformations, including affine transformations, random erasing, feature combination, and style transformation, to independently modify five main components of industrial images, achieving controllable diversification of normal background, defect location, shape, content, and strength. Experimental results show that our method can independently control and diversify five components of defect samples, generating diversified new samples with realistic textures and accurate mask annotations. Finally, our method obtains optimal segmentation performance on various industrial products and demonstrates high application value.}, archive = {J_EAAI}, author = {Jing Wei and Qingfeng Shi and Fei Shen and Zhengtao Zhang}, doi = {10.1016/j.engappai.2025.111963}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111963}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-component controllable diversified augmentation of industrial images based on feature disentanglement}, volume = {160}, year = {2025}, }
• (2025). Online nonparametric identification modeling of ship maneuvering motion based on PSO-optimized incremental gaussian mixture model. EAAI, 160, 111962. (www)

  @article{ , abstract = {Autonomous navigation of intelligent ships is highly dependent on real-time dynamic modeling technology to cope with the inherent nonlinear and time-varying characteristics of the marine environment. In this paper, an online identification method based on particle swarm optimization(PSO) to optimize the incremental Gaussian mixture model(IGMM) is proposed. This method solves the dilemma of manually tuning by deeply analyzing the significant impact of IGMM hyperparameters on model high fidelity and introducing the PSO algorithm to optimize the IGMM hyperparameters. Under the complex conditions of random steering tests and the introduction of wind and wave interference, the IGMM still shows high prediction high fidelity. To verify the performance of the proposed method, the IGMM optimized by PSO is compared with support vector regression(SVR), random forest regression(RFR) and noise input Gaussian process(NIGP) algorithms based on the free navigation test data of the SR108 container ship. The experimental results show that the IGMM optimized by PSO significantly reduces the root mean square error(RMSE) compared with the traditional algorithm in the free navigation test, and can dynamically adjust the number of model components to ensure robustness.}, archive = {J_EAAI}, author = {Xuechao Jiao and Junsheng Ren and Yan Hua and Qinghao Li}, doi = {10.1016/j.engappai.2025.111962}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111962}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Online nonparametric identification modeling of ship maneuvering motion based on PSO-optimized incremental gaussian mixture model}, volume = {160}, year = {2025}, }
• (2025). A transfer learning-based plate shape prediction model with limited samples for roller quenching process. EAAI, 160, 111960. (www)

  @article{ , abstract = {Plate shape is an important indicator in roller quenching process and significantly affects the using performance of steel plates. However, due to the high cost of plate shape detectors, although plate shape detectors play a vital role, many roller quenching production lines still operate without them. This issue severely limits the effectiveness of plate shape control methods and indirectly leads to a decline in plate shape quality. To address this gap and enable automatic feedback in scenarios lacking such detectors, this paper analyzes the limitations in both the accuracy and quantity of samples, and proposes a transfer learning-based plate shape prediction model. Firstly, a data augmentation module is proposed, and a deep sets module is designed for initial feature extraction. Then, a multilayer perceptron module is designed for deep feature extraction, aiming to enhance the ability of the model to capture nonlinear relationships. Next, an improved residual network module is designed to predict the plate shape. Through the special design of the model structure, the problem of different representation methods of plate shape is addressed. Finally, a targeted transfer learning strategy is designed for the proposed model to improve the accuracy in the absence of plate shape detectors. The experimental results demonstrate that the proposed model and strategy achieve high-accuracy plate shape prediction with limited samples, providing a practical plate shape prediction approach tailored for production lines without plate shape detectors.}, archive = {J_EAAI}, author = {Wen Zhang and Min Wu and Sheng Du and Luefeng Chen and Jie Hu and Naoyuki Kubota}, doi = {10.1016/j.engappai.2025.111960}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111960}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A transfer learning-based plate shape prediction model with limited samples for roller quenching process}, volume = {160}, year = {2025}, }
• (2025). Replacing complex transformer with simple attention to achieve hyperspectral and multispectral image fusion. EAAI, 160, 111959. (www)

  @article{ , abstract = {The fusion of hyperspectral (HSI) and multispectral (MSI) images to obtain high-resolution hyperspectral images is crucial for hyperspectral image processing and interpretation. In recent years, the Transformer architecture is extensively utilized in the domain of HSI-MSI fusion, yielding promising results. However, the introduction of various modified Transformer architectures leads to increasingly complex network structures, which impose significant computational resource demands. To address this issue, this paper proposes a Mask and Cross-Attention Network (MCANet) for HSI and MSI fusion. The network comprises three components: masked feature extraction, channel and spatial attention cross-feature fusion, and multi-scale step-by-step reconstruction. Our network abandons the existing patchwork of various advanced attention mechanisms and instead employs only the most straightforward channel spectrum and spatial attention mechanisms. This approach allows us to thoroughly extract spectral and spatial features while minimizing the computational resources required by the model. We conduct experiments on 8 HSI datasets and compare them with state-of-the-art fusion methods. The Pavia Center, Pavia University, Washington DC, Salinas, Houston and Botswana datasets all achieve the best fusion images and metrics. In addition, efficiency experiments confirm that the proposed method saves a significant amount of computational resources while fusing high-quality images. The source code and pre-trained models are available at https://github.com/xiaomudsg/MCANet .}, archive = {J_EAAI}, author = {Kunpeng Mu and Wenqing Wang and Mingze Gao and Han Liu}, doi = {10.1016/j.engappai.2025.111959}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111959}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Replacing complex transformer with simple attention to achieve hyperspectral and multispectral image fusion}, volume = {160}, year = {2025}, }
• (2025). An efficient multi-task forest fire and smoke detection model. EAAI, 160, 111958. (www)

  @article{ , abstract = {The complexity and dynamism of forest ecosystems pose significant challenges to early forest fire detection. Multi-task learning improves detection accuracy by employing a shared feature extraction network that enables tasks to interact and complement each other within a unified framework. In light of this, the paper presents a forest fire detection model employing multi-task learning, referred to as EMFFS-DET. Within the backbone network, a feature extraction module, C2f-KAN (Kolmogorov-Arnold Network), is incorporated to efficiently model complex nonlinear mappings with improved representation capacity. The Hard-Swish (H-Swish) function is employed as a substitute for the Sigmoid Linear Unit (SiLU) to improve performance in contexts involving complex or multi-target detection. A median-enhanced spatial and channel attention module (MECS) is introduced within the neck network, integrating median pooling with global average pooling and max pooling to enhance feature extraction in complex scenes. Furthermore, a content-aware feature upsampling module (CARAFE) is integrated to enhance the receptive field and optimize computational efficiency. In the head section, a segmentation head is added to assist detection by leveraging pixel-level semantics to refine object localization and boundary precision. The proposed model enhances the detection of small targets by introducing the Normalized Wasserstein Distance (NWD), which utilizes probability distributions to achieve precise localization. In comparison to You Only Look Once version 8 (YOLOv8), this approach enhances the Mean Average Precision (mAP) by 5.4 % and the Average Precision for Small objects (APS) by 7.8 %, thereby offering AI solutions for the early detection of forest fires.}, archive = {J_EAAI}, author = {Cong Chen and Yunfei Liu and Chenyu Zhang and Junhui Li and Xingliang Chen}, doi = {10.1016/j.engappai.2025.111958}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111958}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient multi-task forest fire and smoke detection model}, volume = {160}, year = {2025}, }
• (2025). A new paradigm for safe and accurate design of underground coal gasification coal pillars based on physics-informed neural networks. EAAI, 160, 111957. (www)

  @article{ , abstract = {Accurate design of isolated coal pillar width is critical to ensure the stability of underground coal gasification cavities, enhance resource recovery efficiency, and prevent engineering failures. However, conventional pillar design methods often rely on oversimplified approaches stemming from computational limitations, failing to account for the impact of high temperatures on surrounding rock behavior. This leads to inaccurate and overly conservative safety width estimations, severely hindering the large-scale application of underground coal gasification. To address this, this study pioneers the application of Physics-Informed Neural Networks (PINNs) in pillar safety design. A novel PINNs-integrated computational model was developed to precisely quantify the yield zone width of gasification pillars. This model overcomes the complexity of solving thermo-mechanical coupling problems induced by temperature-dependent rock property variations. Furthermore, a PINNs-based safety factor evaluation framework was established, revolutionizing the traditional design paradigm. Field validation conducted at the Ulanqab coal gasification site confirmed the model's reliability. The validated model achieved a reduction in pillar width from 24m to 18.4m while maintaining stability; concurrently, the resource recovery rate increased by 7 %. This research provides a theoretically rigorous and practically viable solution for optimizing pillar design and offering significant implications for the industrial advancement of underground coal gasification technology.}, archive = {J_EAAI}, author = {Chao Tang and Huaizhan Li and Guangli Guo and Jingchun Cao and Jianfeng Zha and Ruiheng Li}, doi = {10.1016/j.engappai.2025.111957}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111957}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new paradigm for safe and accurate design of underground coal gasification coal pillars based on physics-informed neural networks}, volume = {160}, year = {2025}, }
• (2025). Online defect detection method for resistance spot welding based on multi-source information fusion network. EAAI, 160, 111956. (www)

  @article{ , abstract = {Resistance Spot Welding (RSW), a crucial joining technique widely used in various industries, is especially significant in automotive sheet metal assembly. However, the erratic disruptions in fast-paced automotive production lines present challenges to weld spot quality detection and control. In this study, a multi-source information fusion network (MSIFN) that takes the energy and mechanical feature vector extracted based on expert knowledge as inputs is developed for online resistance spot welding (RSW) defect detection. The network comprises three parts: a single-source data high-dimensional feature extraction module based on attention multi-layer perceptron (MLP), a multi-source feature interaction module based on the mutual attention mechanism (MAM), and a feature mixing and enhancement module based on multi-head bilinear fusion. To address the limited and imbalanced nature of real-world data, Focal Loss is used as the training objective. The model classifies four types of welding outcomes: normal weld, cold weld, burn-through, and shrinkage. Compared with the models using only energy features or mechanical features, the proposed MSIFN achieved the highest average classification accuracy (96.3 %), demonstrating the complementarity of the two feature types and can enrich the characterization of welding quality. Meanwhile, compared with other commonly used artificial intelligence classification models, the MSIFN also demonstrated the highest average classification accuracy and better performance in terms of the false negative rate and false positive rate for each defect category. These experiments have verified the applicability of the proposed method with high precision and robustness in online RSW defect detection and can enhance the understanding of RSW process.}, archive = {J_EAAI}, author = {Yitong Fan and Haoyang Huang and Wei Dai and Yongjia Zheng and Ding Tang and Yinghong Peng}, doi = {10.1016/j.engappai.2025.111956}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111956}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Online defect detection method for resistance spot welding based on multi-source information fusion network}, volume = {160}, year = {2025}, }
• (2025). Reinforcement learning based on multi agent value distribution for beamforming optimization in cellular networks. EAAI, 160, 111955. (www)

  @article{ , abstract = {This paper considers multi-input single-output (MISO) homogeneous cellular networks, where multi-antenna base stations (BS) in each cellular network use shared spectrum to transmit data to single-antenna users, and employ downlink beamforming techniques to reduce inter-cell interference and improve network energy efficiency (EE). However, due to the high computational cost and related overhead of collecting global channel state information (CSI), most existing techniques are inefficient and impractical in dynamic wireless environments. In this study, we introduce a distributed reinforcement learning framework based on full-parametric quantile function (DRL-FQF) for downlink beamforming, which utilizes deep reinforcement learning (DRL) to dynamically select the optimal beam direction and power allocation according to the environment. DRL-FQF builds upon reinforcement learning algorithms by dynamically optimizing quantile positions and integrating quantile regression to model the distribution of cumulative returns in reinforcement learning. It employs Wasserstein error and quantile Huber loss for learning and self-adjustment, reducing bias in value distribution estimation, thereby replacing the traditional Q-value approach in reinforcement learning. Additionally, the high sample efficiency of DRL-FQF ensures its scalability even in high-dimensional scenarios with an increasing number of base stations, while prioritized experience replay further accelerates convergence. The proposed beamforming framework is efficient and scalable for high-dimensional scenarios with an increasing number of base stations, while greatly reducing information exchange. Simulation results show that the proposed DRL-FQF algorithm achieves higher performance compared to algorithms of existing methods deep Q network (DQN), deep recurrent Q-network (DRQN), and stochastic schemes.}, archive = {J_EAAI}, author = {Zhibin Liu and Jindong Wang}, doi = {10.1016/j.engappai.2025.111955}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111955}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reinforcement learning based on multi agent value distribution for beamforming optimization in cellular networks}, volume = {160}, year = {2025}, }
• (2025). Sharpness-aware multidomain imbalance generalization with external adversarial learning and intrinsic balanced entropy regularization for intelligent fault diagnosis. EAAI, 160, 111954. (www)

  @article{ , abstract = {In practical industrial scenarios, due to the diversity of operating conditions and the complexity of equipment operations, target condition data are often difficult to obtain. Moreover, mechanical faults occur sporadically and unpredictably, leading to complex class imbalance and domain imbalance phenomena in data across different conditions. The coupling of source domain data imbalance and target condition unknowability forms an imbalance domain generalization (IDG) problem, severely limiting the accuracy and industrial application of intelligent fault diagnosis models. To address this, this paper proposes a sharpness-aware multidomain imbalance generalization method, incorporating external adversarial learning and intrinsic balanced entropy regularization, to construct a class-unbiased and domain-invariant intelligent fault diagnosis model. Specifically, a novel spatiotemporal feature extraction module is designed to achieve an efficient fusion of multi-order spatial and multi-channel features, significantly enhancing the model's feature extraction capability. Subsequently, through the joint domain generalization with the balanced regularization approach, the model aligns the marginal and conditional probabilities across domains under data imbalance conditions, effectively mitigating the negative impact of data imbalance. Finally, the sharpness-aware minimization training strategy is employed to optimize the loss landscape during IDG, further improving the model's generalization performance. Experimental results on four typical IDG engineering scenarios and various imbalance degrees demonstrate that the proposed method achieves state-of-the-art performance, highlighting its significant advantages and application potential in addressing IDG engineering problems.}, archive = {J_EAAI}, author = {Shuaiqing Deng and Zihao Lei and Guangrui Wen and Yu Su and Zhifen Zhang and Houcheng Su and Xuefeng Chen and Chunsheng Yang}, doi = {10.1016/j.engappai.2025.111954}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111954}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sharpness-aware multidomain imbalance generalization with external adversarial learning and intrinsic balanced entropy regularization for intelligent fault diagnosis}, volume = {160}, year = {2025}, }
• (2025). Ensemble fractional fuzzy dispersion entropy: A low bias approach for data analysis. EAAI, 160, 111953. (www)

  @article{ , abstract = {Existing entropy methods exhibit strengths and limitations in capturing the irregularity and uncertainty of nonlinear systems. Dispersion entropy (DE), fuzzy dispersion entropy (FuDE), and ensemble dispersion entropy (EnsDE) are particularly vulnerable to parameters, noise, and outliers due to deficiencies in their mapping strategies and fuzzy classification, the inherent weaknesses of the Shannon entropy framework and suboptimal implementations lead to reduced accuracy in entropy estimation. Thus, we propose the ensemble fractional fuzzy dispersion entropy (EnsFrFuDE) and its two novel variants, E n s F r F u D E n p d f and E n s F r F u D E m e a n , designed to deliver more robust, accurate, and low-bias dispersion patterns for entropy estimation. The establishment of a generalized Gaussian distribution mapping, coupled with the meticulous refinement of five distinct mapping methodologies and the integration of three fuzzy functions, synergistically mitigates the sensitivity of entropy methods to noise and outliers. The fractional-order entropy framework, combined with n p d f process and overall m e a n , captures more nuanced hierarchical structures and multiscale features. Analysis of three groups of synthetic signals and real mechanical vibration datasets reveals that proposed methods exhibit a lower coefficient of variation for synthetic signals and greater variability across different fault patterns. Notably, our approach is effective not only for short signals and small sample sizes but also significantly outperforms 16 existing state-of-the-art entropy methods in noise robustness and estimation accuracy.}, archive = {J_EAAI}, author = {Chengjiang Zhou and Longkun He and Xuanyu Liao and Jie Li and Hao Song and Yang Yang}, doi = {10.1016/j.engappai.2025.111953}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111953}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ensemble fractional fuzzy dispersion entropy: A low bias approach for data analysis}, volume = {160}, year = {2025}, }
• (2025). A novel hybrid clustering and classification framework for multi-level depression detection in social media texts. EAAI, 160, 111952. (www)

  @article{ , abstract = {Depression is a prevalent psychiatric condition worldwide, with significant social and economic implications. Despite its high incidence, many individuals with depression remain undiagnosed and untreated. Meanwhile, people increasingly use social media platforms to express their emotions and thoughts. Consequently, leveraging these platforms for depression detection may help address several related challenges. This paper proposes a three-stage methodology, based on text mining techniques, to determine the severity of depression in individuals who post textual content on social media. In the proposed framework, each post is transformed into a vector of numerical features using established feature extraction methods. Principal Component Analysis is then applied to select the most informative features for identifying whether a post indicates depression via a classification algorithm. If depression is detected, the method clusters the relevant posts based on their characteristics, grouping similar texts together. A second classification model is then applied within each cluster to determine the level of depression. To equip the model with effective algorithms at each stage, the Taguchi method is used to identify the best combination of feature extraction, clustering, and classification techniques. Specifically, Bidirectional Encoder Representations from Transformers (BERT) is used for deep contextual feature extraction, Deep Embedded Clustering (DEC) is employed for clustering, and Support Vector Machine (SVM) is used for classification. Numerical results show that the proposed approach can accurately classify individuals’ posts into one of four depression levels: non-depressed, low, moderate, and severe. These findings suggest that social networks offer a platform for assessing mental health through textual analysis.}, archive = {J_EAAI}, author = {Parisa Khodabakhshi and Masoud Mahootchi and Hadi Mosadegh}, doi = {10.1016/j.engappai.2025.111952}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111952}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel hybrid clustering and classification framework for multi-level depression detection in social media texts}, volume = {160}, year = {2025}, }
• (2025). Improved discrete particle swarm optimization algorithm for solving fuzzy flexible job shop machines and automated guided vehicles fusion scheduling problem. EAAI, 160, 111951. (www)

  @article{ , abstract = {Automated Guided Vehicles (AGVs) are widely used in personalized and multi-batch flexible manufacturing job shops. However, uncertainties in processing and transportation time can arise due to various factors, such as changes in production processes, variations in personnel capabilities, and equipment maintenance requirements. To address this issue, this paper introduces the use of triangular fuzzy numbers to represent the uncertainty associated with processing and transportation time. Subsequently, a fuzzy flexible job shop scheduling model considering the AGV transportation process was established, with the goal of minimizing the maximum completion time. Additionally, an improved discrete particle swarm optimization (IDPSO) algorithm is designed to solve this model. Firstly, the classical particle swarm optimization algorithm is discretized to account for the discrete nature of the problem. Secondly, different particle updating strategies are employed to adjust the exploration and exploitation capabilities of the particles. Subsequently, four neighborhood search strategies are introduced to improve the performance of local search. Finally, the comparison experiment results suggest that the proposed IDPSO has competitive performance compared to other selected efficient algorithms. Moreover, a case study based on the actual situation of an automobile equipment manufacturing enterprise demonstrates that IDPSO is an effective method to solve the proposed problem.}, archive = {J_EAAI}, author = {Rui Wu and Zheng Tian and Xixing Li and Chenchen Wu and Hongtao Tang and Yibing Li}, doi = {10.1016/j.engappai.2025.111951}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111951}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improved discrete particle swarm optimization algorithm for solving fuzzy flexible job shop machines and automated guided vehicles fusion scheduling problem}, volume = {160}, year = {2025}, }
• (2025). A focusing-attention deformable convolution and transformer network with multi-scale contour-render for land cover classification in high-resolution remote-sensing images. EAAI, 160, 111949. (www)

  @article{ , abstract = {Land cover classification in High-resolution remote sensing image (HRRSI) aims to achieve pixel-level identification of ground objects on earth. However, the existing methods encounter difficulties in promoting accuracy due to vague contours of HRRSI. This paper proposes a semantic segmentation network called the Focusing-Attention Deformable Convolution and Transformer Network with Multi-Scale Contour-Render (MCFDTNet) to address two main causes of blurred or indistinct contours in HRRSIs. First, we design a Focusing-Attention Module (FAM) in the deformable convolution to enhance the representation of focused regions, mitigating blurring caused by out-of-focus regions and alleviating the foreground-background imbalance. Second, to fully utilize the semantic information in contours, we introduce a novel Double-Branch Multi-Scale Fusion Module (DMFM) to fuse local and global semantic features in the encoder using Multi-Scale Input (MI). We also propose a contour re-rendering mechanism, which dynamically controls the re-rendering region of blurred contour points (CR, ContourRend), aligning the input image edges with the edges in the label image. Comparative experiments were conducted with advanced semantic segmentation models on two datasets: the ISPRS Potsdam dataset and the Gaofen Image Dataset (GID). Ablation studies confirm the effectiveness of the proposed modules. Experimental results demonstrate that MCFDTNet outperforms existing models, achieving improvements of 0.88 % and 5.35 % in Mean Intersection over Union (MIoU) on the Potsdam and GID datasets, respectively.}, archive = {J_EAAI}, author = {Yiming Fan and Dongmei Zhang and Jiang Li and Jianghui Xiao}, doi = {10.1016/j.engappai.2025.111949}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111949}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A focusing-attention deformable convolution and transformer network with multi-scale contour-render for land cover classification in high-resolution remote-sensing images}, volume = {160}, year = {2025}, }
• (2025). Enhancing probabilistic photovoltaic power forecasting with parallel feature interaction and bayesian correction. EAAI, 160, 111946. (www)

  @article{ , abstract = {Photovoltaic power forecasting holds significant importance for solar energy grid integration and real-time dispatching in power systems. However, existing models struggle to extract sufficiently effective features and photovoltaic power is highly dependent on weather conditions, posing two major challenges in forecasting: the underutilization of features and the difficulty of forecasting under weather uncertainty. To address these challenges, this study proposes a parallel interactive deep evidential regression with Bayesian uncertainty correction model to enhance the accuracy of photovoltaic power forecasts. The proposed model incorporates a parallel feature interaction module that employs a bidirectional flow feature extraction, followed by an attention free mechanism, addressing feature underutilization in photovoltaic power forecasting. Additionally, the deep evidential regression module is introduced to capture the heavy-tailed characteristics of photovoltaic power by using the Student's t-distribution, enabling the estimation of both epistemic and aleatoric uncertainties. Since high weather uncertainty leads to high aleatoric uncertainty. Therefore, to mitigate weather uncertainty, a Bayesian weather uncertainty correction module is designed to adjust results exceeding a certain aleatoric uncertainty threshold with outputs from a Bayesian linear regression model trained on weather-excluded variables. The effectiveness of the proposed model is demonstrated through experiments conducted in Australia across four seasons, encompassing both deterministic forecasts and probabilistic forecasts. The performance of models is validated through comparisons with nine other models and the Diebold-Mariano test, confirming its efficacy in both deterministic and probabilistic photovoltaic power forecasting.}, archive = {J_EAAI}, author = {Yun Wang and Guang Wu and Fan Zhang and Runmin Zou and Jie Wan}, doi = {10.1016/j.engappai.2025.111946}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111946}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing probabilistic photovoltaic power forecasting with parallel feature interaction and bayesian correction}, volume = {160}, year = {2025}, }
• (2025). Uncertainty-guided bayesian active learning for cost-effective fault diagnosis with minimal labeled data. EAAI, 160, 111945. (www)

  @article{ , abstract = {Data-driven approaches have become the mainstream in machine fault diagnosis. If there is substantial amount of high-quality labeled samples, supervised learning methods can achieve commendable fault diagnosis results. Whereas, machines often operate in a normal state for extended periods, resulting in abundant healthy data but scarce fault data. Moreover, for specific large and complex mechanical systems, disassembly is challenging, and expert knowledge is required for fault diagnosis, leading to high labeling costs for samples. In such cases, developing robust fault diagnosis models using conventional data-driven supervised learning methods is challenging. To tackle this challenge, this paper proposes a probabilistic Bayesian active learning method for cost-effective fault diagnosis. This approach uses a probabilistic Bayesian neural network to assess the uncertainty of samples and selects those with the highest uncertainty for labeling, which are then incorporated into the training set to realize active learning. By actively selecting samples, the goal of improving classifier performance through minimal labeled data is achieved. Validation experiments have been performed on two datasets. The results demonstrate that the probabilistic Bayesian active learning approach can find important data samples for labeling, thereby minimizing the labeled data and reducing labeling costs. The comparison experiment demonstrates that the active learning effect of this approach surpasses other representative state-of-the-art methods. The research findings have significant value in developing advanced data-driven models, which can greatly reduce labeling costs and achieve cost-effective fault diagnosis.}, archive = {J_EAAI}, author = {Yuejian Chen and Liang Zhao and Te Han}, doi = {10.1016/j.engappai.2025.111945}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111945}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Uncertainty-guided bayesian active learning for cost-effective fault diagnosis with minimal labeled data}, volume = {160}, year = {2025}, }
• (2025). Cyclical weighting: A new training strategy for long-tailed recognition. EAAI, 160, 111944. (www)

  @article{ , abstract = {A common type of imbalanced data distribution in nature is the long tail distribution. Models trained on such imbalanced data tend to bias towards head classes with many samples. Some researchers have focused on instance-level or category-level difficulty reweighting for addressing data imbalance. However, these two kinds of methods did not handle the long-tailed problem well. The first method increases the model’s focus on tail class samples, but does not address the bias towards head classes. The second method, on the other hand, overlooks the impact of intra-class differences on the model. Therefore, we propose a new reweighting strategy named Instance-wise and Category-wise Difficulty Weighting (ICDW) that cyclically performs the two kinds of methods to address these issues. We design two core components specifically for ICDW: Category-wise Difficulty Momentum Weighting (CDMW) and Instance-wise Difficulty Momentum Weighting (IDMW), which respectively belong to the two kinds of methods. CDMW allocates weights to each class based on category-level difficulty, significantly improving the classification of tail data and addressing the model bias towards head classes. IDMW assigns weights to each sample based on instance-level difficulty, taking intra-class difference into account. Momentum Category Accuracy (MCA) and Momentum Instance Loss (MIL) are designed to measure the difficulty of classes and instances, respectively. Cyclical training strategy is introduced to balance the contributions of CDMW and IDMW in the objective loss. Through extensive experiments on four long-tailed datasets, we demonstrate the effectiveness of the proposed method, outperforming the state-of-the-art methods.}, archive = {J_EAAI}, author = {Qiong Chen and Sixun Feng and Tianlin Huang}, doi = {10.1016/j.engappai.2025.111944}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111944}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cyclical weighting: A new training strategy for long-tailed recognition}, volume = {160}, year = {2025}, }
• (2025). Adaptive nonlinear integral-backstepping control for frequency stabilization in cyber-physical shipboard microgrids using double deep Q-learning. EAAI, 160, 111943. (www)

  @article{ , abstract = {—The integration of diverse distributed energy resources is increasingly being enabled by shipboard microgrids (SbμGSs) as a promising solution to address the challenges posed by evolving maritime power systems toward decentralized and sustainable architectures. Open communication networks have been recently integrated with sustainable energy resources (SERs) for secure operation, regulation, and management of modern maritime power systems. However, deploying such emerging technologies introduces some faults and delays in communication channels which degrades the performance of marine power systems. This paper focuses on the development of a robust maritime controller-based nonlinear integral-backstepping (NI-BSC) for frequency stabilization of SbμGSs with high penetration of SERs, non-sensitive loads, and battery systems. Particularly, the double deep Q-learning (DDQL) is developed to adaptively design the NI-BSC and dynamically respond to the frequency stabilization challenges of SbμGS. By maximizing a long-term reinforcement signal, the neural networks (NNs) of DDQL (i.e., evaluation and target neural networks) are trained to reduce frequency fluctuations of SbμGS considering communication delays. The considered topology of SbμGS utilizes a separate maritime controller to regulate the solid-oxide fuel cell (SOFC) and thermostatic loads (heat pumps and freezer) while providing the possibility for sustainable units (e.g., wave energy and solar energy) to generate their maximum power. To conduct more realistic examinations, several typical scenarios of shipboard microgrids are carried out in a real-time setup. The real-time simulation outcomes reveal that the controller designed by DDQL outperforms the existing methodologies such as optimal controller, backstepping controller, and model-free sliding mode control, with an improvement in dynamic behaviors ranging from 54.57 % to 74.17 %.}, archive = {J_EAAI}, author = {Mai The Vu and Duc Hung Pham and Van-Truong Nguyen and Quang Thang Do and Abdullah K. Alanazi and Tuan Hai Nguyen}, doi = {10.1016/j.engappai.2025.111943}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111943}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive nonlinear integral-backstepping control for frequency stabilization in cyber-physical shipboard microgrids using double deep Q-learning}, volume = {160}, year = {2025}, }
• (2025). Adaptive weighted multiple imputation with generative adversarial networks for improving wind speed data integrity. EAAI, 160, 111942. (www)

  @article{ , abstract = {The integrity of wind speed data is crucial for energy planning and management. Generative Adversarial Imputation Network (GAIN) provides a new approach to address the significant challenge in the field of data science of accurately reconstructing missing data. However, the autocorrelation in time series often leads to reliance on time windows for data imputation, which tends to ignore global correlations within the data. To this end, this paper presents to perform high-dimensional mapping of the data and embed high-dimensional positional encoding, using a Euclidean norm penalty (L2 penalty) to constrain the generator's high-dimensional parameters, to improve the global correlations ability of the traditional GAIN model. This approach not only overcomes the limitations of time windows but also effectively captures the autocorrelation in time series, providing a global perspective for the imputation task. To enhance the imputation performance under high missing rate, an adaptive weighted multiple imputation mechanism is proposed, which assigns appropriate weights to each imputation result, thereby significantly reducing randomness and uncertainty. Experimental results, based on wind speed data with varying missing rates and distributions across four seasons, demonstrate that the proposed imputation framework achieves superior imputation accuracy and performs exceptionally well across all scenarios. Compared to the traditional GAIN model, our approach achieves average improvements of 44.97 %, 48.35 %, and 61.00 % under low, medium, and high missing rate conditions, respectively. The corresponding average Root Mean Square Error (RMSE) values are reduced to as low as 0.50, 0.58, and 0.75, highlighting the robustness and effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Weirui Jiang and Jinxing Che and Kun Hu and Yifan Xu and Wei Dong}, doi = {10.1016/j.engappai.2025.111942}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111942}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive weighted multiple imputation with generative adversarial networks for improving wind speed data integrity}, volume = {160}, year = {2025}, }
• (2025). Adaptive fault prototype network with joint learning transferability and discriminability for cross-domain bearing fault diagnosis. EAAI, 160, 111941. (www)

  @article{ , abstract = {Capturing the status of mechanical bearings in real time and promptly identifying latent faults are critical for the safety of mechanical equipment. However, the scarcity of typical fault data in special scenarios may lead to cases in which the states of fault samples are imperfect, and some fault categories may exhibit cross-domain issues. Although traditional algorithms can address these problems, existing models consider cross-domain transferability and fault discriminability independently, which results in insufficient fault feature learning in cross-domain fault diagnosis in the case of imperfect samples. Therefore, we propose an adaptive fault prototype network with joint learning transferability and discriminability (AFPN-JLTD) for bearing cross-domain fault diagnosis under imperfect samples. First, we develop a dual-branch feature extraction network by designing a sliced pyramid network and employing a transformer to extract fault features simultaneously. Next, we construct a similarity-learning discriminant model for the prototype network to learn fault prototypes with cross-domain transferability and fault discriminability. Finally, we devise transfer strategies based on soft and hard prototypes to facilitate accurate prototype transfer and construct the AFPN-JLTD model. This model mitigates the accumulation of prediction errors caused by using predicted labels during model training, thereby enabling cross-domain fault diagnosis under imperfect samples. Numerous experiments have confirmed the superior performance of AFPN-JLTD in diagnosing mechanical bearing faults, even when dealing with imperfect samples. In cross-domain fault diagnosis tasks without noise, our model consistently achieves the highest diagnostic accuracy (94.28 %) across all four datasets. Furthermore, our approach excels in cross-domain fault diagnosis under noisy conditions, delivering robust performance with minimal computational overhead.}, archive = {J_EAAI}, author = {Jian Wang and Wengang Ma and Yadong Zhang and Ruiqi Liu}, doi = {10.1016/j.engappai.2025.111941}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111941}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive fault prototype network with joint learning transferability and discriminability for cross-domain bearing fault diagnosis}, volume = {160}, year = {2025}, }
• (2025). Multifunctional health status assessment based on decentralized federated temporal domain adaptation for rotating machinery. EAAI, 160, 111940. (www)

  @article{ , abstract = {Based on the training with the large amounts of labeled data, intelligent data-driven approaches are able to assess the health status of rotating machinery. Traditional methods usually lack the multifunctionality required for health status assessment (HSA), resulting in incomplete assessment results. Furthermore, considering the challenge of machinery data silos, existing methods use collaborative model training solutions with multiple users, which place high demands on data privacy protection due to conflict of interests. To tackle this issue, a multifunctional HSA method based on decentralized federated temporal domain adaptation is proposed in this paper. First, a novel multifunctional HSA framework is designed for comprehensive assessment of rotating machinery, which performs synchronous health stage division and recognition, remaining useful life prediction, and reliability evaluation. Then, a temporal adversarial domain adaptation model is proposed to align temporal feature distributions through adversarial training between cross-domain encoders and a discriminator, both integrated with channel attention to enhance temporal feature extraction capability. Meanwhile, data privacy is ensured through decentralized federated learning involving multiple users. Experimental results show that the proposed method exhibits better generalization and stability, achieving an average error of 0.087 in dual prediction functions and an accuracy of 0.94 in recognition functionality.}, archive = {J_EAAI}, author = {Wei Zhou and Yong Zhang and Zuowei Ping and Cheng Cheng and Jiahua Sun}, doi = {10.1016/j.engappai.2025.111940}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111940}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multifunctional health status assessment based on decentralized federated temporal domain adaptation for rotating machinery}, volume = {160}, year = {2025}, }
• (2025). Automated detection of stages, zones, and plus diseases of retinopathy of prematurity using quantum convolutional networks in neonatal fundus images. EAAI, 160, 111938. (www)

  @article{ , abstract = {The proposed work develops Quantum Mobile Vision Transformer (QMViT) designed with the principles of quantum mechanics, vision transformer and pretrained Convolutional Neural Network (CNN) network to classify stages, zones and severity of Retinopathy of Prematurity (ROP). The proposed work aims to i) design and validate a light-weight quantum network QMViT to predict the characteristic stages, zones and plus disease of ROP; ii) classify the images into three multiclass classifications using state-of-the-art networks like Vision Transformer (ViT), Swin Transformer, Residual Vision Transformer (ResViT) and Mobile ViT and machine learning (ML) classifiers. The study utilises the HVDROPDB-BV (ROP Blood vessel) dataset in training the segmentation networks, and a real-time collected dataset with 2400 images (2000 ROP and 400 Normal) in training the classification models. The retinal blood vessels are segmented using transformer-based SwinUNet and MultiResUNet, from which 1000 Scale Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF), and Oriented Fast and Rotated Brief (ORB) features are extracted, fused, and dimensionally reduced to 25 components using Principal Component Analysis (PCA). Three ML classifiers classify 25 components into different ROP stages, zones, and severity. However, transformer-based networks such as ViT, Swin Transformer, ResViT, MobileViT, and Quantum-based QMViT classified retinal images directly into characteristic stages, zones, and Plus disease of ROP. The novel QMViT surpasses all other state-of-the-art deep learning and ML classifiers, with an accuracy of 95.5 %, 96.88 %, and 96.67 % in classifying stages, zones, and plus disease of ROP, respectively. QMViT is a hardware-efficient network that leverages quantum-inspired features for medical image classification.}, archive = {J_EAAI}, author = {Raja Sankari VM and Snekhalatha Umapathy and Ashok Chandrasekaran and Prabhu Baskaran and Varun Dhanraj}, doi = {10.1016/j.engappai.2025.111938}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111938}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated detection of stages, zones, and plus diseases of retinopathy of prematurity using quantum convolutional networks in neonatal fundus images}, volume = {160}, year = {2025}, }
• (2025). Machine learning-optimized stochastic voronoi lattices for enhanced mechanical performance. EAAI, 160, 111937. (www)

  @article{ , abstract = {Lattice structures, traditionally composed of periodic networks of interconnected struts, offer an excellent balance of high strength and low density. However, their periodicity limits adaptability to complex and unpredictable loading conditions. Stochastic Voronoi lattices, characterized by their irregular, non-periodic geometry, provide a promising alternative with enhanced energy absorption and mechanical robustness. In this study, we present a machine learning (ML)-driven framework integrating finite element analysis (FEA), a multilayer perceptron (MLP) neural network, and three-dimensional (3D) printing to optimize Voronoi lattice structures for targeted mechanical properties. To systematically control structural disorder, we introduce Relaxation Iteration (RI), an ordering parameter inspired by Lloyd’s algorithm. Based on RI, we show that there is a range of RI ≈ 1500 − 2000 which gives enhanced mechanical performance for Voronoi lattices. The ML-FEA optimized Voronoi lattices demonstrate double the stiffness and four times the toughness compared to conventional periodic lattices. These findings underscore the potential of ML-driven design strategies in developing tailored architected materials for applications requiring high energy absorption and structural integrity, including aerospace, automotive crash protection, and biomedical implants.}, archive = {J_EAAI}, author = {Michael Thompson and Hamidreza Yazdani Sarvestani and Ahmad Sohrabi-Kashani and Elham Kiyani and Enzo Filippi and Derek Aranguren van Egmond and Meysam Rahmat and Behnam Ashrafi and Mikko Karttunen}, doi = {10.1016/j.engappai.2025.111937}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111937}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-optimized stochastic voronoi lattices for enhanced mechanical performance}, volume = {160}, year = {2025}, }
• (2025). Edge-assisted framework for instant anomaly detection and cloud-based anomaly recognition in smart surveillance. EAAI, 160, 111936. (www)

  @article{ , abstract = {Anomaly detection in surveillance systems, identifying events such as fighting, shooting, or vandalism, remains challenging due to limited anomalous occurrences impacting model generalization and accuracy. Additionally, the significant variability in anomalies compared to normal actions complicates precise detection. Current state-of-the-art methods typically process all frames captured by surveillance cameras using resource-intensive, centralized systems, resulting in high computational costs and network bandwidth usage unsuitable for real-time, edge-based Artificial Intelligence of Things environments. Moreover, unclear anomaly definitions and computationally demanding real-time analyses hinder effective monitoring. To address these challenges, we propose an efficient Artificial Intelligence of Things-based anomaly detection framework employing a two-phase approach. Initially, a lightweight neural architecture search network classifies events as normal or anomalous directly on edge devices. When an anomaly is detected, the edge device alerts relevant authorities and transmits the relevant video frames to the cloud for further evaluation. In the second phase, a next-generation convolutional neural network extracts spatial features, refined by Spatial Attention Modules and further processed using a transformer-based multi-head attention network to capture contextual relationships. A Multi-Scale Feature module then classifies the anomalies into specific types. Extensive experiments conducted on the University of Central Florida Crime, Large-scale Anomaly Detection-2000, and Real-world Fighting-2000 datasets demonstrate that our method achieves accuracy of 53.60%, 80.02%, and 94.05%, with improvements of 1.8%, 1.2%, and 0.8%, respectively. Moreover, we report area under the curve scores of 0.76, 0.98, and 0.98 for all datasets, highlighting the method’s effectiveness and robustness in enhancing anomaly detection and recognition across various datasets.}, archive = {J_EAAI}, author = {Adnan Hussain and Noman Khan and Zulfiqar Ahmad Khan and Hikmat Yar and Sung Wook Baik}, doi = {10.1016/j.engappai.2025.111936}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111936}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Edge-assisted framework for instant anomaly detection and cloud-based anomaly recognition in smart surveillance}, volume = {160}, year = {2025}, }
• (2025). Predicting the ultimate strength of rectangular concrete-filled steel tube columns under eccentric loading using a knowledge-enhanced machine learning framework. EAAI, 160, 111935. (www)

  @article{ , abstract = {Rectangular concrete-filled steel tubes (RCFST) are widely used as structural columns in the field of structural engineering due to their exceptional mechanical properties. In practice, RCFST columns often experience eccentric loads, resulting in combined compression and bending. Although several empirical equations describe the mechanical behavior of CFST columns in terms of mechanical mechanism, their predictive ability is still limited by the necessary simplifying assumptions and application scope. This paper presents the development of a knowledge-enhanced machine learning (KeML) framework, with 15 input parameters, to address the limitations of conventional deterministic approaches and existing machine learning approaches, both suffering from low accuracy and robustness. To facilitate this, a comprehensive database consisting of 447 experimental data is established and divided into training and test sets, enabling the training and evaluation of the proposed KeML model. Comparative analyses against original machine learning models reveal that embedding knowledge can improve predictive accuracy and weak robustness, as well as reduce the required number of training data. Furthermore, a sensitivity analysis demonstrates that the cross-section size predominantly influences the ultimate strength of the RCFST columns subjected to eccentric loading. Finally, based on the sensitivity analysis and correlation analysis, a predictive equation for the ultimate strength of RCFST columns is developed using genetic programming.}, archive = {J_EAAI}, author = {Junbin Lou and Yixuan Li and Shicheng Zheng and Qian Feng and Guannan Wang and Rongqiao Xu and Xudong Qian}, doi = {10.1016/j.engappai.2025.111935}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111935}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predicting the ultimate strength of rectangular concrete-filled steel tube columns under eccentric loading using a knowledge-enhanced machine learning framework}, volume = {160}, year = {2025}, }
• (2025). Simplified syntax-guided domain-shared representation learning for cross-domain aspect-based sentiment analysis. EAAI, 160, 111934. (www)

  @article{ , abstract = {Cross-domain aspect-based sentiment analysis (ABSA) aims to identify aspect terms and their sentiment polarities in the target domain using annotated data from the source domain. Previous studies have shown that leveraging syntactic knowledge can partially bridge domain gaps. However, routine syntactic labels are diverse and fail to directly reflect the connection between aspects and opinions. This limitation hinders their ability to assist the model in perceiving aspects-sentiments in different domains. To overcome this challenge, we present a cross-domain ABSA approach based on simplified syntactic-guided representation learning. Initially, we design a dependency syntax simplification strategy that unifies syntactic labels of aspect terms and their corresponding opinion terms. We employ these simplified syntactic labels to guide the model's self-supervised learning process, thereby obtaining the domain-shared representations from unlabeled data in both the source and target domains. Using the domain-shared representation, we train the model on annotated data from the source domain and then apply it to make predictions in the target domain. Experimental results across ten transfer tasks using four public datasets demonstrate that our proposed method consistently outperforms other baseline models.}, archive = {J_EAAI}, author = {Jiqun Zhang and Yan Xiang}, doi = {10.1016/j.engappai.2025.111934}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111934}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Simplified syntax-guided domain-shared representation learning for cross-domain aspect-based sentiment analysis}, volume = {160}, year = {2025}, }
• (2025). Ego-centric multiple-correlation and temporal graph neural networks based residential load forecasting. EAAI, 160, 111933. (www)

  @article{ , abstract = {Accurate Residential Load Forecasting (RLF) is pivotal for the operation and decision-making in modern power systems. Recently, as the typical implementation and application of Artificial intelligence (AI) in energy field, Graph Neural Network (GNN) based RLF has emerged as a promising paradigm, since GNN can learn from graph-structured data, and capture complex interactions among nodes in a graph. However, it is challenging to build graphs that can effectively characterize the multiple unknown dependencies among residential users. To address the above issue, this paper proposes an effective residential load forecasting framework, based on intentionally constructed multiple ego-centric networks as well as multiple correlations and temporal graph neural networks. This work's contributions are given as follows. First, from two aspects: correlation and causality, multiple personalized ego-centric networks are intentionally constructed through data-mining manner, which respectively characterize the electricity consumption similarity between households, and direct influences on ego from the ego's neighbors (so-called alters) who essentially affect the ego's RLF. Second, multiple-correlation and temporal graph neural networks are adopted to forecast the ego's load. In detail, at each timestep, the ego node's hidden feature is embedded by multiple GNNs to represent multi-correlation dependencies between the ego and its alters, then the formed feature is sent to a recurrent neural network for further learning the spatial-temporal features. Finally, thorough experiments on real datasets demonstrate that our proposal outperforms the state-of-the-art spatial-temporal GNN-based forecasting schemes. Moreover, the empirical results verify that, for the load forecasting of single ego household, data-mining based personalized graphs can indeed significantly improve the forecasting accuracy, while the formulated personalized graphs are really sparsification and locality, which reflects the intuition that there are only relatively few useful relations in graphs based RLF. The source codes are available at https://github.com/tianxuHan/Residential-Load-Forecasting .}, archive = {J_EAAI}, author = {Yufeng Wang and Tianxu Han and Lingxiao Rui and Jianhua Ma and Qun jin}, doi = {10.1016/j.engappai.2025.111933}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111933}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ego-centric multiple-correlation and temporal graph neural networks based residential load forecasting}, volume = {160}, year = {2025}, }
• (2025). An improved genetic algorithm based deep learning model with you only look once framework for driver distraction detection. EAAI, 160, 111932. (www)

  @article{ , abstract = {The growing reliance on automobiles as the most common form of transportation and the increase in traffic has led to the need to implement essential safety measures. The driver’s alertness is critical to the safety of passengers in the vehicle. Existing methods for driver distraction detection face significant challenges as they rely on manually crafted features using traditional machine learning approaches, which are time-consuming to design, require domain expertise, and often fail to adapt to diverse real-world conditions. Although deep learning models have addressed some of these limitations by automatically extracting discriminative features, manually designed deep learning models struggle to handle complex driver distraction scenarios such as texting, yawning, and talking on the phone. The Deep Neural Network (DNN) models are well-known for their effectiveness across various computer vision tasks, including object localization and classification. However, manually designing efficient DNN models requires expertise. Even this may not always lead to an optimal model. To overcome this limitation, we propose the utilization of Neural Architecture Search (NAS) to design an automated method for generating DNN models. This work presents a framework for building a single-stage object detection model based on a NAS using an improved Genetic Algorithm (GA) for search strategy. The improved GA consists of Evaluation Correction based Selection (ECS) and Species Protection based Next Generation Population (SPNGP) to efficiently explore the search space and identify optimal backbone and training parameters. The designed search space includes the You Only Look Once (YOLO) backbone architecture parameters and associated training parameters. The experimental results suggest that the DNN model identified by the proposed approach has a smaller size and achieved better performance than the existing one-stage and two-stage object detection models, demonstrating the efficacy of our approach in designing driver distraction detection systems.}, archive = {J_EAAI}, author = {Medipelly Rampavan and Earnest Paul Ijjina}, doi = {10.1016/j.engappai.2025.111932}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111932}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An improved genetic algorithm based deep learning model with you only look once framework for driver distraction detection}, volume = {160}, year = {2025}, }
• (2025). Unsupervised wear detection for abrasive tools using audio features and dual-masked graph autoencoder. EAAI, 160, 111931. (www)

  @article{ , abstract = {Abrasive tool wear detection is a critical technology for advancing Industry 4.0, ensuring high-quality and efficient machining in industries such as aerospace and optoelectronics. However, developing a high-accuracy, non-contact detection method that performs reliably under complex operating conditions remains a significant challenge. To address this problem, this paper presents a novel unsupervised approach for abrasive tool wear detection, leveraging audio-cepstral features and a dual-masked graph autoencoder, marking the first application of graph-based structures in audio-driven tool wear detection. The proposed method models audio signals with a cross-graph structure, fully integrating both numerical and structural characteristics of mel-frequency cepstral coefficients. A dual-masked graph autoencoder is introduced to learn wear-specific features, enhancing robustness against by masking the nodes again in the decoding. Additionally, a novel loss function for wear detection is designed by combining mean squared error and cosine similarity with focal and weighting factors, enabling a comprehensive reconstruction of both value differences and directional relationships in non-Euclidean graph structures. Finally, a lightweight wear detection network uses the learned features for accurate classification of wear states. Experimental results demonstrate robust and accurate detection performance under eight complex operating conditions (including various operating conditions and noise types). The proposed method achieves an area under the curve of 0.9461 in mixed conditions, outperforming state-of-the-art methods by 1.48–21.59%. Ablation studies on backbone network, masking schemes, graph transformations, and loss functions further validate the effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Shuangjin Shi and Lili Tang and Hui Tian and Hui Huang and Ching-Chun Chang and Chin-Chen Chang}, doi = {10.1016/j.engappai.2025.111931}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111931}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised wear detection for abrasive tools using audio features and dual-masked graph autoencoder}, volume = {160}, year = {2025}, }
• (2025). Automated detection and quantification of leakage areas in shield tunnel linings using laser scanning data and deep learning network. EAAI, 160, 111930. (www)

  @article{ , abstract = {This study proposes an automated approach for identifying and quantifying water leakage in shield tunnels by integrating light detection and ranging (LiDAR) data obtained from laser scanning with the You Only Look Once (YOLO) deep learning network. The method transforms LiDAR data into greyscale images through a rasterisation-based process, followed by image labelling to train a YOLO model that detects and segments leakage areas on tunnel linings. The proposed approach combines two key technologies: (i) LiDAR data acquisition and processing, including a novel noise filtering technique, and (ii) leakage identification and segmentation using the YOLO model. LiDAR data processing involves data acquisition and pre-processing, segment unfolding, and noise reduction to enhance the computational visibility of leakage details. The leakage identification process involves dataset preparation, YOLO model training, leakage detection, and quantification. The model is validated using multiple shield tunnel sections, including those in the Guang-Fo intercity railway, Chunfeng tunnel, and Xuzhou metro line No. 4, among others. Results demonstrate outstanding model performance, achieving an accuracy of 92.1 % by comparing predicted leakage areas with manual observations. The LiDAR data acquisition, leakage identification, and quantification process for a tunnel section consumes only one-eighth of the time required for human inspection, thereby highlighting its efficiency.}, archive = {J_EAAI}, author = {Hao-Yuan Liang and Shui-Long Shen and Annan Zhou and Wei-Wei Zhao}, doi = {10.1016/j.engappai.2025.111930}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111930}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated detection and quantification of leakage areas in shield tunnel linings using laser scanning data and deep learning network}, volume = {160}, year = {2025}, }
• (2025). Mediating between filter and wrapper via probabilistic models, a hybrid feature selection framework for multi-label data. EAAI, 160, 111927. (www)

  @article{ , abstract = {Traditional hybrid feature selection methods rely on the collaboration between filter and wrapper techniques but often lack proper interaction mechanisms, limiting their ability to capture complex feature dependencies. Filter methods evaluate features individually and may overlook important combinations, whereas wrapper methods account for feature interactions but are computationally intensive and prone to overfitting. Without an Artificial Intelligence-powered probabilistic cohesive mechanism to align filter and wrapper components, these methods suffer from inconsistent decision-making, leading to local convergence and reduced effectiveness. In this paper, a novel three-component framework of filter-interface-wrapper is introduced, incorporating an interface layer between filter and wrapper. This layer leverages the inherent differences between objectives of filter and wrapper to enhance exploration capabilities and facilitate escape from local optima. It manages the transition procedure during collaboration between filter and wrapper by initially focusing on filter insights and gradually converging to the wrapper as it matures. The interface employs learnable Importance Probability Models (IPMs) that start with filter information and iteratively refine feature significance through population generation and mutation in the wrapper. These IPMs are updated after each iteration based on wrapper outputs to stay informed of its performance. By combining multiple IPMs with an evolutionary wrapper, the framework improves the exploration–exploitation balance in the solution space. Experiments on 15 multi-label datasets demonstrate significant improvements in feature selection, balancing efficiency and predictive power in complex scenarios.}, archive = {J_EAAI}, author = {Barzan Saeedpour and Fardin Akhlaghian Tab and Mohsen Ramezani and Eghbal Hosseini}, doi = {10.1016/j.engappai.2025.111927}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111927}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mediating between filter and wrapper via probabilistic models, a hybrid feature selection framework for multi-label data}, volume = {160}, year = {2025}, }
• (2025). Hierarchical spatial-temporal transformer with local aggregation for traffic demand prediction. EAAI, 160, 111926. (www)

  @article{ , abstract = {Accurate traffic demand prediction is essential for improving the safety and efficiency of intelligent transportation systems, and it has consequently received increasing attention in transportation research. Deep learning-based methods have been widely used in recent traffic demand prediction studies due to their flexibility and scalability. The exploratory analysis reveals that traffic demand data exhibit global temporal and spatial dependencies, while also demonstrating temporal locality and spatial locality. However, the existing methods have difficulty in capturing both global spatial-temporal dependencies and local similarities in traffic data, which limits the performance of these methods. To this end, this paper proposes a Hierarchical Spatial-Temporal Transformer with Local Aggregation for traffic demand prediction. The proposed method uses temporal and spatial multi-head attention to capture global spatial-temporal dependencies, while employs segment-wise spatial-temporal embedding and spatial local aggregation modules to incorporate temporal and spatial locality inductive biases. In addition, the proposed method adopts a hierarchical encoder-decoder architecture to utilize multi-scale feature representations for further enhancing its performance on the traffic demand prediction. Comprehensive experiments and ablation studies are conducted to evaluate the proposed method using three large-scale real-world traffic trip datasets. The evaluation results validate the effectiveness of the proposed method for short-term and long-term traffic demand prediction tasks. The comparative experiments show that our method outperforms the existing prediction methods.}, archive = {J_EAAI}, author = {Qianxi Zhao and Xiying Li}, doi = {10.1016/j.engappai.2025.111926}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111926}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical spatial-temporal transformer with local aggregation for traffic demand prediction}, volume = {160}, year = {2025}, }
• (2025). A lightweight segmentation model based on dilated multi-scale residual attention U-net for brain tumor segmentation. EAAI, 160, 111925. (www)

  @article{ , abstract = {To address the limited computational capacity of current clinical medical devices, which hampers the effective use of complex high-performance segmentation algorithms, this paper proposes a new lightweight brain tumor segmentation algorithm: Lightweight Dilated Multi-Scale Residual Attention U-Net (LDMRA U-Net). This model is based on the previously proposed Dilated Multi-Scale Residual Attention U-Net (DMRA U-Net). The overall network architecture incorporates the proposed Lightweight Channel Aggregation (LCA) mechanism to reduce the number of channels, along with a multi-level decoder aggregation strategy to minimize key information loss. Additionally, to enhance more direct information transfer between different levels of the encoder-decoder, the proposed Cross Enhanced Attention (CEA) module is incorporated into the skip connection part, improving segmentation performance. We evaluated our proposed method on the Brain Tumor Segmentation Challenge (BraTS) 2019 dataset, and the experimental results show that the number of parameters in LDMRA U-Net is only about 7.41 % of that in DMRA U-Net, with Dice Similarity Coefficients (DSC) of 0.8882, 0.8685, and 0.8622 in the Whole Tumor (WT), Tumor Core (TC), and Enhancing Tumor (ET) regions, respectively. LDMRA U-Net demonstrates significant performance improvements compared to existing lightweight methods. Our method helps to improve brain tumor segmentation accuracy with very small parameter sizes, and has the potential to be applied in a clinical setting.}, archive = {J_EAAI}, author = {Lihong Zhang and Yuzhuo Li and Yingbo Liang and Tong Liu and Wenwu Zhang and Junding Sun}, doi = {10.1016/j.engappai.2025.111925}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111925}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight segmentation model based on dilated multi-scale residual attention U-net for brain tumor segmentation}, volume = {160}, year = {2025}, }
• (2025). Early prediction of lithium-ion battery health in electric vehicles using small-sample learning and hybrid models. EAAI, 160, 111924. (www)

  @article{ , abstract = {Data collection during electric vehicle operation is costly, limited, and often of low quality, making early prediction of battery health a challenging task. This paper presents an early prediction method for lithium-ion battery health using small-sample learning. Features are categorized into four types, and a comprehensive correlation calculation method is introduced to address feature fluctuation. An innovative hybrid model, incorporating an attention mechanism and Bayesian optimization, is designed to enhance feature contribution and optimize hyperparameters. Experimental results show that with only 10% of the training data and 20 iterations of optimization, the method achieves an average prediction error of less than 3%, demonstrating high accuracy and efficiency in small-sample data learning and multi-feature selection.}, archive = {J_EAAI}, author = {Changsheng Zhao and Lei Yao and Yanqiu Xiao and Guangzhen Cui and Changhui Qu and Huilin Dai and Zhen Chen and Tiansi Wang and Zhigen Fei}, doi = {10.1016/j.engappai.2025.111924}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111924}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Early prediction of lithium-ion battery health in electric vehicles using small-sample learning and hybrid models}, volume = {160}, year = {2025}, }
• (2025). Multiple attribute group decision-making using p, q-quasirung orthopair fuzzy credibility sets: Application to data literacy evaluation of scientific researchers. EAAI, 160, 111923. (www)

  @article{ , abstract = {In multi-attribute group decision-making (MAGDM) problems, ensuring the trustworthiness of decision outcomes is essential, especially as real-world scenarios become increasingly complex and uncertain. Existing fuzzy set (FS) models, while effective in handling ambiguity, often neglect the credibility of membership and non-membership degrees, limiting their reliability. To address this issue, we propose an enhanced framework called the p, q-quasirung orthopair fuzzy credibility set (p, q-QOFCS), which incorporates membership degree fuzzy credibility numbers (MDFCNs) and non-membership degree fuzzy credibility numbers (NMDFCNs) into the conventional p, q-quasirung orthopair fuzzy sets (p, q-QOFS), thereby explicitly modeling the credibility of fuzzy information. This model enhances fuzzy decision-making with artificial intelligence (AI)-based credibility reasoning and supports more reliable decision outputs. We further develop novel distance measures and aggregation operators (q-QOFCWA and q-QOFCWG) to support robust information fusion, and employ the Criteria Importance Through Intercriteria Correlation (CRITIC) method for objective attribute weighting. To reflect decision-makers’ psychological behaviors under uncertainty, the cumulative prospect theory (CPT) and behavioral Technique for Order Preference by Similarity to Ideal Solution (BT) method are integrated into a new MAGDM model termed p, q-QOFC-CRITIC-CPT-BT. The proposed approach is validated through a case study on data literacy evaluation of scientific researchers (DLESR), demonstrating its effectiveness and robustness via parameter sensitivity analysis and comparison with existing methods. This study not only addresses the limitations of existing FS-based decision models by incorporating credibility measures, but also provides a versatile and reliable decision-making tool for complex group decision problems under behavioral and uncertain environments.}, archive = {J_EAAI}, author = {Baoquan Ning}, doi = {10.1016/j.engappai.2025.111923}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111923}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multiple attribute group decision-making using p, q-quasirung orthopair fuzzy credibility sets: Application to data literacy evaluation of scientific researchers}, volume = {160}, year = {2025}, }
• (2025). Joint class attention knowledge and self-knowledge for multi-teacher knowledge distillation. EAAI, 160, 111922. (www)

  @article{ , abstract = {Intelligent applications using large-scale deep neural networks face significant challenges due to their high storage and computational demands, hindering deployment on resource-limited edge devices. Knowledge distillation addresses this by transferring knowledge from an extensive teacher network to a smaller student network, thereby reducing computational costs while preserving performance. Multi-teacher Knowledge Distillation (MKD) further enhances this by allowing the student to learn from multiple teachers. However, MKD methods have two key limitations: (1) They typically use non-interpretable logits or features as knowledge, limiting the transparency of the learning process. (2) They focus primarily on teacher-guided learning, neglecting the potential of combining teacher supervision with self-learning. To address these limitations, this study presents a novel method, Joint Class attention knowledge and Self-knowledge for Multi-teacher Knowledge Distillation (JCS-MKD) , which combines both teacher supervision and self-learning. Our method introduces two key innovations: (1) A class attention mechanism that integrates class activation maps from multiple teachers to deliver more interpretable knowledge to the student. Additionally, an adaptive weighting scheme is employed to assign greater importance to teacher predictions that are closer to the ground truth, ensuring the student primarily learns from high-quality teacher knowledge. (2) A self-knowledge mechanism that decouples the student's logit into target and non-target components, customizing soft labels respectively to achieve adaptive self-supervision, enabling the student to refine their understanding independently. Experimental results on standard benchmark datasets demonstrate that JCS-MKD consistently outperforms state-of-the-art distillation methods across various teacher-student architectures. The code is available at: https://github.com/EifelTing/JCS-MKD .}, archive = {J_EAAI}, author = {Yifeng Ding and Gaoming Yang and Xinxin Ye and Xiujun Wang and Zhi Liu}, doi = {10.1016/j.engappai.2025.111922}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111922}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Joint class attention knowledge and self-knowledge for multi-teacher knowledge distillation}, volume = {160}, year = {2025}, }
• (2025). Enhancing driver assistance systems with field-programmable gate array based image filtering: Implementation of adaptive image processing framework for real-time filtering. EAAI, 160, 111921. (www)

  @article{ , abstract = {Advanced Driver Assistance Systems (ADAS) aim to enhance driving safety and comfort by automating various vehicle functions. These systems rely heavily on sensor data, particularly from cameras, which requires robust image enhancement to operate effectively under real-world conditions. However, real-time image enhancement in ADAS is challenging due to the need for high-speed processing, low power consumption, and resilience to visual degradations such as impulse noise, haze, and low-light environments. To address these challenges, an Adaptive Image Processing Framework (AIPF) is proposed which composed of three key modules. The Dynamic Thresholding and Adaptive Median Filtering Module (DTAFM) effectively removes impulse noise by adjusting thresholds based on local image statistics. The Context-Aware Haze and Light Enhancement Module (CA-HLEM) adapts contrast and brightness enhancement regionally based on haze density, contrast, and illumination. Finally, the Sobel Vector Directional Filter (SVDF) performs precise edge detection using multi-directional gradient analysis and lightweight Convolutional Neural Network (CNN) aids post-enhancement classification. The modular design of AIPF allows for real-time adaptability. The system is implemented using Field Programmable Gate Array (FPGA) architectures and optimized through non-pipelined, pipelined, and dual parallel-pipelined configurations to balance processing speed and power consumption. Performance evaluations show that AIPF can process 512 × 512 images in 12 milli seconds (ms) and 4096 × 4096 images in 45 ms, with latency ranging from 14 ms to 50 ms. Throughput is maintained at 16.67 images/sec under all operational loads. These results validate the effectiveness of AIPF, making it a robust and efficient solution for real-time ADAS applications.}, archive = {J_EAAI}, author = {A. Arul Edwin Raj and Nabihah Binti Ahmad and Jeffin Gracewell J. and R. Renugadevi and C.T. Kalaivani}, doi = {10.1016/j.engappai.2025.111921}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111921}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing driver assistance systems with field-programmable gate array based image filtering: Implementation of adaptive image processing framework for real-time filtering}, volume = {160}, year = {2025}, }
• (2025). Design of intelligent neuro-structures optimized with Levenberg–Marquardt and bayesian distribution for dynamical analysis of Caputo–Fabrizio fractional electric circuit models. EAAI, 160, 111920. (www)

  @article{ , abstract = {Electrical engineering models utilize interconnected circuits that consist of charged particles to enable the simulation and study of electric devices and systems, taking advantage of effective electron transfer across a completed circuit. Electric circuits involving the Caputo-Fabrizio (CF) fractional derivative have been precisely modeled recently by known solutions, efficiently capturing the system's response. The paper talks about the application of electrical circuit models for analyzing fractional stiff differential equations in an effort to explore different properties of the fractal Resistor-Capacitor (RC) and Resistor-Inductor (RL) circuits. The study employs artificial intelligence-based neurocomputing techniques and backpropagation networks for the purposes of increasing the knowledge on fractal circuit models. The Bayesian Regularization backpropagated neural networks (BR-BNNs) and Levenberg–Marquardt backpropagated neural networks (LM-BNNs) are utilized as efficient procedure for the training. The mathematical equations of CF-fractional RC and RL circuits were implemented to generate synthetic reference datasets and these information's were then used as a target for execution of LM-BNNs and BR-BNNs to find approximate solutions for the models. To validate and compare the accuracy of BR-BNNs and LM-BNNs in solving CF-fractional electric circuit models, the convergence curves on iterative adaptation of mean squared error (MSE) are employed. Results show that BR-BNNs yields MSE of approximately 10 −12 to 10 −13 and absolute error within the range of 10 −6 to 10 −8 , which providing strong evidence for the effectiveness of BR-BNNs approach than that of LM-BNNs algorithm. To further validate and endorse precision of the results, a performance evaluation via absolute error, statistical instance distribution in error histogram, regression analysis, convergence stability test, and Wilcoxon signed-rank test were exploited for CF-fractional electric circuit models that shows the statistical adequacy.}, archive = {J_EAAI}, author = {Aneela Kausar and Chuan-Yu Chang and Sidra Naz and Muhammad Asif Zahoor Raja and Rooh Ullah Khan and Muhammad Safiullah and Saeeda Naz}, doi = {10.1016/j.engappai.2025.111920}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111920}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Design of intelligent neuro-structures optimized with Levenberg–Marquardt and bayesian distribution for dynamical analysis of Caputo–Fabrizio fractional electric circuit models}, volume = {160}, year = {2025}, }
• (2025). Multi-modal self-supervised contrastive representation learning for three-dimensional point cloud understanding. EAAI, 160, 111919. (www)

  @article{ , abstract = {Three-dimensional point cloud understanding is widely used in autonomous driving and has become a hot research topic. However, labeling large scale point cloud data is challenging, and single modal unlabeled data often provides limited information, which severely limits its development. Multi-modal self-supervised learning can learn relevant features from unlabeled data and incorporate more modal information to provide rich semantic information for point cloud data, which shows great potential but is difficult to effectively correct the semantic bias between modalities. In order to address above issues, we propose a multi-modal self-supervised contrastive learning model (named MSCLM) for three-dimensional point cloud understanding, which aims to enhance the understanding of three-dimensional point cloud by jointly learning from multiple modalities of unlabeled data. MSCLM trains on a large amount of point cloud–image–text triplet data and learns better feature representations for point cloud in a self-supervised manner from positive and negative samples. To compensate for semantic biases caused by multiple modalities with different granularity, we design a global point cloud feature fusion mechanism, which reduces the negative impact of biases from pre-trained networks on each modality by integrating global point cloud features with other modal features. To fully integrate the learning information from different modalities, we adopt intra-modal enhanced contrastive learning and cross-modal joint contrastive learning to further strengthen multi-modal information sharing. The experimental results demonstrate the effectiveness of the proposed algorithm and promote the development of autonomous driving.}, archive = {J_EAAI}, author = {Weichao Ding and Zehao Yang and Fei Luo and Chunhua Gu and Wenbo Dong}, doi = {10.1016/j.engappai.2025.111919}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111919}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modal self-supervised contrastive representation learning for three-dimensional point cloud understanding}, volume = {160}, year = {2025}, }
• (2025). Mitigating batch normalization bias for single domain generalizable person re-identification. EAAI, 160, 111918. (www)

  @article{ , abstract = {Domain generalizable person re-identification (ReID) poses a significant challenge in real scenarios, as it aims at transferring the knowledge learned from single- or multi-source domains to unseen target domains. Due to the deterministic values of feature statistics (mean and standard deviation) learned from source-domain data, Batch Normalization (BN) exhibits a severe bias towards source domain. This bias is particularly pronounced in single-source domain scenarios, often leading to catastrophic performance degradation. In this paper, a novel Debiasing Batch Normalization (DBN) approach is proposed to alleviate the source-domain bias caused by deterministic values. The DBN is composed of two novel components, a Dynamical Transformed Module (DTM) component and a Gallery to Query Test-time Adaptation (G2QTA) component. Specifically, DTM can flexibly generate the feature statistics relying on the input samples, which is beneficial to adapting to unseen domain and addressing the domain shift. Note that DTM is only embedded in high level stages to replace several BN modules. A novel G2QTA is designed to revise deterministic values, which are not modified by DTM, and adapt to the style of unseen target domain. Extensive experiments demonstrate that our DBN outperform most methods on the task ‘Market1501 → DukeMTMC-ReID’ and ‘DukeMTMC-ReID → Market1501’.}, archive = {J_EAAI}, author = {Luyifu Chen and Yanfeng Li and Jia Sun and Houjin Chen and Minjun Wang and Wanru Peng}, doi = {10.1016/j.engappai.2025.111918}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111918}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mitigating batch normalization bias for single domain generalizable person re-identification}, volume = {160}, year = {2025}, }
• (2025). Machine learning-based investigation of activation energy for bio-convective boundary layer flow inspired by challenges in aerospace heat transfer systems. EAAI, 160, 111917. (www)

  @article{ , abstract = {In this study, a machine learning-based investigation to measure boundary layer flow even when the wall is at different temperatures and interacts with motile gyrotactic microorganisms is conducted with direct applications in optimizing fuel combustion, improving biofuel production, and designing advanced thermal management systems. To deal with this complex flow scenario and its variants, a nonlinear autoregressive with Exogenous inputs model is implemented and trained by the Levenberg-Marquardt method. The model is verified by comparing it to the reference dataset created with Adams' numerical methods. Similarity transformations are put to work in this framework, they reduce governing equations down to a manageable form that is then solved iteratively. According to the findings, wall temperature changes significantly affect the distribution and flow characteristics of microorganisms within a boundary layer. This, in turn, affects flow stability and heat transfer functions. The machine learning model achieved a high degree of accuracy, as indicated by low Mean Squared Error values, and successfully maintained its performance across diverse flow scenarios. The model achieved negligible MSE magnitudes, with scenarios reaching a Mean Square Error (MSE) of approximately 8.718 × 10 −11 to 1.073 × 10 −10 consistently indicating high prediction accuracy and alignment with reference solutions. This means that the machine learning procedure portrayed a good generalization capability. The research offers an effective tool for studying complex biofluid systems, delivering increased predictive accuracy combined with decreased processing time from simulations that involve microorganism-laden flows at different temperatures. The findings offer valuable contributions to sustainable energy technologies, environmental conservation, and the development of novel fuel systems, aligning with the growing need for energy efficiency and pollution control.}, archive = {J_EAAI}, author = {Sidra Shaheen and Tariq Mahmood and Muhammad Asif Zahoor Raja and Fuad A.M. Al-Yarimi and Muhammad Bilal Arain and Junhui Hu}, doi = {10.1016/j.engappai.2025.111917}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111917}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-based investigation of activation energy for bio-convective boundary layer flow inspired by challenges in aerospace heat transfer systems}, volume = {160}, year = {2025}, }
• (2025). Physically interpretable stockwell weight initialization and adaptive fusion average threshold for intelligent fault diagnosis of rolling bearing under noisy environment. EAAI, 160, 111916. (www)

  @article{ , abstract = {Deep learning technology has significantly advanced the field of rolling bearing fault diagnosis, delivering impressive improvements in diagnostic accuracy in recent years. These breakthroughs have revolutionized intelligent fault diagnosis, allowing for the extraction of valuable information from large datasets without manual intervention. However, despite the progress, there remains limited research in tailored weight initialization methods and noise-reduction threshold algorithms, especially in noisy environments. To tackle these challenges, we propose an innovative fault diagnosis network termed as Stockwell adaptive fusion average threshold network (SAFATN), which leverages Stockwell weight initialization to capture fault-related features and provide interpretability. This approach integrates prior physical knowledge into the first convolutional layer, making it more suitable for bearing fault diagnosis in noisy conditions. Furthermore, an adaptive fusion average threshold algorithm is introduced, which designs to enhance interactions between spatial and channel dimensions, thereby reducing noise interference. Experimental results from two distinct bearing datasets underscore that SAFATN consistently outperforms other state-of-the-art methods, showcasing superior diagnostic accuracy and robustness in noisy environments.}, archive = {J_EAAI}, author = {Lijie Zhang and Junhui Hu and Pengfei Liang and Xuefang Xu and Guoqiang Li and Zhongliang Xie and Suiyan Wang}, doi = {10.1016/j.engappai.2025.111916}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111916}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physically interpretable stockwell weight initialization and adaptive fusion average threshold for intelligent fault diagnosis of rolling bearing under noisy environment}, volume = {160}, year = {2025}, }
• (2025). Influence maximization on competitive and complementary independent cascade model based on rating-aware bidirectional gated recurrent unit. EAAI, 160, 111915. (www)

  @article{ , abstract = {Influence maximization aims to identify a set of influential users in social networks to maximize the spread of target entities, such as products or ideas. While extensive research has been conducted on single-entity propagation and competitive entity scenarios, the mechanisms of influence propagation in environments where complementary and competitive entities coexist remain underexplored. To address this limitation, we propose a novel User Historical Behavior-based Cost-Effective Lazy Forward (UHBCELF) algorithm, which integrates the Competitive and Complementary Independent Cascade (CCIC) model and the Rating-aware Bidirectional Gated Recurrent Unit (R-BiGRU) to capture complex relationships between entities. Specifically, R-BiGRU leverages users’ historical behavior sequences and associated ratings to predict the conditional probability of adopting a target entity, while the CCIC model systematically incorporates both competitive and complementary relationships into the propagation process. Based on these predictions, UHBCELF selects optimal seed nodes that improve propagation effectiveness. Experimental results on four public datasets demonstrate that our method significantly outperforms state-of-the-art approaches, achieving a 6.8% improvement in F1-score and a 7.4% increase in expected spread. This work provides a practical solution for influence maximization in multi-entity social networks, offering valuable insights for real-world applications in marketing, recommendation systems, and opinion control.}, archive = {J_EAAI}, author = {Chunlong Fu and Gang Wang and Yurui Jiang and Yajun Du}, doi = {10.1016/j.engappai.2025.111915}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111915}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Influence maximization on competitive and complementary independent cascade model based on rating-aware bidirectional gated recurrent unit}, volume = {160}, year = {2025}, }
• (2025). Resource-efficient automatic software vulnerability assessment via knowledge distillation and particle swarm optimization. EAAI, 160, 111914. (www)

  @article{ , abstract = {The increasing complexity of software systems has led to a surge in cybersecurity vulnerabilities, necessitating efficient and scalable solutions for vulnerability assessment. However, the deployment of large pre-trained models in real-world scenarios is hindered by their substantial computational and storage demands. To address this challenge, we propose a novel resource-efficient framework that integrates knowledge distillation and particle swarm optimization to enable automated vulnerability assessment. Our framework employs a two-stage approach: First, particle swarm optimization is utilized to optimize the architecture of a compact student model, balancing computational efficiency and model capacity. Second, knowledge distillation is applied to transfer critical vulnerability assessment knowledge from a large teacher model to the optimized student model. This process significantly reduces the model size while maintaining high performance. Experimental results on an enhanced MegaVul dataset, comprising 12,071 CVSS (Common Vulnerability Scoring System) v3 annotated vulnerabilities, demonstrate the effectiveness of our approach. Our approach achieves a 99.4% reduction in model size while retaining 89.3% of the original model’s accuracy. Furthermore, it outperforms state-of-the-art baselines by 1.7% in accuracy with 60% fewer parameters. The framework also reduces training time by 72.1% and architecture search time by 34.88% compared to traditional genetic algorithms.}, archive = {J_EAAI}, author = {Chaoyang Gao and Xiang Chen and Jiyu Wang and Jibin Wang and Guang Yang}, doi = {10.1016/j.engappai.2025.111914}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111914}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Resource-efficient automatic software vulnerability assessment via knowledge distillation and particle swarm optimization}, volume = {160}, year = {2025}, }
• (2025). A computation-efficient network with feature aggregation for cancer subtype classification on histopathological images. EAAI, 160, 111913. (www)

  @article{ , abstract = {Histopathology whole-slide images (WSI) capture detailed structural and morphological features of tumor tissue, offering rich histological and molecular information. Deep learning (DL) methods have emerged to assist in automatically examining histopathology WSIs and supporting tumor classification. Traditional DL approaches for WSI images face challenges due to the intrinsic complexity of tumor tissue characteristics and the extremely large image size. Multiple instance learning (MIL) methods have been proposed to address these issues by splitting the WSI images into small non-overlapping tiles and aggregating predictions from selected informative tiles for the final classification outcome. However, MIL methods still face challenges such as the need for accurate pseudo-labels, the risk of losing local information, or the failure to learn explicit class-relevant information. To address these limitations, we propose a novel framework that uses a lightweight convolutional neural network (CNN)-based tile encoder (CTE) to extract local tile features and a Transformer-based feature aggregator (TFA) to fuse local features into a representative global feature for WSI classification. Three key contributions of our framework are as follows. Firstly, we design a two-stage training strategy that decouples a lightweight CTE pre-training (using sparsely sampled tiles for efficiency) and TFA fine-tuning (using all tiles for accuracy). It significantly reduces computational costs compared to existing MIL methods while alleviating local information loss. Secondly, dynamic self-attention-based aggregation is designed in TFA, leveraging the Transformer’s self-attention to weigh all local tile features without accurate pseudo-labels. It ensures comprehensive integration of local information from both tumor and ambiguous non-tumor regions to enrich global representations from input WSIs, which benefits the final classification performance. Finally, interpretable saliency maps are generated from TFA attention scores, highlighting histopathologically relevant regions to align model decisions with clinical reasoning. Comprehensive experiments on three cancer subtype datasets demonstrate the effectiveness of our proposed method over existing MIL approaches. We also conduct further investigations into the impacts of various factors on model performance, gaining in-depth insights into our method. Our framework achieves higher classification accuracy while maintaining computational efficiency, making it a promising tool for histopathology image analysis.}, archive = {J_EAAI}, author = {Zong Fan and Chaojie Zhang and Lulu Sun and Wade Thorstad and Hiram Gay and Xiaowei Wang and Hua Li}, doi = {10.1016/j.engappai.2025.111913}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111913}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A computation-efficient network with feature aggregation for cancer subtype classification on histopathological images}, volume = {160}, year = {2025}, }
• (2025). Forecasting steam mass flow in power plants using the parallel hybrid network. EAAI, 160, 111912. (www)

  @article{ , abstract = {Efficient and sustainable power generation is a crucial concern in the energy sector. In particular, thermal power plants grapple with accurately predicting steam mass flow, which is crucial for operational efficiency and cost reduction. In this study, we use a parallel hybrid neural network architecture that combines a parameterized quantum circuit and a conventional feed-forward neural network specifically designed for time-series prediction in industrial settings to enhance predictions of steam mass flow 15 min into the future. Our results show that the parallel hybrid model outperforms standalone classical and quantum models, achieving more than 5.7 and 4.9 times lower mean squared error loss on the test set after training compared to pure classical and pure quantum networks, respectively. Furthermore, the hybrid model demonstrates smaller relative errors between the ground truth and the model predictions on the test set, up to 2 times better than the pure classical model. These findings contribute to the broader scientific understanding of how integrating quantum and classical machine learning techniques can be applied to real-world challenges faced by the energy sector, ultimately leading to optimized power plant operations. To our knowledge, this study constitutes the first parallel hybrid quantum–classical architecture deployed on a real-world power-plant dataset, illustrating how near-term quantum resources can already augment classical analytics in the energy sector.}, archive = {J_EAAI}, author = {Andrii Kurkin and Jonas Hegemann and Mo Kordzanganeh and Alexey Melnikov}, doi = {10.1016/j.engappai.2025.111912}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111912}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forecasting steam mass flow in power plants using the parallel hybrid network}, volume = {160}, year = {2025}, }
• (2025). Secure federated dataset distillation. EAAI, 160, 111911. (www)

  @article{ , abstract = {Dataset Distillation (DD) is a powerful technique for reducing large datasets into compact, representative synthetic datasets, accelerating Machine Learning training. However, traditional DD methods operate in a centralized manner, which poses significant privacy threats and reduces its applicability. To mitigate these risks, we propose a Secure Federated Data Distillation (SFDD) framework to decentralize the distillation process while preserving privacy. Unlike existing Federated Distillation techniques that focus on training global models with distilled knowledge, our approach aims to produce a distilled dataset without exposing local contributions. We leverage the gradient-matching-based distillation method, adapting it for a distributed setting where clients contribute to the distillation process without sharing raw data. The central aggregator iteratively refines a synthetic dataset by integrating client-side updates while ensuring data confidentiality. To make our approach resilient to inference attacks perpetrated by the server that could exploit gradient updates to reconstruct private data, we create an optimized Local Differential Privacy approach, called LDPO-RLD (Label Differential Privacy Obfuscation via Randomized Linear Dispersion). Furthermore, we assess the framework’s resilience against malicious clients executing backdoor attacks (such as Doorping) and demonstrate robustness under the assumption of a sufficient number of participating clients. Our experimental results demonstrate the effectiveness of SFDD and that the proposed defense concretely mitigates the identified vulnerabilities, with minimal impact on the performance of the distilled dataset. By addressing the interplay between privacy and federation in dataset distillation, this work advances the field of privacy-preserving Machine Learning making our SFDD framework a viable solution for sensitive data-sharing applications.}, archive = {J_EAAI}, author = {Marco Arazzi and Mert Cihangiroglu and Serena Nicolazzo and Antonino Nocera}, doi = {10.1016/j.engappai.2025.111911}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111911}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Secure federated dataset distillation}, volume = {160}, year = {2025}, }
• (2025). Harnessing large language models for question answering over complex tables. EAAI, 160, 111910. (www)

  @article{ , abstract = {Complex table question answering requires deriving precise answers to natural language questions by analyzing structurally complex tables, where cells are arbitrarily merged and header locations are flexible. Despite significant advancements in the large language model (LLM) era, current approaches for complex table reasoning suffer from two critical limitations: either (1) directly feeding the entire table into prompts, which introduces irrelevant information, or (2) requiring additional structures (e.g., graph-based representations) alongside LLMs, thereby incurring unnecessary system complexity. To address these limitations, we propose MaCTab , a Ma sked language modeling framework that decomposes C omplex Tab le question answering into cloze-style tasks. Specifically, MaCTab employs two modular decomposition strategies: (1) row-/column-wise retrieval and (2) cell-wise retrieval. Both strategies explicitly retrieve answers from serialized complex tables, enabling strong standalone performance. Meanwhile, their confidence-based combination filters low-confidence outputs to yield complementary gains. Comprehensive experiments on the IM-TQA benchmark — a Table Question Answering dataset with Implicit and Multi-typed structures comprising 1,200 real-world tables from publicly available business documents and 5,000 questions — demonstrate MaCTab’s effectiveness. MaCTab achieves 96.8% exact match accuracy, marking a 43.4% absolute improvement over the previous state-of-the-art (53.4%), and surpassing the average performance of five trained human annotators (95.1%) recruited by the dataset authors. This work establishes a hybrid approach that maintains the simplicity of full-table prompting while enabling graph-free information filtering, offering a simple yet effective paradigm for multi-industry applications requiring precise interpretation of complex tabular data.}, archive = {J_EAAI}, author = {Jian Lyu and Zuobin Ying and Ning Shi and Jingfeng Xue and Zijie Pan and Weiping Ding and Wanlei Zhou}, doi = {10.1016/j.engappai.2025.111910}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111910}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Harnessing large language models for question answering over complex tables}, volume = {160}, year = {2025}, }
• (2025). An approach to optimizing semantic consistency for text-to-digital human generation. EAAI, 160, 111909. (www)

  @article{ , abstract = {Digital human generation models play a crucial role in industrial applications, offering innovative directions to reduce costs and enhance benefits across fields such as education, entertainment, and marketing. However, most existing methods focus on isolated modalities within multi-modal processes, unable to synchronize semantic information with the required consistency and interdependency through prompts generated for each modality, resulting in semantic misalignments in visual appearance, auditory timbre, and emotional expressions. To address these issues, we propose VividTalker, a unified multimodal approach for text-to-digital human generation that prioritizes semantic consistency through a fully automated pipeline encompassing the synthesis of text-to-image, text-to-audio, and (image & audio)-to-video. Specifically, VividTalker generates consistent and interdependent prompts for each modality during the pre-generation stage. Then, it utilizes an Artificial Intelligence (AI) agent driven by a Small Language Model (SLM) to dynamically coordinate visual representation, auditory features, and semantic logic, ensuring improved cross-modal semantic alignment in the generated videos. Since the SLM-driven AI agent requires lower computational power and offers higher throughput, Vividtalker achieves faster generation compared to the Large Language Model (LLM) driven one. Moreover, to assess the effectiveness and stability of VividTalker, we incorporate the task success rate as one of the evaluation metrics, reducing the influence of subjective evaluation. Experimental results indicate that VividTalker achieves an 87.76% similarity to real human videos while improving the generation speed by 18.49%. This strikes a well-balanced compromise between generation speed and quality while delivering exceptional cross-modal consistency.}, archive = {J_EAAI}, author = {Xiaoyan Liu and Xingyu Liu and Peng Yang and Zhenqi Wang and Fengyuan Liu}, doi = {10.1016/j.engappai.2025.111909}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111909}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An approach to optimizing semantic consistency for text-to-digital human generation}, volume = {160}, year = {2025}, }
• (2025). Enable importance-aware model cacheability for inference serving. EAAI, 160, 111908. (www)

  @article{ , abstract = {Inference serving systems are leveraged to deploy deep learning (DL) models as services. Accelerators such as Graphics Processing Units (GPUs) have been extensively used in these systems to reduce model execution time. As accelerators become more powerful and expensive, GPU sharing among DL models across different inference requests is a common practice. However, GPU memory capacity becomes a bottleneck when the number of collocated models increases, making this approach unsustainable. At the same time, collocated models may vary in popularity levels — some are accessed frequently and others are not, leading to low resource efficiency and system performance. While some existing inference serving systems offer the capability to dynamically load and cache model in memory, they are typically locality-aware and exhibit poor performance for DL inference serving. To this end, we present mCache, a novel inference serving-oriented caching system to dynamically manage a set of collocated models with diverse popularity for efficient use of memory. mCache treats each DL model as a cacheable object, and loads model on demand and unloads models when not in use. Rather than using recency or frequency, we manage models in GPU memory based on rank-based importance scores, which jointly consider cache access patterns and model-specific factors, serving as a unified metric to compare different models. During model serving, the importance scores of cached models are dynamically updated, and the least important model is evicted to make room for a newly targeted model when the cache is full. Evaluation with representative DL models shows that mCache reduces memory footprint by nearly a half with a modest inference latency increase. Compared to existing serving system using Least Frequently Used (LFU) caching algorithm, mCache improves throughput by up to 1.5 × and 2.39 × given the 40% and 80% GPU memory capacity.}, archive = {J_EAAI}, author = {Hao Mo and Didier El Baz and Ligu Zhu and Suping Wang and Songfu Tan and Hongning Zhao and Lei Shi}, doi = {10.1016/j.engappai.2025.111908}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111908}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enable importance-aware model cacheability for inference serving}, volume = {160}, year = {2025}, }
• (2025). Loosen attention: Integrating localized channel and coarse spatial attention for enhanced analysis of complex aurora images. EAAI, 160, 111907. (www)

  @article{ , abstract = {Auroral images present a significant challenge for automated analysis due to their highly variable and dynamic morphology, influenced by complex interactions between the solar wind and Earth’s magnetosphere. These natural phenomena exhibit considerable randomness in shape, brightness, and motion, making them a unique and challenging signal source for artificial intelligence methods. In this work, we propose Loosen Attention (LA), a novel and lightweight attention mechanism tailored to capture the unpredictable and fluid-like nature of auroral patterns. LA integrates localized channel attention with coarse-grained spatial attention, forming a flexible attention framework that enhances the robustness and adaptability of feature extraction in deep learning models. The LA module is engineered around four key strategies: volumetric feature generation, volume-wise attention computation, refinement and reconstruction, and enhanced feature fusion, enabling efficient focus on subtle yet significant auroral structures while tolerating less informative regions. Unlike conventional attention methods that may struggle with complex visual patterns, LA is explicitly designed to process diverse auroral images in a structured and computationally efficient way. We validate the LA mechanism on multiple vision tasks to demonstrate consistent performance improvements over state-of-the-art attention modules. The results highlight the potential of LA to improve deep learning pipelines in geospace imaging and other domains dealing with similarly complex natural signals. This work offers a promising approach for intelligent processing of Earth–space interaction data and other visually ambiguous scientific images.}, archive = {J_EAAI}, author = {Qian Wang and Shihao Jing and Rui Yang and Zhenpei Liu and Yao Tang and Han Pan}, doi = {10.1016/j.engappai.2025.111907}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111907}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Loosen attention: Integrating localized channel and coarse spatial attention for enhanced analysis of complex aurora images}, volume = {160}, year = {2025}, }
• (2025). An ensemble-based transfer testing method for large language models. EAAI, 160, 111906. (www)

  @article{ , abstract = {Large Language Models (LLMs) can pose serious risks in real-world applications due to their potential for erroneous behavior, necessitating comprehensive and effective testing of LLMs. To assess the robustness of LLMs, adversarial attacks are typically conducted by constructing adversarial examples. Previous methods often require extensive queries and access to the internal information of the victim model. However, the internal information of most black-box LLMs is not accessible, rendering these testing methods infeasible. In addition, excessive queries to commercial black-box LLMs may incur substantial costs. To address these issues, this paper proposes an E nsemble-based T ransfer T esting method for L arge L anguage M odels (ETTLLM). In contrast to previous adversarial testing methods for LLMs, ETTLLM queries white-box surrogates rather than the victim model, thereby significantly reducing testing costs. Moreover, it enhances the transferability and generalization of adversarial examples across diverse real-world classification tasks. Compared to baselines, ETTLLM significantly reduced the number of queries to the victim model, with an average of 1.6 queries, just 1.2% of the baselines. Furthermore, the textual similarity and modification rate of the adversarial examples generated by ETTLLM differ from the baselines by no more than 1.6%, while achieving 70% of the attack success rate compared to the baselines.}, archive = {J_EAAI}, author = {Yuanxin Qiao and Yong Liu and Xiang Chen and Zhanqi Cui}, doi = {10.1016/j.engappai.2025.111906}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111906}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An ensemble-based transfer testing method for large language models}, volume = {160}, year = {2025}, }
• (2025). Simple yet robust markerless motion capture system using deep learning. EAAI, 160, 111905. (www)

  @article{ , abstract = {Absolute Three Dimensional human pose estimation (3DHPE) is essential for a wide range of applications, including virtual reality, animation, sports analytics, and biomechanical studies. Traditional marker-based motion capture systems, however, remain expensive, complex to set up, and dependent on specialized hardware, limiting their broader adoption. In this work, we present a unified and accessible marker-less motion capture system designed to overcome these limitations by leveraging a multi-view camera setup composed of low-cost Internet of Things (IoT) devices, such as Raspberry Pi units and a deep learning algorithm. Our pipeline begins by extracting two dimension (2D) poses from each camera view, which are then fused using a learned triangulation method to generate an initial set of absolute three dimension (3D) pose points. To further enhance accuracy, these poses are processed by a spatio-temporal transformer that models the dynamics and relationships among pose points across consecutive frames, ensuring both spatial precision and temporal coherence. The network is trained with additional constraints on bone length consistency and absolute pose error to improve robustness in real-world scenarios. For experimentation, the system is deployed on a local server to demonstrate real-time performance and ease of deployment. We validate our system on the Human3.6 m and Panoptic datasets, achieving state-of-the-art results while significantly reducing the cost and complexity of 3D human motion capture. This work opens up new possibilities for democratizing high-quality motion capture in both research and industry settings.}, archive = {J_EAAI}, author = {Avinash Upadhyay and Ankit Shukla and Manoj Sharma and Ajith Abraham}, doi = {10.1016/j.engappai.2025.111905}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111905}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Simple yet robust markerless motion capture system using deep learning}, volume = {160}, year = {2025}, }
• (2025). A study of the efficacy of generative flow networks for robotics and machine fault-adaptation. EAAI, 160, 111904. (www)

  @article{ , abstract = {Advancements in robotics have opened possibilities to automate tasks in various fields such as manufacturing, emergency response and healthcare. However, a significant challenge that prevents robots from operating in real-world environments effectively is out-of-distribution (OOD) situations, wherein robots encounter unforseen situations. One major OOD situations is when robots encounter faults, making fault adaptation essential for real-world operation for robots. Current state-of-the-art reinforcement learning algorithms show promising results but suffer from sample inefficiency, leading to low adaptation speed due to their limited ability to generalize to OOD situations. Our research is a step towards adding hardware fault tolerance and fast fault adaptability to machines. In this research, our primary focus is to investigate the efficacy of generative flow networks in robotic environments, particularly in the domain of machine fault adaptation. We simulated a robotic environment called Reacher in our experiments. We modify this environment to introduce four distinct fault environments that replicate real-world machines/robot malfunctions. The empirical evaluation of this research indicates that continuous generative flow networks (CFlowNets) indeed have the capability to add adaptive behaviors in machines under adversarial conditions. Furthermore, the comparative analysis of CFlowNets with reinforcement learning algorithms also provides some key insights into the performance in terms of adaptation speed and sample efficiency. Additionally, a separate study investigates the implications of transferring knowledge from pre-fault task to post-fault environments. Our experiments confirm that CFlowNets has the potential to be deployed in a real-world machine and it can demonstrate adaptability in case of malfunctions to maintain functionality.}, archive = {J_EAAI}, author = {Zahin Sufiyan and Shadan Golestan and Shotaro Miwa and Yoshihiro Mitsuka and Osmar Zaiane}, doi = {10.1016/j.engappai.2025.111904}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111904}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A study of the efficacy of generative flow networks for robotics and machine fault-adaptation}, volume = {160}, year = {2025}, }
• (2025). Deep learning-based modeling of stakeholders preferences and balancing through multi-objective optimization in a multi-stakeholder recommendation system. EAAI, 160, 111903. (www)

  @article{ , abstract = {Satisfying the stakeholders' preferences and establishing a trade-off recommendation environment is a challenging problem in a multi-stakeholder recommendation system. Recent techniques have maximized consumer satisfaction by delivering diverse and novel item suggestions but do not satisfy provider and system-level goals. The multi-stakeholder recommendation methods exploit highly sparse user-item interaction to compute providers' and systems' utility and deliver recommendations without establishing trade-offs among stakeholders' conflicting preferences. It hinders their comprehension of stakeholders' preferences and minimizes the likelihood of item consumption, providers' exposure, and stakeholders' retention in the system. Therefore, this study proposes a Deep learning-based Multi-Stakeholder recommendation model and develops a personalized Consumption, Exposure, and Retention maximization-based objective function for consumer, provider, and system stakeholders. This work employs a deep autoencoder to learn latent implicit features of stakeholder-item interactions. Next, we propose a deep neural network-based probabilistic modeling of the objective functions that maximize the likelihood of item consumption, providers' exposure, and stakeholders' retention. Finally, this study employs evolutionary multi-objective optimization to establish the trade-off recommendation solution between stakeholders' conflicting objectives. The simulations and evaluation results using standard evaluation metrics for each stakeholder across four benchmark datasets validate the proposed models' performance over baseline methods.}, archive = {J_EAAI}, author = {Rahul Shrivastava and Dilip Singh Sisodia and Naresh Kumar Nagwani}, doi = {10.1016/j.engappai.2025.111903}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111903}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning-based modeling of stakeholders preferences and balancing through multi-objective optimization in a multi-stakeholder recommendation system}, volume = {160}, year = {2025}, }
• (2025). SCB-ADAE: An attention-based deep autoencoder for ground penetrating radar signal denoising. EAAI, 160, 111902. (www)

  @article{ , abstract = {In buried object detection, recorded signals of a ground penetrating radar (GPR) inevitably include noise interference owing to complex underground environments. Existing rule- and data-driven denoising methods struggle to handle non-Gaussian and real-world noise because the rule-driven ones rely on the assumptions of simplified noise characteristics and the data-driven ones cannot capture fine- and global-scale features of a GPR signal well. To address the problem, this study proposes an attention-based denoising model called the Swin-Conv Block with Attention Denoising Autoencoder (SCB-ADAE). The model first feeds a GPR signal into a SCB module, which extracts a tensor with the fine-scale features in the signal, such as sharp reflective interfaces and abrupt amplitude variations. The feature tensor then passes through an ADAE module that uses encoder-decoder structure with the self-attention to enhances the representation of the global-scale signal features. Finally, the feature tensor from the ADAE module is decoded by another SCB module to generate a denoised GPR signal, where the tensor includes the fine-scale and global features of the raw signal. An experiment with three types of GPR signals demonstrates the effectiveness of the proposed model: radar signals with Gaussian noise, radar signals with inhomogeneous-material noise, and real-world signals. radar signals with Gaussian noise, radar signals with inhomogeneous-material noise, and real-world signals. Experimental results demonstrate that the proposed model outperforms other state-of-the-art denoising methods on denosing the three types of GPR signals, where the signal-to-noise ratio, peak signal-to-noise ratio, and structural similarity index are improved to 20.64, 14.59, and 0.366, respectively.}, archive = {J_EAAI}, author = {Jiahao Liu and Yiming Zhang and Liang Song and Zheng Tong}, doi = {10.1016/j.engappai.2025.111902}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111902}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SCB-ADAE: An attention-based deep autoencoder for ground penetrating radar signal denoising}, volume = {160}, year = {2025}, }
• (2025). GHR-2D: Gaze and head redirection via disentanglement and diffusion for gaze estimation. EAAI, 160, 111901. (www)

  @article{ , abstract = {The performance of appearance-based gaze estimation models significantly decreases during cross-domain evaluation due to domain gaps. Gaze redirection edits the gaze direction and head pose of existing faces to expand the dataset and bridge the inter-domain gap. However, the generated results are contaminated by gaze-irrelevant information. In addition, physiological structures can cause gaze shifts. We introduce a face image redirection method based on a diffusion model that interprets head pose and gaze redirection in the latent space. An encoder maps two-dimensional vectors to the latent space, aligning them with image features while preserving angle information. Relative gaze compensates for deviations in the visual and optical axes caused by physiological structures. We also propose a training pipeline for disentangling identity, gaze, and head pose embeddings. An encoder mines the relationship between gaze and head, mapping input pitch and yaw into conditions to guide the generation of gaze and head pose embeddings in a diffusion model. The decoder projects the connected identities and modified embeddings back into the feature space. Finally, the redirected samples expand the original dataset and facilitate cross-domain evaluation. Compared to state-of-the-art methods, our approach significantly improves gaze accuracy.}, archive = {J_EAAI}, author = {Daosong Hu and Mingyue Cui and Kai Huang}, doi = {10.1016/j.engappai.2025.111901}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111901}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {GHR-2D: Gaze and head redirection via disentanglement and diffusion for gaze estimation}, volume = {160}, year = {2025}, }
• (2025). A cross-domain multi-scale feature fusion network based on graph convolution for intelligent fault diagnosis. EAAI, 160, 111900. (www)

  @article{ , abstract = {Fault diagnosis of mechanical equipment plays a critical role in enhancing system stability and ensuring operational safety. Multi-modal monitoring data provides a more comprehensive view of the equipment’s condition, enabling more precise state diagnosis through the processing of such data. A major challenge in contemporary multi-modal information fusion lies in the effective integration of cross-modal information. To overcome the inherent limitations of single-view approaches often found in existing multi-modal methods, this paper presents a multi-view, multi-modal information fusion approach based on graph convolutional networks. This method not only extracts key features from signals efficiently but also uncovers the interrelationships between different modal signals through a multi-view interaction mechanism, achieving more robust information fusion and enhanced fault diagnosis performance. This method consists of three core modules: feature extraction, information fusion, and classification. The feature extraction module utilizes a multi-level residual architecture, Mini-Long Short Term Memory, and self-attention mechanisms to capture both local and global signal features, model temporal dependencies, and refine feature selection. The information fusion module combines data and feature-level signal correlations using graph convolutional networks for cross-modal interaction. The classification module employs a two-layer fully connected network for fault diagnosis. This method quantitatively assesses the contribution of each modality, providing a theoretical basis for understanding their physical significance. Experimental results and ablation studies demonstrate its superior performance and enhanced accuracy in fault diagnosis, offering a novel approach for condition monitoring of complex equipment.}, archive = {J_EAAI}, author = {Quanyu Zhong and Qiang Li and Junxiao Ren and Xin Chen and Dong Liu and Qiang Yang}, doi = {10.1016/j.engappai.2025.111900}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111900}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cross-domain multi-scale feature fusion network based on graph convolution for intelligent fault diagnosis}, volume = {160}, year = {2025}, }
• (2025). Dual-phase feature selection using adaptive neighborhood rough sets and hybrid sine-cosine optimization for classification. EAAI, 160, 111899. (www)

  @article{ , abstract = {High-dimensional, multi-class, and imbalanced datasets present significant challenges in classification tasks across various industries, including healthcare, finance, and image processing. Existing feature selection methods, particularly those based on neighborhood rough sets, often struggle with handling both feature redundancy and noisy samples, making it difficult to capture the complex distribution of features and samples across different classes. To address this, we propose a dual-phase feature selection method that performs joint optimization in both horizontal (feature-level) and vertical (sample-level) dimensions. In the first phase, adaptive neighborhood rough set theory is used for horizontal feature selection. By adjusting the neighborhood radius ( δ ) and inclusion degree ( λ ) through cross-validation, the method selects relevant feature subsets tailored to the granularity of each dataset, thereby improving generalization. In the second phase, a hybrid sine cosine algorithm is employed for vertical processing to optimize sample selection. This algorithm iteratively removes noisy or misleading samples based on fitness evaluation, enhancing the model’s robustness. Furthermore, the framework integrates an enhanced fuzzy k-nearest neighbor classifier that leverages feature subset weights for each class to better address class imbalance during classification. Extensive experiments on 21 public datasets, using three types of classifiers, show that the proposed method outperforms seven benchmark feature selection algorithms in terms of classification accuracy, weighted precision, weighted recall, and weighted F1-score. Statistical tests, including the Wilcoxon signed-rank test, confirm significant improvements. This dual-phase horizontal and vertical optimization approach offers a robust and effective solution for real-world classification tasks involving complex data distributions.}, archive = {J_EAAI}, author = {Chengfeng Zheng and Mohd Shareduwan Mohd Kasihmuddin and Zhizhong Yan and Mohd. Asyraf Mansor and Yuan Gao and Ju Chen}, doi = {10.1016/j.engappai.2025.111899}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111899}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-phase feature selection using adaptive neighborhood rough sets and hybrid sine-cosine optimization for classification}, volume = {160}, year = {2025}, }
• (2025). Iterative contextual and adaptive strategies for enhanced monocular depth estimation. EAAI, 160, 111898. (www)

  @article{ , abstract = {Monocular Depth Estimation (MDE) is an important computer vision task within artificial intelligence (AI). It is widely used in fields such as autonomous navigation, robotics, and augmented and virtual reality. Current approaches often face difficulties in accurately estimating depth, particularly in scenes with complicated shapes, occlusions, and diverse depths. To overcome these challenges, we introduce an innovative AI framework that combines two main methods: Iterative Elastic Bins (IEBins) and a Contextual Transformer-based Feature Extractor (CTFE). The IEBins method improves depth predictions by repeatedly refining the depth estimation bins according to the scene's depth uncertainty. The CTFE method, using a Swin Transformer architecture, extracts both local and global context information, which significantly helps refine depth predictions. Our integrated approach, called the Progressive Refinement Strategy, effectively reduces errors like blurred edges and inconsistent depth predictions. Comprehensive testing on standard datasets—including Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI), New York University Depth Dataset Version 2 (NYU-Depth-v2), and Scene Understanding RGB-Dataset (SUN RGB-D)—demonstrates that our method outperforms existing models in accuracy and efficiency. Moreover, our AI model generalizes effectively to new, unseen scenarios, and is computationally efficient enough for real-time use in practical applications such as autonomous vehicles, robotics systems, and immersive virtual and augmented reality experiences.}, archive = {J_EAAI}, author = {Sabina Umirzakova and Sevara Mardieva and Shakhnoza Muksimova and Jushkin Baltayev and Young Im Cho}, doi = {10.1016/j.engappai.2025.111898}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111898}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Iterative contextual and adaptive strategies for enhanced monocular depth estimation}, volume = {160}, year = {2025}, }
• (2025). Mastering autonomous assembly in fusion application with learning-by-doing: A peg-in-hole study. EAAI, 160, 111897. (www)

  @article{ , abstract = {Robotic peg-in-hole assembly represents a critical area of investigation in robotic automation. Traditional approaches primarily rely on optical or force/torque (F/T) sensors, each with inherent limitations: restricted assembly accuracy or inefficient peg-hole alignment. Deep Reinforcement Learning (DRL) based methods have the potential to combine data from both types of sensors to achieve improved results. However, our application scenario is situated inside a fusion reactor, where the radiation environment and the abundance of smooth metal surfaces make commonly used three-dimensional (3D) sensors face operational challenges. To address this, we propose a novel DRL-based approach that, unlike conventional methods, integrates data from a two-dimensional (2D) camera and an F/T sensor. This approach trains the agent to perform peg-in-hole assembly tasks by mimicking human hand-eye coordination. It utilizes multi-input branch neural network to fuse multi-sensor data with significant differences and automatically adjusts the weights of multi-source data at different stages of assembly, thereby simultaneously meeting the requirements of fast alignment and high-precision assembly. Real-world experimental results demonstrate the effectiveness of this multi-sensor fusion approach, particularly in rigid peg-in-hole assembly tasks. It achieves rapid assembly with a peg-hole clearance of less than 0.1 mm (smaller than the repeatable accuracy of the robotic arm we used) while maintaining a high success rate and avoiding risky behaviors. This study shows that high-precision peg-in-hole assembly operations can be successfully accomplished using only a 2D camera as the optical sensor. This advancement significantly contributes to the development of high-precision automated operations in challenging environments, such as radiation-exposed settings.}, archive = {J_EAAI}, author = {Ruochen Yin and Huapeng Wu and Ming Li and Yong Cheng and Yuntao Song and Hongtao Pan and Heikki Handroos}, doi = {10.1016/j.engappai.2025.111897}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111897}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mastering autonomous assembly in fusion application with learning-by-doing: A peg-in-hole study}, volume = {160}, year = {2025}, }
• (2025). Collaborative multiple attention mechanisms for vehicle fault prediction. EAAI, 160, 111896. (www)

  @article{ , abstract = {Accurate prediction of vehicle faults is crucial for extending the service life of vehicles and enhancing their safety. However, current research predominantly focuses on predicting a subset of vehicle faults and falls short in providing a comprehensive prediction of all potential issues. To address these limitations, we propose a vehicle fault prediction method that leverages the collaboration of multiple attention mechanisms. To capture the temporal correlation between vehicle faults across different periods, we introduce a dual temporal correlation learning module that learns the fault correlation from multiple perspectives. Additionally, to efficiently model the interdependencies between different types of faults, we propose a graph attention correlation perception module. Experimenting with real vehicle breakdown records, our model achieves 41.90% weighted F 1 score ( w-F 1 ; a class-balanced metric combining precision and recall) with recall at top- k predictions ( R@k , proportion of relevant items in top- k results) reaching 57.55% ( k =5), 65.17% ( k =10), 70.20% ( k =15), 73.75% ( k =20), 76.67% ( k =25), 79.16% ( k =30), and 81.23% ( k =35). The results, which all exceed the performance of the best models available today, highlight the system’s superiority in predicting vehicle faults. The source code and dataset are available at https://github.com/FanghuaX/CoMAM}, archive = {J_EAAI}, author = {Fanghua Chen and Deguang Shang and Gang Zhou and Ke Ye and Fujie Ren and Guofang Wu}, doi = {10.1016/j.engappai.2025.111896}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111896}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Collaborative multiple attention mechanisms for vehicle fault prediction}, volume = {160}, year = {2025}, }
• (2025). Dual Stimulus–Organism–Response paradigms cross-domain aware model for rumor detection. EAAI, 160, 111895. (www)

  @article{ , abstract = {How to effectively detect rumors on social media has gained much attention in recent years. However, relatively little focus has been given on exploiting psychological crowd signal in user comments left for posts in rumor detection. Therefore, based on Stimulus–Organism–Response (SOR) theory, we first construct the dual Stimulus–Organism–Response paradigms between post and commenting users, and then propose the dual Stimulus–Organism–Response paradigms cross-domain aware model for rumor detection. Particularly, the model is composed of three components: (a) dual SOR paradigms Feature Extractor, which extracts both the factual refutation and emotional resonance SOR paradigm features. We concatenate these two features to form the dual SOR feature. (b) Rumor Classifier. (c) Domain Classifier. In general, the model conducts a minimax game between the rumor classifier and the domain classifier, and learns domain-invariant dual SOR paradigms features for cross-domain rumor classification. Extensive experiments conducted on real-world datasets demonstrates the effectiveness of the proposed model for rumor detection and give explainable results.}, archive = {J_EAAI}, author = {Yunyun Li and Jiahua Jin}, doi = {10.1016/j.engappai.2025.111895}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111895}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual Stimulus–Organism–Response paradigms cross-domain aware model for rumor detection}, volume = {160}, year = {2025}, }
• (2025). Multi-task learning for multi-label electrocardiogram disease detection via point-level delineation-guided fusion. EAAI, 160, 111894. (www)

  @article{ , abstract = {Despite substantial research over the past decade on analyzing electrocardiograms (ECGs) using deep learning models, disease detection in complex scenarios remains challenging, particularly for co-occurring diseases. Additionally, existing methods lack sufficient interpretability for multi-label disease classification. To address these issues, we propose the Point-level Delineation Guided Multi-Task Learning network (PDG-MTL), integrating the main task of ECG classification with an auxiliary task of ECG delineation. Firstly, during training, fiducial points are generated using the ECGDeli tool and converted into labels for the auxiliary task, including three segments and three waves of ECG delineation. Secondly, we proposed a novel decoder-focused multi-task architecture comprising initial task predictions and an attention-guided fusion module. The initial prediction heads provide deep supervision, obtaining multi-modal predictions that include both regions of interest for ECG delineation and meaningful diagnostic features. The attention-guided module fuses the feature maps of waveforms and delineation information, which bridge the dimensional gap between heterogeneous features. In the results, PDG-MTL outperforms prior models across a diverse patient cohort on a large-scale dataset, particularly in multi-label classification involving over 20 classes. The average Area under the Receiver Operating Characteristic (AUROC) increases by up to 3.4 % for morphological cardiac diseases. It also significantly improves performance for low-sample diseases, with F1 scores increasing by nearly 60 %. The model demonstrates exceptional robustness, with minimal confidence intervals across six experiments. The attention map reveals alignment between the model's focus on delineation regions and expert expectations across disease groups, highlighting the potential of interpretability.}, archive = {J_EAAI}, author = {Mengxiao Wang and Yuanyuan Tian and Zhiyuan Li and Xiaoyang Wei and Yanrui Jin and Chengliang Liu and Liqun Zhao}, doi = {10.1016/j.engappai.2025.111894}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111894}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-task learning for multi-label electrocardiogram disease detection via point-level delineation-guided fusion}, volume = {160}, year = {2025}, }
• (2025). Metamodeling elastic wave propagation using a mixed factorized fourier encoder–decoder for online laser-ultrasound testing in additive manufacturing. EAAI, 160, 111893. (www)

  @article{ , abstract = {Laser-ultrasound (LU) testing has emerged as a promising technique for characterizing the polycrystalline microstructure of metal components produced by wire-laser additive manufacturing (WLAM), with potential for real-time online application. Numerical models simulating elastic waves propagation provide valuable insights into the relationship between microstructural properties and laser-induced displacements, but their computational cost renders them impractical for automated high-throughput characterization. To overcome this limitation, we build a metamodel that maps a wide variety of two-dimensional anisotropic polycrystalline microstructures — simplified but representative of features commonly observed in WLAM — to simulated surface displacements. Addressing this challenging high-dimensional regression problem, several neural network surrogates relying on a novel combination of layers are investigated. Their architectures include usual convolutional encoder–decoder elements and spectral layers inspired from the Fourier neural operator (FNO) framework. These layers are adapted and some variants are proposed to provide versatility in network design. All metamodels can run both a forward and backward pass at least 100 times faster than a single forward call of the original model. The best architecture implies a trade-off between computational cost and accuracy. Notably, the architecture involving the channel-wise factorized variant of the spectral layers, which is characterized by a relatively small number of parameters, achieved the lowest approximation error. The metamodel successfully captures the primary effects of anisotropy on wave propagation, even for low-anisotropy inputs not included in the training data. These findings represent a promising initial step towards addressing inversion problems and facilitating the development of online LU testing protocols in additive manufacturing.}, archive = {J_EAAI}, author = {Frédéric Allaire and Filippo Gatti and Roberto Miorelli and Vincent Dorval and Jérôme Laurent}, doi = {10.1016/j.engappai.2025.111893}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111893}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Metamodeling elastic wave propagation using a mixed factorized fourier encoder–decoder for online laser-ultrasound testing in additive manufacturing}, volume = {160}, year = {2025}, }
• (2025). Multi-task driver gaze estimation in real world driving scenes. EAAI, 160, 111892. (www)

  @article{ , abstract = {Driver gaze estimation task is pivotal for safe driving. However, challenges persist when dealing with changing illumination, eyeglasses, adjacent zones, or personal behavior and appearance. To tackle these problems, we introduce a driver gaze estimation approach, including gaze zone and direction estimation. We propose a global facial feature extraction convolutional neural network (gCNN) embedded with attention network for driver gaze zone estimation. The incorporation of attention mechanisms in different dimensions (channel or spatial) at various stages facilitates the network in efficiently capturing overall generic features in early stages and concrete representations in later stages. This network is also applied to extract facial features in gaze direction estimation task. While a local eye feature extraction convolutional neural network (LeCNN) is proposed for fine-grained eye features extraction. The facial and eye features, as well as head pose, are concatenated and fused to regress the finer gaze direction. The experimental results show that the network achieves an error of 2.43° and 4.36° on MPIIFaceGaze and EyeDIAP datasets, respectively, outperforming the prior arts. Furthermore, in driver gaze zone estimation task, our method achieves accuracy of 98.87 % on Laboratory for Intelligent and Safe Automobiles (LISA) Gaze dataset, with 3.91 % improvement over prior arts. It also achieves a compatible performance of 82.80 % on Driver Gaze in the Wild (DGW) dataset.}, archive = {J_EAAI}, author = {Xinmei Wu and Lin Li and Gang Zhou and Qilong Wu and Xinkai Zuo and Haihong Zhu and Shen He}, doi = {10.1016/j.engappai.2025.111892}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111892}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-task driver gaze estimation in real world driving scenes}, volume = {160}, year = {2025}, }
• (2025). Learning-centered multi-modal federated learning complimented with filter rotation for clinical image diagnosis. EAAI, 160, 111891. (www)

  @article{ , abstract = {The demand for accurate and efficient diagnostic tools in medical imaging, highlighted by the COronaVIrus Disease of 2019 (COVID-19) pandemic, has emphasized the need for robust computer-aided diagnostic systems. However, the challenge of privacy preservation in data sharing across institutions remains a hurdle. To address these, this study introduces Federated Network (FedNet), a federated learning (FL)-based framework designed to facilitate decentralized training of medical data while preserving privacy. FedNet incorporates innovative methodologies: a filter rotation technique to enhance feature extraction by capturing image features at varying orientations, and Dark-mixer image augmentation to improve image contrast and feature visibility. A novel learning-centered multi-modal learning (L-CML) approach, focusing on learning of modalities rather than traditional data modalities, to improve learning efficiency is introduced. These techniques are implemented within an FL setup, using the PySyft encryption and decryption technique for privacy protection. The objective of FedNet is to improve accuracy and efficiency in medical imaging through decentralized and privacy-preserving learning. This study aims to demonstrate the adaptability of FedNet to diverse medical imaging applications. Through extensive experiments, FedNet demonstrated promising results, achieving training and validation accuracies of 96.47 % and 95.50 % , respectively. FedNet outperformed frameworks such as Federated Averaging (FedAvg), Federated Proximal (FedProx), and Federated Batch Normalization (FedBN) in key metrics, such as precision, recall, and computational efficiency, attaining accuracies of 94.01 % , 93.50 % , and 95.92 % , respectively. FedNet's filter rotation, Dark-mixer, and L-CML integration enhanced feature extraction and model performance, offering a promising solution for advancing medical image analysis toward accurate, efficient, and privacy-preserving diagnostics.}, archive = {J_EAAI}, author = {Isaac Adjei-Mensah and Qingxian Wang and Isaac Osei Agyemang and Adu Asare Baffour and Sophyani Banaamwini Yussif and Collins Sey and Linda Delali Fiasam and Ijeoma Amuche Chikwendu and Weidong Wang}, doi = {10.1016/j.engappai.2025.111891}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111891}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Learning-centered multi-modal federated learning complimented with filter rotation for clinical image diagnosis}, volume = {160}, year = {2025}, }
• (2025). Research on mobile robot path planning based on improved Q-evaluation ant colony optimization algorithm. EAAI, 160, 111890. (www)

  @article{ , abstract = {This paper proposes an improved Q-evaluation ant colony optimization (IQACO) algorithm to address the shortcomings of traditional ant colony optimization (ACO) algorithm in solving path planning problems, such as the difficulty of effectively reflecting the quality of nodes through the concentration of pheromones, and the problem of poor search stability caused by excessive hyperparameters at different map sizes. Firstly, by analyzing the optimization process of the Q-Learning algorithm, the feasibility of evaluating nodes with Q-value was demonstrated. Secondly, an update method from path transformation to node Q-value was studied, and its superiority over pheromone concentration was demonstrated. Finally, a node selection strategy based on Q-evaluation was proposed to reduce the hyperparameters of the algorithm and improve its stability in optimizing environments of different sizes. To validate the effectiveness of the proposed improvement strategy, both theoretical analysis and repetitive experimental results demonstrate a significant reduction in the time complexity of the proposed algorithm compared to the traditional Ant Colony Optimization (ACO) algorithm. Additionally, simulations conducted on four different map sizes show that the reduction in hyperparameters enhances the algorithm's stability across various map scales. In addition, comparative experiments with other Q-learning-enhanced ACO variants further highlight the advantages of the proposed IQACO algorithm in terms of solution quality, convergence stability, and real-time performance. Furthermore, through comparisons with related work and recent studies, the proposed algorithm's superiority in terms of search capability and stability is clearly demonstrated.}, archive = {J_EAAI}, author = {Dongdong Li and Lei Wang}, doi = {10.1016/j.engappai.2025.111890}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111890}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on mobile robot path planning based on improved Q-evaluation ant colony optimization algorithm}, volume = {160}, year = {2025}, }
• (2025). A neuro-inspired approach for continual multi-label learning with evolving spiking networks. EAAI, 160, 111889. (www)

  @article{ , abstract = {Real-world data is often accompanied by multiple labels per sample and generally arrives in a continuous stream without well-defined task boundaries. This is the problem of Continual Multi-Label Learning (CMLL). The high imbalance in multi-label data, coupled with privacy and storage constraints, renders traditional memory-based approaches ineffective for CMLL. Therefore task-agnostic and memory-free CMLL methods with minimal catastrophic forgetting are needed to learn from real-world data streams. Artificial Neural Networks (ANNs) update all the weights leading to a poor stability vs plasticity tradeoff. As a result, ANN-based Continual Learning (CL) approaches might suffer from more catastrophic forgetting in multi-label data streams due to high imbalance. While Spiking Neural Networks (SNNs) employing biologically inspired local learning rules are known to provide good stability vs. plasticity tradeoff, multi-label classification and CMLL are underexplored areas in SNN literature. We propose Neuro-Inspired Continual Multi-Label Learning (NIMBLE), an architecture-based, task-agnostic, and memory-free CMLL framework based on SNNs to address these limitations. The evolving feature space of NIMBLE provides enough plasticity to learn new patterns and retain feature correlations while employing a winner-takes-all mechanism to unambiguously predict multi-labeled outputs. NIMBLE also uses the winner-takes-all mechanism to apply localized updates to selective interneurons, enforcing the stability of the representations learned by the other interneurons. Experimental results across 15 real-world datasets demonstrate that NIMBLE outperforms previous ANN-based CL and CMLL benchmarks on the micro-F1 and Jaccard scores, with the highest gains of 0.26 and 0.18, respectively, on the GNegative dataset.}, archive = {J_EAAI}, author = {Sourav Mishra and Suresh Sundaram and P.Md. Thousif}, doi = {10.1016/j.engappai.2025.111889}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111889}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A neuro-inspired approach for continual multi-label learning with evolving spiking networks}, volume = {160}, year = {2025}, }
• (2025). Motion control strategy for robotic arm with deep cascade feature bayesian broad learning system. EAAI, 160, 111888. (www)

  @article{ , abstract = {The motion control strategy based on broad learning system (BLS) is a typical technology of intelligent control. It simplifies the complex and time-consuming process of parameter tuning and offers significant advantages over conventional methods. Therefore, it has great potential for application in the field of robotic arm control. However, current BLS-based control strategies suffer from model instability and limited prediction accuracy. In this study, a novel motion control strategy based on deep cascade feature Bayesian broad learning system is proposed to solve these problems. Firstly, the sparse self-encoder is introduced to optimize the BLS-based motion control network to improve the performance of feature extraction. Secondly, Bayesian regression is employed to obtain the network parameters of the control strategy, which improves the accuracy of the robot arm motion control. Finally, the constraints are used to guide the generation of network parameters, ensuring control system stability. Simulation experiments are designed and compared with existing learning-based control strategies. And, physical experiments are performed with the AUBO_i5 robotic arm. The proposed method is validated as feasible and superior by the experimental results. Compared to the state of the art, the root mean square error of the proposed method is reduced to 0.133 rad, which is a reduction of 54.91%.}, archive = {J_EAAI}, author = {Jiyong Zhou and Guoyu Zuo and Shuwen Zhang and Shuangyue Yu and Chunfang Liu}, doi = {10.1016/j.engappai.2025.111888}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111888}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Motion control strategy for robotic arm with deep cascade feature bayesian broad learning system}, volume = {160}, year = {2025}, }
• (2025). Unified bayesian representation for high-dimensional multi-modal biomedical data for small-sample classification. EAAI, 160, 111887. (www)

  @article{ , abstract = {The increasing availability of multi-modal medical data, including neuroimaging, genetic profiles, and clinical measurements, offers unprecedented opportunities for advancing disease diagnosis and prognosis. However, integrating these heterogeneous data sources poses significant challenges due to their high dimensionality, redundancy, and small sample sizes, which hinder the effectiveness of traditional machine learning models. To overcome these challenges, we present the BAyesian Latent Data Unified Representation model (BALDUR), a novel Bayesian algorithm designed to deal with multi-modal datasets and small sample sizes in high-dimensional settings while providing explainable solutions. To do so, the proposed model combines within a common latent space the different data views to extract the relevant information to solve the classification task and prune out the irrelevant/redundant features/data views. Furthermore, to provide generalizable solutions in small sample size scenarios, BALDUR efficiently integrates dual kernels over the views with a small sample-to-feature ratio. Finally, its linear nature ensures the explainability of the model outcomes, allowing its use for biomarker identification. This model was tested over two different neurodegeneration datasets, outperforming the state-of-the-art models and detecting features aligned with markers already described in the scientific literature.}, archive = {J_EAAI}, author = {Albert Belenguer-Llorens and Carlos Sevilla-Salcedo and Jussi Tohka and Vanessa Gómez-Verdejo and Alzheimer’s Disease Neuroimaging Initiative}, doi = {10.1016/j.engappai.2025.111887}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111887}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unified bayesian representation for high-dimensional multi-modal biomedical data for small-sample classification}, volume = {160}, year = {2025}, }
• (2025). Computer vision in warehouse management automation: A survey on implemented methods with prototyping hardware. EAAI, 160, 111886. (www)

  @article{ , abstract = {In the context of the fourth industrial revolution, warehouse management systems are increasingly integrating advanced computer vision technologies, such as Autonomous Mobile Robots (AMRs), drones, Unmanned Aerial Vehicles (UAVs), robotic arms, and smart glasses. Despite their growing adoption, a critical gap remains in understanding the specific hardware configurations and computer vision methodologies used in prototyping for warehouse automation. This survey offers a comprehensive analysis of recent prototyping efforts by exploring the relationship between computer vision, vision sensor technologies, and Artificial Intelligence (AI) hardware. Through rigorous manual screening of 1,570 papers, we identify 50 key studies that explicitly detail the implementation of computer vision techniques in warehouse automation tasks since 2013. Unlike existing reviews, our work uniquely highlights the gap between computer vision methods and their deployment on edge AI hardware for warehouse automation. We analyze the adopted techniques in image processing, object detection, classification, segmentation, and navigation, alongside the latest advancements in AI accelerators and robotic platforms. Our findings stress the necessity of transitioning to next-generation AI hardware, implementing standardized benchmarking methods, leveraging photorealistic synthetic datasets, and deploying vision-based AI models on advanced hardware platforms. This surveying method ensures robustness and easy reproducibility for future use, serving as a critical reference for researchers and industry practitioners seeking to stay continuously updated on warehouse automation prototypes with computer vision.}, archive = {J_EAAI}, author = {Chaouki Tadjine and Abdelkrim Ouafi and Azeddine Benlamoudi and Abdelmalik Taleb-Ahmed}, doi = {10.1016/j.engappai.2025.111886}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111886}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Computer vision in warehouse management automation: A survey on implemented methods with prototyping hardware}, volume = {160}, year = {2025}, }
• (2025). Analyzing food bolus formation during chewing through image segmentation. EAAI, 160, 111884. (www)

  @article{ , abstract = {Mastication, as the initial stage of digestion, transforms food into swallowable boluses through complex mechanical and biochemical processes that are critical for overall digestive health. Traditional evaluation methods such as particle size distribution analysis provide quantitative measurements but require destructive sample processing that compromises food bolus structural integrity, making continuous monitoring impossible and eliminating valuable spatial relationship information between particles. This study presents the first implementation of a deep learning system using Mask Region-based Convolutional Neural Networks to achieve real-time, non-destructive evaluation of food breakdown patterns during mastication. We compared four different Residual Network (ResNet) backbone architectures (ResNet50, ResNet101, ResNeXt101, and Panoptic ResNet101) to determine optimal performance for food particle detection and bolus formation analysis. High-resolution images of peanut boluses across twelve chewing cycles were captured and analyzed using controlled experimental protocols with standardized mastication conditions. Results show that ResNet101 achieved the best accuracy and consistency in tracking chewing-related changes, providing a significant improvement over traditional method. The system successfully identified three distinct chewing stages: initial breakdown (cycles 1–3) with high particle count, gradual consolidation (cycles 4–8) with decreasing particle surface area, and final bolus formation (cycles 9–12) with 100% bolus composition. Independent validation through traditional particle size distribution measurements confirmed our computational findings, with median particle sizes decreasing from 0.47 mm to 0.27 mm across the mastication sequence. This computational methodology represents the first successful application of artificial intelligence for quantitative, non-destructive food bolus analysis, establishing a critical foundation for developing real-time monitoring systems in robotic in vitro chewing simulators.}, archive = {J_EAAI}, author = {Shiyi Ren and Bangxiang Chen and Jaspreet S. Dhupia and Martin Stommel and Weiliang Xu}, doi = {10.1016/j.engappai.2025.111884}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111884}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Analyzing food bolus formation during chewing through image segmentation}, volume = {160}, year = {2025}, }
• (2025). SkeletonDETR: A novel multimodal fusion based object detection framework for chemical safety applications. EAAI, 160, 111883. (www)

  @article{ , abstract = {Regarding the object detection algorithm in field operations in chemical plants, how to accurately detect objects carried and used by construction workers has become a crucial challenge in safety monitoring. Current object detection algorithms usually perform well for large objects but can easily ignore small objects, particularly under partial occlusion. Moreover, existing methods fail to recognize the significance of workers’ pose information during the construction process, which provides significant benefits for the detection task, especially when the targets for detection are closely associated with the construction workers. To solve this problem, we proposed a novel multimodal fusion based object detection framework, which can effectively use human pose information to improve the detection effect of small targets and occlusions. Furthermore, we propose a multimodal sampling module to fully utilize the features of different modalities to enhance the encoder’s ability to aggregate features. Compared with the baseline model, our proposed method achieves an 8.3% improvement in small object detection performance. Comprehensive experiments demonstrate that our proposed method outperforms existing efficient models, especially in field operation in chemical plants.}, archive = {J_EAAI}, author = {Yudi Tang and Bing Wang and Wangli He and Feng Qian and Zhen Liu}, doi = {10.1016/j.engappai.2025.111883}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111883}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SkeletonDETR: A novel multimodal fusion based object detection framework for chemical safety applications}, volume = {160}, year = {2025}, }
• (2025). Research on hull form optimization at multiple speeds based on machine learning and ship model experiments. EAAI, 160, 111882. (www)

  @article{ , abstract = {In order to improve the scientificity, efficiency and systematicness of ship form optimization, the multi-objective optimization research on the David Taylor Model Basin (DTMB) 5512 ship is carried out. First, the ship model experiment quantified the still water resistance of DTMB 5512 at six speeds at Froude number (Fr) as 0.25–0.40, demonstrating an almost linear resistance velocity relationship. Meanwhile, the DTMB 5512 ship is subjected to numerical simulations using the Computational Fluid Dynamics (CFD) method and the calculated results are compared with the experimental results. Then, Random Forest (RF)-based approximate models were developed for multi-speed resistance prediction, and verified its feasibility using Maximum Absolute Error (MAE). Finally, the parametric modeling method, the CFD method, and the optimization algorithm are integrated to construct a multi-objective optimization design system for ship forms. The resistance performance of the DTMB 5512 ship is optimized using the Multi-Objective Particle Swarm Optimization (MOPSO) algorithm. The results show that under the constructed hull form optimization framework, the optimized hull forms that meet the constraint conditions can be obtained. The total resistance of the obtained optimized ship at six speeds is reduced by 2.95 %, 4.44 %, 3.71 %, 5.22 %, 5.51 % and 4.83 % respectively. The research results indicate that the optimized hull forms with improved resistance performance can be obtained through the proposed methods, significantly enhancing the optimization efficiency. It also verifies the effectiveness of the random forest method in addressing the challenges of actual engineering optimization.}, archive = {J_EAAI}, author = {Jie Liu and Baoji Zhang and Lifen Hu and Junying Bi and Zheng Tian and Yingkai Dong}, doi = {10.1016/j.engappai.2025.111882}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111882}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on hull form optimization at multiple speeds based on machine learning and ship model experiments}, volume = {160}, year = {2025}, }
• (2025). Instance segmentation pipeline for etch pit detection and prismatic slip characterization on silicon carbide substrates. EAAI, 160, 111881. (www)

  @article{ , abstract = {Silicon carbide, a wide-bandgap semiconductor, is well-suited for high-frequency and high-temperature applications. The performance of devices fabricated from silicon carbide is negatively affected by extended defects such as dislocations. Therefore, it is crucial to measure and characterize these defects to ensure the reliability of the devices. The most commonly employed method for visualizing dislocations on the surface is defect-selective etching using molten potassium hydroxide. To detect and characterize various types of etch pits on the surface of single crystal substrates, a Mask Region-based Convolutional Neural Network (Mask R-CNN) is utilized. An inference pipeline based on slicing the images into overlapping tiles assures that large images of whole wafers can be processed. Masks generated by the network are further used to extract the dislocation line direction of basal plane dislocations, enabling the characterization of the extent of prismatic slip. The method performs well, even for overlapping etch pits, and can easily be integrated into existing characterization workflows, provided that segmentation masks of the etch pits are available.}, archive = {J_EAAI}, author = {Georg Holub and Sebastian Hofer and Thomas Obermüller and Elmar Rueckert and Lorenz Romaner}, doi = {10.1016/j.engappai.2025.111881}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111881}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Instance segmentation pipeline for etch pit detection and prismatic slip characterization on silicon carbide substrates}, volume = {160}, year = {2025}, }
• (2025). Artificial-intelligence assisted advanced exergy analysis of a preheat train in a petroleum refinery. EAAI, 160, 111880. (www)

  @article{ , abstract = {The heat exchanger network (HEN) is an essential component in the petroleum refinery, particularly within the crude distillation unit. It plays a critical role in minimizing furnace heat loads by harnessing the residual heat of product streams from the Crude Oil Distillation Unit (CDU). While this process helps in energy recovery, there is an ongoing effort to further optimize its efficiency. In this study, the quantification of exergy efficiency and exergy destruction within the HEN of the crude distillation unit under uncertain process conditions is conducted using an advanced exergy analysis technique in combination with artificial neural networks (ANN). Initially, an advanced exergy analysis of the HEN in the crude distillation unit under steady-state conditions was performed. Subsequently, an uncertainty of ±10% was incorporated into the process variables, converting the process model into the dynamic mode. From that dynamic model, a data set of 600 samples was generated. ANN models were developed using datasets to predict exergy efficiency, exergy destruction, unavoidable and avoidable exergy destruction, and modified exergy efficiency. The resulting performance indicators, with correlation coefficients (R), are 0.9923, 0.99266, 0.9829, 0.99243, and 0.99135 for overall exergy efficiency, overall exergy destruction, avoidable exergy destruction, unavoidable exergy destruction, and modified exergy destruction respectively. This study lays the groundwork for advancing Refinery 4.0, holding promising potential for future initiatives to enhance energy efficiency in the industry.}, archive = {J_EAAI}, author = {Asad Ayub and Iftikhar Ahmad and Hakan Caliskan and Hiki Hong}, doi = {10.1016/j.engappai.2025.111880}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111880}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial-intelligence assisted advanced exergy analysis of a preheat train in a petroleum refinery}, volume = {160}, year = {2025}, }
• (2025). Active learning with gaussian process regression for solving non-linear time-dependent partial differential equations. EAAI, 160, 111879. (www)

  @article{ , abstract = {The efficient and accurate solution of nonlinear time-dependent partial differential equations (PDEs) remains a challenging problem in various scientific and engineering domains. Traditional numerical methods often require fine-grained discretizations and become computationally expensive as the complexity of the PDE increases. This paper aims to present a novel approach for solving nonlinear time-dependent PDEs using active learning with Gaussian Process Regression (GPR). GPR provides a probabilistic prediction of the solution, which includes a mean (the predicted value) and uncertainty (variance). Active learning with GPR in solving PDEs is a data-efficient approach that strategically selects the most informative points for model training, thus reducing the overall computational burden. Nonlinearity in PDEs is addressed by using kernel functions, such as the Matern and Radial Basis kernels. We predict numerical solutions for four distinct PDEs, namely the Burger’s equation, the Allen–Cahn equation, the parabolic interface PDE (Stefan equation) and the Korteweg–de Vries equation. By integrating active learning with GPR, the proposed method intelligently chooses the most informative data points for training, optimizing the model’s accuracy and efficiency. Our numerical results reveal that active learning with GPR significantly reduces the number of data points required for accurate predictions, offering a substantial reduction in computational costs compared to conventional numerical methods. Furthermore, the accuracy of the predictions is maintained, even with a limited number of labeled samples.}, archive = {J_EAAI}, author = {Soumen Sinha and Neha Bharill and Om Prakash Patel and Mahipal Jetta}, doi = {10.1016/j.engappai.2025.111879}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111879}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Active learning with gaussian process regression for solving non-linear time-dependent partial differential equations}, volume = {160}, year = {2025}, }
• (2025). Kidney stone and tumor segmentation by analyzing medical images using deep learning technique. EAAI, 160, 111878. (www)

  @article{ , abstract = {The segmentation of kidney tumors is a critical activity in medical imaging since it aids in effective diagnosis, treatment, and follow-ups of renal disorders. However, the segmentation process faces challenges related to variability in the tumor's size, shape, and intensity, as well as the noise and artifacts in medical images. This study aims to address the challenge of designing an effective and automated Deep Neural Model (DNM) analysis for Computed Tomography (CT) images of kidney stones and tumor segmentation, which is more accurate, faster, and more efficient than current state-of-the-art models. The DNM utilizes the U-Net structure to extract cross-scale features from the CT images. The extracted features are further explored with the aid of a transformer model, which identifies and extracts local and global context features to enhance mask segmentation efficiency. The obtained results reveal a considerable enhancement in segmentation results, achieving an 8 % increase in the Dice similarity coefficient (DSC) compared to standard techniques. This approach primarily focuses on segmenting renal cell carcinoma, a pathology commonly associated with kidney tumors, and demonstrates strong potential to assist clinical diagnosis, surgical planning, and treatment monitoring in nephrology, contributing to improved assessment and management of chronic kidney diseases (CKD). The proposed DNM model increases the precision ratio by 98.89 %, the recall ratio by 97.12 %, the accuracy ratio by 98.43 %, the F1-score ratio by 98.5 %, and the IoU by 99.18 % compared to existing models.}, archive = {J_EAAI}, author = {Fangfang Ye and Congcong Liu and Jinming Wang and Qingrong Sun and Somia Asklany}, doi = {10.1016/j.engappai.2025.111878}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111878}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Kidney stone and tumor segmentation by analyzing medical images using deep learning technique}, volume = {160}, year = {2025}, }
• (2025). Multi-atlas functional and effective connectivity attention fusion method for autism spectrum disorder diagnosis. EAAI, 160, 111877. (www)

  @article{ , abstract = {The brain connectivity patterns, such as functional connectivity (FC) and effective connectivity (EC), describe the complex interaction patterns in brain networks. By combining FC and EC information, essential biomarkers for characterizing brain abnormalities can be further extracted. However, existing FC–EC fusion methods often process FC and EC networks independently, neglecting the intrinsic interrelationships between them. In addition, current FC–EC fusion methods rely on single segmentation atlas data, whereas multi-atlas data provides a more comprehensive display of brain information. To address these issues, we propose a novel multi-atlas dual branch graph convolution functional effective connectivity attention fusion (MaFECAF) method. It comprises three modules: multi-atlas data fusion, functional and effective connectivity construction, and functional and effective connectivity attention fusion. First, two atlases are fused using the cosine similarity matrix as weights. Next, we use the fused data to construct the FC and EC networks via spatio-temporal graph convolutional network and variational autoencoder parameterized by graph neural network architecture, respectively. Subsequently, the FC and EC networks are merged via a dynamic graph convolutional network and an adaptive self-attention feature extraction unit, followed by the application of the obtained features to autism spectrum disorder (ASD) classification. The experimental results have demonstrated the effectiveness of the proposed method and the benefits of integrating FC and EC. The proposed method achieved a 74.7% classification accuracy for ASD and outperformed state-of-the-art methods.}, archive = {J_EAAI}, author = {Minqi Yu and Jinduo Liu and Junzhong Ji}, doi = {10.1016/j.engappai.2025.111877}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111877}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-atlas functional and effective connectivity attention fusion method for autism spectrum disorder diagnosis}, volume = {160}, year = {2025}, }
• (2025). Neural network adaptive force control for pneumatic polishing end-actuator with external disturbances and full-state constrains. EAAI, 160, 111876. (www)

  @article{ , abstract = {In pneumatic polishing, the nonlinear, time-varying, and uncertain contact characteristics introduce significant modeling inaccuracies, posing substantial challenges to the realization of precise and robust force control. This paper proposes a neural-network adaptive force control strategy for a pneumatic polishing end-actuator under external disturbances and full-state constraints. To estimate the unmeasurable states and enhance the ant disturbance capability, a composite observer is developed to estimate the internal states and external disturbances in real time. Under the adaptive backstepping design framework, a radial-basis-function–neural-network–based adaptive learning mechanism is employed to approximate the nonlinear uncertainties, and a dynamic surface-control structure is introduced to avoid the complexity explosion in conventional recursive designs. Furthermore, a barrier Lyapunov function is integrated to ensure compliance with the full-state constraints throughout the control process. The convergence of the controller is verified through stability analyses, and the effectiveness and superiority of the control scheme is verified via experiments in four different polishing scenarios. The results show that the proposed control method achieves an average force tracking error less than 0.07 N and convergence time less than 2.35 s, showing higher control accuracy, faster transient response, and stronger robustness, than similar control algorithms.}, archive = {J_EAAI}, author = {Zhiguo Yang and Jiange Kou and Zhanxin Li and Wenbo Zhao and Yushan Ma and Yixuan Wang and Yan Shi}, doi = {10.1016/j.engappai.2025.111876}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111876}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural network adaptive force control for pneumatic polishing end-actuator with external disturbances and full-state constrains}, volume = {160}, year = {2025}, }
• (2025). Empirical method for developing a hyper-personalization artifact. EAAI, 160, 111875. (www)

  @article{ , abstract = {This research synthesized an artifact that applied Artificial Intelligence to enable telecommunication businesses to offer hyper-personalized products and services. The study was based on a database of the customers’ digital demography, and a range of telecommunication products and services the customer used. Relevant attributes were included in a Google Form to collect customer usage data. The data collected was screened for bad data and normality checked by conducting multivariable Mahalanobis outlier detection and normality tests. Outlier data was removed and multivariable normality of the data was ensured. Using the preprocessed database, several procedures were conducted to determine the best artificial intelligence algorithm for subsequent analysis, namely, the Logistic Model Tree algorithm. Using this algorithm and the customer digital demography dataset, the telecommunication business offerings for the customers were predicted with a 97.6 % accuracy. The proof of concept was developed using the Waikato Environment for Knowledge Analysis software. The artifact created was named Hypersona. Implemented within telecommunication systems, the model can be integrated into customer relationship management platforms allowing real-time adaptation to user needs. The methodology ensures feasibility by leveraging existing data infrastructures, while scalability is achieved through automated learning mechanisms that adapt to changing user environments. The research contributions lie in the applicability of real-time identification of changing personalized products and services. This research highlights the potential of artificial intelligence driven hyper-personalization in the telecommunications sector. Further research can be extended to other contemporary artificial intelligence methods and exploring the scaling of the artifact across diverse businesses.}, archive = {J_EAAI}, author = {Umapathy Sivan G. Murugasu and Anusuyah Subbarao}, doi = {10.1016/j.engappai.2025.111875}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111875}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Empirical method for developing a hyper-personalization artifact}, volume = {160}, year = {2025}, }
• (2025). Robust memory-based graph neural networks for noisy and sparse graphs. EAAI, 160, 111874. (www)

  @article{ , abstract = {Real-world graphs often suffer from structural noise and label sparsity, two critical challenges that severely impair the performance of graph neural networks (GNNs). While prior methods have focused on addressing either of these problems in isolation, their co-occurrence in practical scenarios remains largely unsolved. To address this challenge, we propose a robust memory-based GNN for noisy and sparse graphs that stores and updates node similarity information within a memory module to assist in predicting missing edges and eliminating noisy ones. This reconstruction densifies the graph structure, effectively mitigating the impact of noisy edges and alleviating the challenges posed by label sparsity through enhanced information propagation. Furthermore, the reconstructed graph adopts an edge regularization strategy that models the confidence of predicted edges and suppresses uncertain connections during training, thereby smoothing the label propagation for unlabeled nodes and improving the robustness of GNN training. Extensive evaluations conducted on real-world benchmark datasets, including Cora, Citeseer, and Pubmed, demonstrate that our proposed method, the robust memory graph neural network (RMGNN), significantly enhances GNN performance on noisy graphs with limited labeled nodes, with a notable performance boost of up to 17.8% on the Cora dataset. Our experimental analysis further confirms the effectiveness and efficiency of the proposed memory-based graph structure learning approach in the presence of edge noise and sparse labels, validating the robustness of the framework in complex graph scenarios.}, archive = {J_EAAI}, author = {Linling Jiang and Wenchang Zhang and Hua Wang and Fan Zhang}, doi = {10.1016/j.engappai.2025.111874}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111874}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust memory-based graph neural networks for noisy and sparse graphs}, volume = {160}, year = {2025}, }
• (2025). Strategy tuning of synthetic jet control over supercritical airfoil based on deep reinforcement learning. EAAI, 160, 111873. (www)

  @article{ , abstract = {In this article, the author couples Soft Actor-Critic (SAC), a deep reinforcement learning (DRL) algorithm, and synthetic jet (SJ) control, to explore the application of SAC algorithm in the field of active flow control (AFC). The author develops a framework of coupling DRL algorithms and computational fluid dynamics (CFD), in which a novel DRL environment is build based on ANSYS Fluent, and interacts with DRL agent using PyFluent kit as a bridge. In the case of two-dimensional (2D) cylinder laminar flow with Reynolds number (Re) = 100, SAC algorithm achieves 33.3% improvement of drag reduction compared with Proximal Policy Optimization (PPO) algorithm, another DRL algorithm, with 70% of training time consumption. The newly developed framework is adopted in tuning the strategy of SJ control over super critical airfoil. With the objective to enhance lift, agent's strategy gains 3.15% extra lift compared with the baseline which is not stimulated by SJ, and 29.03% improvement of lift enhancement compared with stable SJ frequency strategy. In another case involving a double-objective optimization, where both lift enhancement and drag reduction are considered as objectives, agent's strategy achieves 49.088 of time-averaged lift-drag ratio, which is 10.31% more than that of the baseline, and 1.58% – 14.7% improvement compared with stable SJ frequency strategy. With the foundation of the DRL environment introduced in this article, researchers can easily migrate other DRL algorithms into this framework and do further application in AFC.}, archive = {J_EAAI}, author = {Fangzhou Han and Jingying Wang and Mingrui Li and Chunhian Lee}, doi = {10.1016/j.engappai.2025.111873}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111873}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Strategy tuning of synthetic jet control over supercritical airfoil based on deep reinforcement learning}, volume = {160}, year = {2025}, }
• (2025). An on-line global–local defect detection framework for wide cold-rolled strip steel. EAAI, 160, 111872. (www)

  @article{ , abstract = {Wide cold-rolled strip steel possess distinctive characteristics, including a large surface area, high reflectivity, and multiple defect scales. These attributes have posed a significant challenge for achieving efficient and accurate online detection of small irregular defects in existing research. To address this, this paper proposes a Global–Local Defect Detection Framework (GLDDF) based on machine vision. The GLDDF comprises three parts: an image feature extractor, a global detector and a local detector. A feature extraction backbone network has been introduced to effectively capture multiple, subtle defect features on the steel strip surface. The global detector leverages normalizing flow to estimate the probability density of extracted features and quickly localize anomalous regions based on anomaly scores, effectively reducing computational overhead. And the local detector classifies and localizes defects to improve detection accuracy. The specialized equipment has been developed to overcome high reflection on the strip surface, allowing for on-line collection of high-definition images. A dataset of strip steel images has been produced to validate the algorithm’s effectiveness. The proposed method achieved a mean Average Precision (mAP) of 86.3% and a processing speed of 75.7 Frames Per Second (FPS) on the collected dataset, demonstrating a significant improvement in detection speed compared to existing methods. The method has been initially implemented on the cold-rolled strip steel production line of a partner company, yielding promising results.}, archive = {J_EAAI}, author = {Pan Jiang and Zhenying Xu and Wei Fan and Jin Zhang}, doi = {10.1016/j.engappai.2025.111872}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111872}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An on-line global–local defect detection framework for wide cold-rolled strip steel}, volume = {160}, year = {2025}, }
• (2025). Myocardial ischemic classification using a knowledge-guided polar transformer in two-dimensional echocardiography. EAAI, 160, 111871. (www)

  @article{ , abstract = {Myocardial ischemia, characterized by inadequate blood supply to the heart muscles, is critical to cardiovascular diseases. Timely and accurate identification of ischemic segments is essential for prompt intervention and patient care. This study developed a Transformer-based model to identify myocardial ischemia in left ventricle short-axis (LVSA) two-dimensional echocardiography (2DE) images where a novel Knowledge-Guided Polar Transformer (KGPT) was proposed that integrated the unique characteristics of 2DE images with the prior clinical knowledge. 305 patients (aged 57.6 ± 8.8 years) were selected and underwent transthoracic echocardiography within 1–3 days prior to invasive coronary angiography (ICA). With ICA and quantitative flow ratio as the gold standard of myocardial ischemia, the KGPT model was trained to classify the LVSA 2DE images as ischemia or non-ischemia by capturing spatial features in a radial orientation. Its performance was evaluated with five-fold cross-validation and receiver operating characteristic curve (ROC) analysis. It achieved an area under ROC (AUC) of 0.8326 ± 0.0906, with an accuracy of 79.50 ± 5.40 %, precision of 79.07 ± 6.70 %, recall of 80.79 ± 7.87 %, and F1 score of 78.43 ± 6.56 %. In comparison, the original Swin-Transformer model produced an AUC of 0.7011 ± 0.0334, accuracy of 70.20 ± 1.04 %, precision of 68.58 ± 3.12 %, recall of 63.21 ± 3.60 %, and F1 score of 63.13 ± 3.78 %. The differences were statistically significant (P < 0.05). The KGPT also demonstrated significantly superior performance to radiologists. It effectively classifies ischemic regions in 2DE images, presenting a promising tool for diagnosing myocardial ischemia. The integration of clinical knowledge with Transformer enhances the accuracy and reliability of ischemia classification, potentially revolutionizing the diagnosis and monitoring of myocardial ischemic diseases.}, archive = {J_EAAI}, author = {Ziwei Pang and Yi Du and Yanhui Guo and Shuang Chen and Bo Yu and Siqi Guo and Guo-Qing Du}, doi = {10.1016/j.engappai.2025.111871}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111871}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Myocardial ischemic classification using a knowledge-guided polar transformer in two-dimensional echocardiography}, volume = {160}, year = {2025}, }
• (2025). Robust bi-temporal cross-scene land cover map updating via curriculum-guided self-training and adversarial learning. EAAI, 160, 111870. (www)

  @article{ , abstract = {Bi-temporal land cover maps updating extends the traditional remote sensing image classification task to analyze land cover and land use (LC&LU) with unavailable ground truth and distribution shifts across the time scenarios. Difference-assisted knowledge transfer can alleviate the heavy annotation dependency on the target domain (newly collected image over the same region) to some extent. The inaccurate difference information obtained in an unsupervised manner often introduces a large number of mislabels for the training data on the target domain, while land cover transitions may contribute to the shortage of samples for specific ground categories on the target domain, i.e., imbalance between categories. This paper proposes a collaborative curriculum self-training framework, which is considerably robust to inaccurate difference knowledge transferring, to learn a more generalized classifier for the target domain. Specifically, curriculum guided adversarial learning consists of three players, i.e., curricula-classifier, curricula-generator, and discriminator, which aims at efficiently and reliably adapting to the negative effects of incorrect cross-domain mapping and the imbalanced long-tail distribution on self-training framework. Sample-level and task-level curriculum constraints are designed to collaboratively facilitate sample selection and task learning. The experimental results on four datasets (two pairs of bi-temporal images) show that the above method is able to perform robust learning and achieve remarkable classification performance. The code is released at https://github.com/bling2beyond/Curricula-GAN .}, archive = {J_EAAI}, author = {Zhao Wang and Yue Zhao and Maoguo Gong and Hao Li and Gaogao Liu and Jianlong Tang}, doi = {10.1016/j.engappai.2025.111870}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111870}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust bi-temporal cross-scene land cover map updating via curriculum-guided self-training and adversarial learning}, volume = {160}, year = {2025}, }
• (2025). Physics-informed neural networks for control of single-phase flow systems governed by partial differential equations. EAAI, 160, 111869. (www)

  @article{ , abstract = {The modeling and control of single-phase flow systems governed by Partial Differential Equations (PDEs) present challenges, especially under transient conditions. In this work, we extend the Physics-Informed Neural Nets for Control (PINC) framework, originally proposed for modeling and control of Ordinary Differential Equations (ODEs) without the need of any labeled data, to the PDE case, particularly to single-phase compressible flows, integrating neural networks with physical conservation laws. The PINC model for PDEs is structured into two stages: a steady-state network, which learns equilibrium solutions for a wide range of control inputs, and a transient network, which captures dynamic responses under time-varying boundary conditions. We propose a simplifying assumption that reduces the dimensionality of the spatial coordinate regarding the initial condition, allowing the efficient training of the PINC network. This simplification enables the derivation of optimal control policies using Model Predictive Control (MPC). We validate our approach through numerical experiments, demonstrating that the PINC model, which is trained exclusively using physical laws, i.e., without labeled data, accurately represents flow dynamics and enables real-time control applications. The results highlight the PINC’s capability to efficiently approximate PDE solutions without requiring iterative solvers, making it a promising alternative for fluid flow monitoring and optimization in engineering applications.}, archive = {J_EAAI}, author = {Luis Kin Miyatake and Eduardo Camponogara and Eric Aislan Antonelo and Alexey Pavlov}, doi = {10.1016/j.engappai.2025.111869}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111869}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed neural networks for control of single-phase flow systems governed by partial differential equations}, volume = {160}, year = {2025}, }
• (2025). Bayesian estimation of unified generalized process capability index for logistic-exponential distribution under progressive type II censoring scheme. EAAI, 160, 111868. (www)

  @article{ , abstract = {A unified generalized process capability index, denoted by C p y ( u , v ) for normal as well as non-normal, continuous as well as discrete random variables and unilateral as well as bilateral specifications is recently proposed by Saha et al. (2022). The present article attempts to study the Bayesian inferences of the capability index C p y ( u , v ) for type-II progressive censored sample under five different loss functions where the underlying distribution is logistic-exponential. Since, the obtained Bayes estimators are not in closed form. Thus, the Markov-Chain-Monte-Carlo technique has been efficiently used here to have the approximate solution for the index C p y ( u , v ) . The performance of the Bayes estimates is evaluated in terms of posterior risks and the highest posterior density credible intervals in terms of average width and corresponding coverage probability respectively using extensive simulation studies. The risk of the estimators derived under different loss functions are decreases with the increasing percentage of effective sample size m for the fixed values of n . Additionally, among the loss functions under consideration, no such specific trend is noticed for the Bayes estimators with the same choices of censoring parameters, although the differences among the average posterior risks reported under the considered loss functions are very small in magnitude. Furthermore, the average widths and coverage probabilities are also computed, and it is observed that the average widths decrease with an increase in n , m , and the coverage probabilities approaches its nominal value. Lastly, a real data set example pertaining to the electronic industry has been considered to illustrate its practical applicability.}, archive = {J_EAAI}, author = {Mahendra Saha and Abhimanyu Singh Yadav and Amartya Bhattacharya and Sudhansu S. Maiti and Arindam Gupta}, doi = {10.1016/j.engappai.2025.111868}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111868}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bayesian estimation of unified generalized process capability index for logistic-exponential distribution under progressive type II censoring scheme}, volume = {160}, year = {2025}, }
• (2025). Assessment of a family of recurrent neural network models for flood susceptibility mapping: An explainable glass-box approach. EAAI, 160, 111867. (www)

  @article{ , abstract = {Owing to climate change, the frequency of flood events has increased, causing substantial harm to society. Consequently, identifying flood-prone areas is a fundamental step toward effective flood management and mitigation. This study aims to compare recurrent neural network (RNNs) deep learning families for flood susceptibility mapping in the Gorganrood watershed in the Golestan Province, a hotspot for floods area in Iran. In the present study, an Elman-RNN model was developed, and the results were compared with those of regular RNN, gated recurrent unit (GRU), and long short-term memory (LSTM) models. Geodata comprising 230 historical flood events collected by analyzing Sentinel-1 and 14 geo-environmental characteristics were considered for the modeling. The SHapley Additive exPlanations (SHAP) approach was used to convert the black box results to an explainable glass box. Finally, the effectiveness of the models was assessed using the area under the receiver operating characteristic curve (AUC). The SHAP results revealed that altitude, distance from rivers, normalized differences in vegetation index, rainfall, and drainage density were the most important factors influencing flooding in the Gorganrood Watershed. The validation results revealed that the Elman RNN, which had an AUC testing of 0.969, possessed the highest generalization power, followed by the GRU (AUC testing = 0.952), LSTM (AUC testing = 0.947), and RNN (AUC testing = 0.943) models. The majority of the flood-prone areas were located downstream of the watershed, and approximately 19.92 % of the Gorganrood Watershed was very susceptible to future flooding. These results demonstrate the superiority of the Elman RNN model over other recurrent models, making it a practical and reliable flood modeling tool.}, archive = {J_EAAI}, author = {Shadi Maddah and Khabat Khosravi and Changhyun Jun and Sayed M. Bateni and Dongkyun Kim and Shunlin Liang}, doi = {10.1016/j.engappai.2025.111867}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111867}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Assessment of a family of recurrent neural network models for flood susceptibility mapping: An explainable glass-box approach}, volume = {160}, year = {2025}, }
• (2025). Defect detection and data generation of shunt between three layers of resistance spot welding based on deep convolutional generative adversarial network and conditional variational autoencoder. EAAI, 160, 111866. (www)

  @article{ , abstract = {Resistance spot welding (RSW), as a widely used joining technology in the automotive industry, especially in the welding process of three-layer plates, makes it difficult to concentrate the welding current to the target area due to the inter-plate shunt effect, which reduces the quality of the formation of the molten nucleus and increases the incidence of weld defects. To cope with this problem, this paper proposes a defect generation and classification method based on a Deep Convolutional Generative Adversarial Network-Conditional Variational Autoencoder (DCGAN-CVAE), which is used to generate synthetic defect data and augment limited real-world samples. The optimal welding process parameters are found through one-way experiments on welding current, welding time, and electrode pressure combined with Analysis of Variance and response surface analysis. Based on the optimal parameters, we conduct inter-plate shunt experiments, collect dynamic resistance signals, generate and expand defect samples through the DCGAN-CVAE generation model, and finally verify the classification of the generated samples using the Bidirectional Long Short-Term Memory network-One Dimensional Convolutional Neural Network (BiLSTM-1DCNN) classifier. Experimental results show that the generated data effectively improves the accuracy of defect detection, and the generated data have high similarity with the real data, which can provide data support for quality control in actual production.}, archive = {J_EAAI}, author = {Haofeng Deng and Xiangdong Gao and Dongfang Zhang and Wuqi Lu and Pengyu Gao and Yanxi Zhang}, doi = {10.1016/j.engappai.2025.111866}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111866}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Defect detection and data generation of shunt between three layers of resistance spot welding based on deep convolutional generative adversarial network and conditional variational autoencoder}, volume = {160}, year = {2025}, }
• (2025). Data-driven approaches for surface quality monitoring and prediction based on heterogeneous multi-channel signal fusion in hard part machining. EAAI, 160, 111865. (www)

  @article{ , abstract = {Data-driven systems have demonstrated significant value in real-time surface roughness evaluation and prediction for machining processes. This study presents a comprehensive methodology for evaluating three core elements critical to decision-making in machining monitoring: sensor-based dynamic signal selection and processing, sensory fusion scenarios, and Machine Learning (ML) models. Unlike previous research that often analyzes these elements in isolation, this study emphasizes their collective impact on unified datasets. The proposed methodology is validated using experimental data from machining trials on two distinct machines during the finishing process of hard turning. Hard turning is a widely employed finishing operation in manufacturing that directly affects dimensional accuracy, surface integrity, and surface finish, which are key characteristics of machined parts. Surface roughness, as an essential indicator of surface quality, plays a pivotal role in the functional performance of end products, necessitating accurate monitoring and assessment throughout the process. To optimize performance, Bayesian Optimization was employed for automatic hyperparameter tuning, facilitating efficient exploration of optimal parameters. The predictive capabilities of trained ML models were subsequently evaluated using performance metrics and quantitatively assessed for uncertainty through Prediction Intervals (PIs), calculated via non-parametric Kernel Density Estimation. Results indicate that the Wavelet Packet Transform method significantly enhances the predictive performance across all ML models. Among the evaluated models, Support Vector Regression and K-Nearest Neighbors demonstrated superior predictive accuracy and minimal uncertainty across all signal processing methods. This work provides actionable guidelines for researchers and manufacturers in selecting optimal combinations of methodologies for developing accurate, reliable, and generalizable data-driven surface quality monitoring and prediction systems, particularly for hard part turning applications.}, archive = {J_EAAI}, author = {Yaoxuan Zhu and Tomas Österlind and Amir Rashid and Andreas Archenti}, doi = {10.1016/j.engappai.2025.111865}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111865}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven approaches for surface quality monitoring and prediction based on heterogeneous multi-channel signal fusion in hard part machining}, volume = {160}, year = {2025}, }
• (2025). A vector-based end-effector tracking and obstacle avoidance control for wheeled mobile manipulators under multi-joint position and velocity constraints. EAAI, 160, 111864. (www)

  @article{ , abstract = {Obstacle avoidance is generally an additional task for a wheeled mobile manipulator (WMM) to perform end-effector tracking tasks that ensure the safety of the WMM. The WMM obstacle avoidance ensures that collisions with objects or people are avoided during operation, thus ensuring safety in the work situation and increasing efficiency in accomplishing tasks. A vector inequality-based obstacle avoidance (VIBOA) scheme is developed and analyzed, which addresses the inability of the WMM to safely avoid obstacles while completing trajectory tracking. Compared with traditional obstacle avoidance schemes, the proposed VIBOA scheme avoids the sharp stops caused by obstacle avoidance of the WMM. It enables the WMM to achieve accurate and safe obstacle avoidance, joint physical constraints and end-effector tasks. The VIBOA scheme based on time-varying quadratic programming is eventually transformed into a linear variational inequality, which is solved by linear-variational-inequality primal dual neural network (LVI-PDNN). The validity, accuracy and superiority of the LVI-PDNN-based VIBOA scheme are demonstrated through simulation comparisons with conventional schemes. Finally, experiments are conducted on a practical WMM platform to verify the physical realizability and effectiveness of the LVI-PDNN-based VIBOA scheme.}, archive = {J_EAAI}, author = {Shijun Tang and Yuzhe Fei and Xingtian Xiao and Zhongbo Sun and Yunfeng Hu}, doi = {10.1016/j.engappai.2025.111864}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111864}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A vector-based end-effector tracking and obstacle avoidance control for wheeled mobile manipulators under multi-joint position and velocity constraints}, volume = {160}, year = {2025}, }
• (2025). Multi-task cyclical consistency learning based medical image segmentation. EAAI, 160, 111863. (www)

  @article{ , abstract = {Image segmentation and image reconstruction are two of the most prominent tasks in current computer vision research, with numerous advanced models contributing to increasing task accuracy. However, most of the existing models for these tasks are trained independently, overlooking the complementary potential of these tasks during the training process. In this work, we propose a progressive segmentation refinement strategy by designing a dual-stage joint multi-task consistency learning model based on the Transformer, effectively combining the image segmentation and image reconstruction tasks to achieve fine-grained segmentation of medical images. Specifically, we present a multi-stage joint multi-task consistency learning network, which includes a shared transformer encoder and two independent transformer decoders. These decoders are responsible for image segmentation and lesion region reconstruction tasks, respectively. The image reconstruction task aids the model in learning the feature representations of lesion regions, helping to refine the segmentation boundaries and improve segmentation precision. In addition, the model leverages semi-supervised learning by computing loss on the reconstructed masked lesion regions, further enhancing the generalizability of the model. Experimental results on the Kvasir-SEG, Kvasir-Capsule, ISIC 2016, and ISIC 2018 datasets demonstrate that our method outperforms other state-of-the-art methods.}, archive = {J_EAAI}, author = {Le Han and Jianan Zhang and Yan Hu and Xueyu Liu and Guanghui Yue and Mingqiang Wei and Yongfei Wu}, doi = {10.1016/j.engappai.2025.111863}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111863}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-task cyclical consistency learning based medical image segmentation}, volume = {160}, year = {2025}, }
• (2025). Deep reinforcement learning–based collision avoidance strategy for multiple unmanned aerial vehicles. EAAI, 160, 111862. (www)

  @article{ , abstract = {Unmanned aerial vehicles (UAVs) are used prevalently to perform various tasks related to robot control. UAVs are extensively used to perform different tasks because they are lightweight and have high mobility. Deep reinforcement learning is a common technique used for automatically training UAVs to complete different tasks. This study used a two-stage training process to train three UAVs and enable them to pass through four goals consecutively. The UAVs had to avoid dynamic and static obstacles while maintaining a certain distance from each other. The first stage involved using a self-defined reward function for the UAVs to learn to avoid obstacles. The second stage involved using the tree-structured Parzen estimator to optimize several hyperparameters, generating environments for training the UAVs to complete tasks related to the first stage while maintaining adequate distances from each other. The method proposed in this paper can help UAVs to learn to pass through multiple goals, avoid obstacles, and automatically adjust their distance from other UAVs. This study also compared different algorithms and used proximal policy optimization (PPO), advantage actor critic (A2C), and soft actor critic (SAC) to train the UAVs. The results indicated that the UAVs trained using A2C or PPO completed the relevant tasks easily, with the UAVs trained using A2C completing the tasks faster. By contrast, the UAVs trained using SAC were unable to complete most tasks. The proposed method was experimentally verified and has potential for improving the coordination of UAV control in the future.}, archive = {J_EAAI}, author = {Ping-Huan Kuo and Kuan-Lin Chen and Yu-Sian Lin and Yu-Chih Chiu and Chao-Chung Peng}, doi = {10.1016/j.engappai.2025.111862}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111862}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep reinforcement learning–based collision avoidance strategy for multiple unmanned aerial vehicles}, volume = {160}, year = {2025}, }
• (2025). Statistical significance of cluster membership for categorical data. EAAI, 160, 111861. (www)

  @article{ , abstract = {Clustering algorithms partition data samples into distinct groups in an unsupervised manner, which requires subsequent validation. In post-hoc cluster analysis, clustering quality is typically evaluated at the cluster level, with a focus on metrics such as intra- and inter-cluster distances. However, evaluation at the sample level (i.e., cluster membership) is often overlooked. To assess whether a given sample is correctly assigned to its respective cluster, it is crucial to consider the inevitable effects of random assignments or noisy samples. Unfortunately, the statistical cluster membership evaluation is a largely underexplored problem, with almost no previous research efforts in this direction. In this paper, we propose a new method for assessing the statistical significance of cluster membership for categorical data. Under the null hypothesis that there is no association between one sample and a cluster, we can employ the Fisher’s exact test to derive a p -value with respect to each attribute. By combining p -values from all attributes via meta-analysis, we can obtain a consensus p -value to quantify the cluster membership, i.e., if the sample is statistically associated with the target cluster. To show the benefit of such a cluster membership evaluation technique, we deploy our algorithm to several applications, ranging from cluster validation to cluster refinement and enhancement. Experimental results on real categorical data sets demonstrate the rationale and effectiveness of our method, including its potential to improve the overall clustering accuracy of both classical and state-of-the-art clustering algorithms.}, archive = {J_EAAI}, author = {Lianyu Hu and Zerun Li and Junjie Dong and Mudi Jiang and Zengyou He}, doi = {10.1016/j.engappai.2025.111861}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111861}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Statistical significance of cluster membership for categorical data}, volume = {160}, year = {2025}, }
• (2025). An eXplainable artificial intelligence framework to predict marine scrubbers performances. EAAI, 160, 111860. (www)

  @article{ , abstract = {This study presents an eXplainable Artificial Intelligence (XAI) framework to predict the performance of marine scrubbers used for sulfur dioxide ( SO 2 ) removal from marine diesel engine flue gases. Using an aggregated dataset from a roll-on/roll-off (Ro-Ro) cargo ship equipped with an open-loop scrubber, combined with satellite data, the study constructs and evaluates multiple artificial intelligence models, including ensemble models, which were benchmarked against each other using standard regression metrics such as the coefficient of determination (R 2 ), mean absolute error (MAE), and mean squared error (MSE). Results achieve high accuracy R 2 > 0 . 92 and offer insights for optimizing scrubber operations. Nevertheless, artificial intelligence models lack transparency. To overcome this problem, this research integrates post-hoc explainability techniques to elucidate the contributions of various features to model predictions, thereby enhancing interpretability and reliability. The integration of SHapley Additive exPlanations (SHAP) and Explain Like I’m 5 (ELI5) not only confirmed the consistency of feature importance rankings (e.g. seawater acidity level, SO 2 inlet concentration, outlet temperature) but also aligned with the physical-chemical principles of SO 2 absorption. Quantitative comparisons with theoretical expectations demonstrated the reliability of the XAI insights, enhancing both model transparency and interpretability. This can improve the current capability of designing scrubber units by defining more efficient and less expensive options for environmental regulation compliance.}, archive = {J_EAAI}, author = {Luigi Piero Di Bonito and Lelio Campanile and Mauro Iacono and Francesco Di Natale}, doi = {10.1016/j.engappai.2025.111860}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111860}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An eXplainable artificial intelligence framework to predict marine scrubbers performances}, volume = {160}, year = {2025}, }
• (2025). A transformer-based framework for software vulnerability detection using attention-driven convolutional neural networks. EAAI, 160, 111859. (www)

  @article{ , abstract = {In the realm of software systems and quality assurance, vulnerability identification has become a critical concern in today’s connected world. Vulnerabilities not only make systems less effective, but they also pose significant risks to user privacy and data integrity. Although automated vulnerability detection has seen considerable progress, existing approaches frequently struggle to accurately model syntactic and semantic code relationships. This deficiency leads to undetected vulnerabilities and false positives, thereby undermining the efficacy of software security protocols. To overcome these limitations, we propose a new vulnerability detection technique called CodeGATNet (Code-Gated Attention using Convolutional Features). Our deep learning model is designed to handle static vulnerability detection in source code to enable batch-level examination of codebases. CodeGATNet operates in two phases: Static Code Embedding Generation (SCEG) and Convolutional Attention Network for Feature Refinement (CAN-FR). SCEG utilizes fine-tuned CodeBERT embeddings (a pretrained transformer model for programming languages), which are task-specifically optimized to improve the pretrained model’s capacity to capture application-relevant semantic patterns while maintaining its general code comprehension capabilities. CAN-FR uses a hybrid architecture for feature extraction and refinement combining a gated attention mechanism with one-Dimensional Convolutional Neural Network (1D-CNN). This hybrid approach helps the model to efficiently capture present global contextual relationships in the source code as well as local structural patterns. Extensive experimental evaluations on three large-scale (C/C++) real-world datasets and the results show that CodeGATNet considerably outperforms leading models, obtaining accuracy enhancements of 18.05%, 28.14%, and 13.06%, alongside F1-score improvements of 7.83%, 18.28%, and 13.0%.}, archive = {J_EAAI}, author = {Abdelkarim Smaili and Yanan Zhang and Djamel Eddine Mekkaoui and Mohamed Amine Midoun and Mohamed Zakariya Talhaoui and Meryem Hamidaoui and Weiqiang Kong}, doi = {10.1016/j.engappai.2025.111859}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111859}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A transformer-based framework for software vulnerability detection using attention-driven convolutional neural networks}, volume = {160}, year = {2025}, }
• (2025). Reconstructing three-dimensional conductivity distribution of in-situ maize ears using frequency-enhanced residual encoder-decoder network. EAAI, 160, 111858. (www)

  @article{ , abstract = {Monitoring water content distribution in in-situ maize ears is crucial for agricultural cultivation and crop research, with impedance dynamics correlated to water content changes. However, the combined effects of complex maize ear structure, unstable contact impedance, and diverse environmental noise significantly reduce the quality of measurement signals, posing major challenges for reconstructing three-dimensional (3D) conductivity distribution. Therefore, we propose a 3D conductivity absolute reconstruction model based on a spectrum-enhanced residual encoder-decoder (FRED-Net), enabling stable monitoring of impedance changes in in-situ maize ears in a greenhouse. FRED-Net combines residual blocks and encoder-decoder architectures, utilizing skip connections to preserve low-level features, effectively addressing gradient vanishing and information loss while maintaining low computational complexity. To cope with greenhouse environmental interference, FRED-Net incorporates an information fusion method based on spectral characteristics and volatility features, combined with a noiseless dataset training mode to enrich model input variability. Results from standard water tank simulation experiments indicate that FRED-Net exhibits ideal reconstruction accuracy and noise robustness compared to existing 3D absolute reconstruction algorithms, with a structural similarity index of 0.9970, root mean square error of 0.0019, relative error of 0.0199, and coefficient of determination of 0.9610. Imaging experiments on in-situ maize ears in the greenhouse successfully captured the dynamic characteristics of conductivity changes with water content during growth and the significant differences between tissue structures, providing an effective absolute imaging method for non-invasive visualization and measurement of the physiological state of maize ears during the growing period.}, archive = {J_EAAI}, author = {Hai-Ying Zheng and Bing-Zhou Chen and Yang Xiang and Ke-Lei Xia and Yu-Bin Lin and Jin-Hang Liu and Jia-Jia Zhang and Liu-Deng Zhang and Zhong-Yi Wang and Lan Huang}, doi = {10.1016/j.engappai.2025.111858}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111858}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reconstructing three-dimensional conductivity distribution of in-situ maize ears using frequency-enhanced residual encoder-decoder network}, volume = {160}, year = {2025}, }
• (2025). Anomaly detection in distributed systems based on spatio-temporal causal inference. EAAI, 160, 111857. (www)

  @article{ , abstract = {Detecting anomalies in distributed systems is a complex problem owing to the nature and size of these systems. By utilizing artificial intelligence (AI), this study introduces a new anomaly detection method that combines spatio-temporal causal inference through a Structural Causal Model (SCM) framework of Granger causality as the temporal method for improved performance and interpretability. First, we provide a Spatio-Temporal Causal Model (STCM) that captures distributed systems’ temporal and spatial dependencies. We then generalize an anomaly scoring method that quantifies each event with respect to its causal relationships. These techniques have been integrated into a unified framework, which we refer to as Integrated Causal Anomaly Detection (ICAD), and are optimized for both computational efficiency and scalability to large systems through approaches including parallelization and graph pruning. To evaluate the efficiency and robustness of ICAD, we perform extensive experimentation on synthetic and real-world datasets and compare it against baseline approaches such as Isolation Forest, Long Short-Term Memory (LSTM)-based Autoencoders, and Graph Neural Networks. Experimental results show that ICAD achieves an average F1 -score of 0.92 across three datasets, outperforming baselines by up to 11 percentage points (e.g., 0.89 vs. 0.78 for Isolation Forest on CLOUD-1 ). ICAD maintains a false-positive rate as low as 0.01 under 20% system load (compared with 0.05 for baselines), demonstrates superior robustness with 82% F1-score retention under noise ( σ = 0 . 5 ) versus 54% for traditional methods, and attains 87.5% root-cause identification accuracy with concise explanations (average causal path length of 3.2).}, archive = {J_EAAI}, author = {Chunmao Jiang and Wei Wu}, doi = {10.1016/j.engappai.2025.111857}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111857}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Anomaly detection in distributed systems based on spatio-temporal causal inference}, volume = {160}, year = {2025}, }
• (2025). A crude oil price ensemble forecasting system based on outlier correction and adaptive error-compensation strategy. EAAI, 160, 111856. (www)

  @article{ , abstract = {Crude oil price forecasting provides reliable information for energy policy formulation, corporate strategic planning, and investment decision-making. Its accuracy is often hampered by data outliers and inherent prediction errors, but many existing studies may overlook the significance of outlier preprocessing and error correction and often lack sufficient transparency. To address this issue, we propose a crude oil price ensemble forecasting system based on outlier correction and an adaptive error-compensation strategy. In this system, to identify and correct outliers in the original data, a data-preprocessing module is developed to eliminate negative impacts on model training and forecasting performance. In addition, an effective ensemble modelling module is proposed to combine the results of the diverse crude oil forecasting sub-models provided by the designed multiple-model forecasting module, which demonstrably outperforms other advanced ensemble methods and can perform better than any individual sub-model by fully harnessing their individual strengths. Another key advancement of this study is the proposal of an adaptive error-compensation strategy, which innovatively addresses not only how to correct errors in ensemble results but also whether such corrections are necessary. Moreover, the system incorporates an interpretable analysis module, yielding crucial insights and offering practical implications for decision-makers by enhancing the transparency of the ensemble system. Empirical studies show that the developed system has an average root mean squared error of 3.0810, representing an improvement of up to 11.53 % over the best sub-model. The system consistently outperforms the benchmark models and can be a promising solution for crude oil price forecasting.}, archive = {J_EAAI}, author = {Wendong Yang and Weicheng Ma and Bo Zeng and Yan Hao}, doi = {10.1016/j.engappai.2025.111856}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111856}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A crude oil price ensemble forecasting system based on outlier correction and adaptive error-compensation strategy}, volume = {160}, year = {2025}, }
• (2025). A federated fairness-aware incentive mechanism for medical image classification. EAAI, 160, 111855. (www)

  @article{ , abstract = {Despite the promising progress in federated learning (FL), two major challenges persist when applying FL in real-world medical scenarios: data heterogeneity and class imbalance. Previous efforts to tackle these challenges have primarily focused on local training or global aggregation, often assuming that all clients are willing participants. However, this assumption may not hold in real medical scenarios, where clients act based on rational self-interest. In this paper, we propose a Fed erated F airness-Aware I ncentive M echanism (FedFIM) to tackle these problems. FedFIM consists of two components: S erver-driven M arginal I ncentive (SMI) and P rototype-based S hapley E stimation (PSE). Specifically, SMI ensures the data quality of participants to mitigate the performance degradation caused by data heterogeneity, while PSE leverages prototypes for more fine-grained contribution evaluations. Furthermore, based on these contribution evaluations, a prototype-level aggregation scheme is designed to alleviate the impact of class imbalance. Extensive experiments on five medical image classification tasks demonstrate that FedFIM achieves superior accuracy compared to seven popular benchmark methods, particularly in complex multi-classification scenarios.}, archive = {J_EAAI}, author = {Chunling Chen and Haiwei Pan and Kejia Zhang and Zhe Li and Fengming Yu}, doi = {10.1016/j.engappai.2025.111855}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111855}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A federated fairness-aware incentive mechanism for medical image classification}, volume = {160}, year = {2025}, }
• (2025). Exploring you only look once v8 and v9 for efficient airplane detection in very high resolution remote sensing imagery. EAAI, 160, 111854. (www)

  @article{ , abstract = {Automatic airplane detection from satellite images using deep learning methods produces valuable geospatial information for a wide range of applications, including aviation safety, defence, airport and disaster management. You Only Look Once (YOLO) models have been widely used for various geospatial tasks; however, their application to airplane detection in very high-resolution (VHR) remote sensing imagery, particularly YOLOv8 and YOLOv9, remains underexplored. This study aims to assess the performance of YOLOv8 and YOLOv9 architectures in the context of airplane detection using High Resolution Planes (HRPlanes) dataset. First, we examine the impact of various hyperparameters on the performance of YOLOv8 models to propose optimal hyperparameter and model variant combinations. Second, we compare the best-performing YOLOv8 configurations with their YOLOv9 counterparts to evaluate potential improvements. Third, we assess the generalizability and transferability of the top-performing models by testing them across independent airplane detection datasets. Lastly, we perform an operational assessment of inference performance by analyzing trade-offs between network size, input image resolution and processing time. The optimal performance was achieved with the YOLOv8x model using 960x960 network size and data augmentation, resulting in 98.99 % F1-Score, 99.12 % Precision, 98.86 % Recall, 99.35 % Mean Average Precision (mAP)50, and 89.82 % mAP50-95. YOLOv9e achieved comparable performance with fewer parameters (57.3 vs. 68.2 million) and lower computational cost (189.0 vs. 257.8 giga floating point operations per second (GFLOPS)), offering up to a 27 % reduction in computational cost. These findings highlight the practical potential of both YOLOv8 and YOLOv9 for high-precision airplane detection in VHR remote sensing imagery. The HRPlanes dataset and model weights are publicly available at: https://github.com/RSandAI/Efficient-YOLO-RS-Airplane-Detection .}, archive = {J_EAAI}, author = {Doğu İlmak and Tolga Bakirman and Elif Sertel}, doi = {10.1016/j.engappai.2025.111854}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111854}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploring you only look once v8 and v9 for efficient airplane detection in very high resolution remote sensing imagery}, volume = {160}, year = {2025}, }
• (2025). A unified multi-task model for leaf disease region detection and segmentation. EAAI, 160, 111853. (www)

  @article{ , abstract = {Crop diseases challenge agricultural production. Pinpointing disease spots, assessing infection areas, and gauging infection severity are crucial for effective disease control. However, lesion variations, blurred boundaries, and small, dense lesions make precise detection and segmentation difficult. This paper presents an end-to-end unified multi-task model based on Detection Transformer (DETR) for leaf disease region detection and segmentation. It integrates Convolutional Neural Networks (CNNs) and Transformer, uses the Contextual Transformer Network (CoTNet) for feature extraction, and incorporates innovative mechanisms like box-attention and reference window update. Additionally, we have devised a novel instance segmentation head. This head effectively addresses the misclassification issue between minute disease spots and leaf surfaces. Experiments show the model achieves Average Precision ( AP ). The Average Precision of Bounding Box ( AP box ) of 73.9 %, the Average Precision of Mask ( AP mask ) of 68.2 %, the Average Precision of Small-sized Bounding Box ( AP s box ) of 29.0 %, and the Average Precision of Small-sized Mask ( AP s mask ) of 27.1 % for four diseases, with detection recall reaching 76.8 % and segmentation recall reaching 73.4 %. Meanwhile. The model architecture demonstrates practical feasibility with 40.1 million (m) parameters and 169 Giga Floating-point Operations Per Second (GFLOP) computational complexity. The accuracy of disease grading reaches 92.07 %. In this study, a model based on Artificial Intelligence (AI) was implemented to address the challenges in leaf disease detection and segmentation. The proposed model, which integrates object detection and instance segmentation tasks, can be applied to accurately identify and grade leaf diseases, providing support for disease control strategies in agriculture.}, archive = {J_EAAI}, author = {Tian Zhang and Yanfeng Lu and Chenshuang Li and Rundong Hong and Hengqiang Su and Ce Yang and Ning Yang and Haiying Zhang and Changji Wen}, doi = {10.1016/j.engappai.2025.111853}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111853}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A unified multi-task model for leaf disease region detection and segmentation}, volume = {160}, year = {2025}, }
• (2025). Meta-adversarial transfer learning based on a dual-channel transformer network for aircraft engine remaining useful life prediction. EAAI, 160, 111852. (www)

  @article{ , abstract = {Meta-transfer learning (MTL) is increasingly used for remaining useful life (RUL) of aircraft engines across operating conditions because it enables efficient knowledge transfer under limited data. However, existing MTL-based RUL models still suffer from domain shift because they transfer irrelevant features from pre-training tasks and overlook the joint influence of temporal patterns and multi sensor interactions on RUL. To address these challenges, we propose a meta-adversarial transfer learning method based on a dual-channel Transformer network (MAT-DCT). Specifically, during meta-training, we insert a gradient reversal layer into the inner loop so the extractor and regressor co-learn while purging pre-training noise; during meta-adaptation, we add a second adversarial mechanism in which the extractor contests a domain discriminator, enabling target feature alignment under few-shot conditions. Furthermore, we construct a dual-channel network; One channel emphasizes inter-sensor interactions while the other captures temporal evolution. In each channel, gated convolutions distill subtle degradation cues and feed them into the Transformer, which weaves them into a coherent global context. Experiments on the aircraft turbofan engine dataset across twelve cross domain tasks show that MAT-DCT reduces root mean square error by about 24 % and Score function by 49 % compared to baseline models, confirming its superiority for few-shot RUL prediction under diverse operating conditions.}, archive = {J_EAAI}, author = {Zhiwu Shang and Tianchu Pang and Cailu Pan and Leyi Yao and Zifei Wang}, doi = {10.1016/j.engappai.2025.111852}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111852}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Meta-adversarial transfer learning based on a dual-channel transformer network for aircraft engine remaining useful life prediction}, volume = {160}, year = {2025}, }
• (2025). Cost-effective and real-time landslide monitoring method based on ultra-wideband using ultra-wideband transformer neural network. EAAI, 160, 111851. (www)

  @article{ , abstract = {Landslides rank among the most destructive natural phenomena, posing substantial risks to human safety, infrastructure, and ecological systems. Their frequent occurrence in topographically complex regions demands urgent development in real-time monitoring solutions. Current monitoring methodologies, however, are constrained by prohibitive costs, limited temporal resolution, and high-power consumption. These factors create substantial implementation barriers to implementing landslide monitoring systems. To address these limitations, this study proposes an economical real-time monitoring method leveraging ultra-wideband (UWB) technology for landslide detection. The implementation of a dual-Microcontroller Unit (MCU) distributed hardware architecture enables high-accuracy ranging capabilities and high real-time performance. To enhance the spatial resolution of UWB systems in landslide monitoring, we propose an optimized sensor deployment structure and a novel deep learning architecture called Ultra-wideband Transformer (UWBformer). This network utilizes differential UWB-ranging data to predict spatial displacement at monitoring locations, specifically the displacement distance, horizontal angle, and pitch angle. UWBformer incorporates a spatial multi-head attention mechanism and a dual-channel architecture processing both time-domain and frequency-domain features. It is specifically designed to mitigate ranging error propagation and enhance prediction stability by focusing on relative distance changes rather than absolute ranging accuracy. Empirical results demonstrate UWBformer's superior performance in predicting displacement distance, horizontal angle, and pitch angle, outperforming the conventional Caffery-Taylor (C-T) localization approach and established deep learning benchmarks. Field tests incorporated 3 σ criterion and Kalman filtering alongside to pre-process raw measurements, thereby enhancing data stability. Comprehensive validation across field tests demonstrates UWBformer's capability to maintain accurate spatial displacement estimation under harsh environments.}, archive = {J_EAAI}, author = {Yu Si and Zhaofeng He and Fan Zhang and Xiaoyun Sun and Yong Chen and Haiqing Zheng}, doi = {10.1016/j.engappai.2025.111851}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111851}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cost-effective and real-time landslide monitoring method based on ultra-wideband using ultra-wideband transformer neural network}, volume = {160}, year = {2025}, }
• (2025). Fast and accurate prediction of h2 combustion via data-augmented physics-informed operator learning. EAAI, 160, 111850. (www)

  @article{ , abstract = {This study models hydrogen combustion within a zero-dimensional constant volume reactor through a physics-informed operator approach. 11 thermochemical variables are considered, including 9 chemical species along with two thermodynamic state variables. Due to the short lifetime of the intermediate species, these variables differ up to 11 orders of magnitudes, resulting in stiff differential equations. Therefore, a scalable architecture designed to predict numerous thermochemical variables varying in order of magnitudes and nonlinearities is proposed. Moreover, small amounts of pre-labeled training data are incorporated to increase training efficiency. Additionally, a dedicated set of training strategies is introduced to improve both training efficiency and convergence. As a result, the completely trained model accurately predicts time-evolving profiles of thermochemical variables in combustion phenomena at given initial conditions. Furthermore, a quantitative analysis of the correlation between the amount of pre-labeled training data and the accuracy of the model is investigated. Results show that the data-augmented physics-informed model can achieve accuracy comparable to that of a data-driven model trained with up to a fivefold increase in training datasets. Consequently, the physics-informed model showed approximately 1% relative error for temperature and pressure, while showing errors lower than 1 0 − 3 for the chemical species concentrations when employing only 25 training datasets.}, archive = {J_EAAI}, author = {Junhyoung Jo and Kyungrae Kang and Paxton Wiersema and Jihyung Yoo and Eric Mayhew and Tonghun Lee}, doi = {10.1016/j.engappai.2025.111850}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111850}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast and accurate prediction of h2 combustion via data-augmented physics-informed operator learning}, volume = {160}, year = {2025}, }
• (2025). Transformer-based deep neural networks for short-term solar power prediction in the middle east and north africa regions. EAAI, 160, 111848. (www)

  @article{ , abstract = {This paper examines the integration of renewable energy (RE) sources into national power grids, with a focus on solar energy in the Middle East and North Africa (MENA) region. The study highlights the potential and variability of solar energy in Algeria, Tunisia, and Saudi Arabia, three countries with significant solar exposure. To improve the accuracy of solar power forecasting, we propose a novel hybrid model called the Long Short-Term Memory (LSTM)-Transformer, which combines LSTM networks with the Encoder Transformer architecture. Extensive testing under two distinct scenarios across diverse regional settings demonstrates the model’s robustness and scalability. Remarkable performance was achieved in Mostaganem (Algeria), Sfax (Tunisia), and Riyadh (Saudi Arabia), with Mean Squared Errors (MSE) of 0.0047 , 0.0045, and 0.0023 , Root Mean Squared Errors (RMSE) of 0.0680, 0.0670, and 0.0512 , and coefficients of determination (R 2 ) of 93%, 94%, and 97%, respectively. Comparative analysis with existing literature confirms that our proposed model outperforms previously reported approaches. These findings underscore the effectiveness of integrating LSTM and Transformer technologies for solar forecasting while demonstrating the model’s adaptability to different geographic locations with varying solar intensities. This research represents a significant step toward advancing renewable energy adoption by providing a reliable predictive tool to enhance the stability and efficiency of photovoltaic (PV) solar power integration into energy systems.}, archive = {J_EAAI}, author = {Ghizlene Cheikh and Boudour Ammar and Naourez Benhadj and Kamil Amine Bentounes and Hadj Adda Bentounes and Amel Ksibi and Ala Saleh Alluhaidan}, doi = {10.1016/j.engappai.2025.111848}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111848}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer-based deep neural networks for short-term solar power prediction in the middle east and north africa regions}, volume = {160}, year = {2025}, }
• (2025). A physics-informed machine learning approach to piezoelectric plate modelling. EAAI, 160, 111847. (www)

  @article{ , abstract = {This paper develops the physics-informed neural network (PINN) framework for solving steady-state piezoelectric equations in plates of various geometries without requiring labelled data. The neural network loss function is designed to capture five dependent variables, including deformation, electric potential, and carrier concentration change, while ensuring the satisfaction of a coupled system of partial differential equations (PDEs). This system comprises five conservation equations, nine constitutive equations, and five relevant boundary conditions. To address the significant disparities in magnitudes among the dependent variables, the governing PDEs are reformulated using dimensionless variables. The performance of the proposed PINN method is benchmarked against traditional numerical approaches implemented in COMSOL Multiphysics, demonstrating comparable accuracy. The spatial average relative errors, with respect to the COMSOL results, for the predicted quantities across the three cases range from 0.36% to 1.23%, with larger errors observed in the electrical quantities. Furthermore, this study investigates the impact of the complexity of neural networks, the number of random training points, and the weight of the boundary loss on the accuracy of the predictions to provide insights into the optimisation of the PINN loss function for piezoelectric problems.}, archive = {J_EAAI}, author = {Narges Tandis and Emad Tandis}, doi = {10.1016/j.engappai.2025.111847}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111847}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A physics-informed machine learning approach to piezoelectric plate modelling}, volume = {160}, year = {2025}, }
• (2025). Path optimization of a flexible robot with a spatial compressed and a direction-guided exploring method. EAAI, 160, 111846. (www)

  @article{ , abstract = {Optimal path planning of a measuring robot is a crucial component in automatic measurement. However, it is hard to obtain a stable optimal path solution when dealing with multiple environments. To overcome this problem, a spatial compressed and direction-guided exploring method is proposed based on Rapidly-exploring Random Tree (RRT). First, to enhance the consistency of the search path, a spatial compaction strategy is incorporated into the direction guidance Rapidly-exploring Random Tree (DG_RRT) method, which optimizes the exploration space for initial path generation. Second, invalid spaces are simplified by model constraints. Third, to obtain more suitable spatial compression models for different environmental features, the generated compressed space is further optimized by adjusting the functional coefficient (FC) and iterations. Path optimization consists of two key steps: linearization for redundant node elimination and curve-fitting for path smoothing. Subsequent discretization is then applied to ensure compatibility with robotic execution constraints. To verify the suggested spatial compressed and direction-guided RRT (SC_DG RRT) method, both numerical simulation and experimental analysis are carried out. The experimental results reveal that the average path length (APL) of SC_DG RRT is 11.4 % lower than that of the DG_RRT and 42.3 % lower than that of Q-learning (QL). The standard deviation (SD) of the path length of SC_DG RRT is 74.3 % lower than that of DG_RRT. It demonstrates that the SC_DG RRT is superior to other methods.}, archive = {J_EAAI}, author = {Yan Wang and Wensong Jiang and Zai Luo and Li Yang and Jiafu Li and Hongzhe Lu}, doi = {10.1016/j.engappai.2025.111846}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111846}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Path optimization of a flexible robot with a spatial compressed and a direction-guided exploring method}, volume = {160}, year = {2025}, }
• (2025). A non-negative garrote shrinkage network with adaptive swish for rotating machinery fault diagnosis under noisy environment. EAAI, 160, 111845. (www)

  @article{ , abstract = {The strong noise existing in the vibration signals has a negative impact on rotating machinery fault diagnosis. To solve the noise problem in the engineering applications of fault diagnosis, a deep residual network, named non-negative garrote shrinkage network with adaptive Swish (NNGSN-AS), is proposed for rotating machinery fault diagnosis under noisy environment. In the NNGSN-AS, the non-negative garrote shrinkage function (NNGSF) is integrated into residual building blocks as nonlinear transformation layers, and the residual building block is named the non-negative garrote shrinkage building unit (NNGSBU). In the NNGSBU, the threshold of the NNGSF is adaptively learned by the thresholding module, so that different thresholds can be assigned to different data samples. The thresholding module is close to the NNGSBU input, enabling early noise handling. The depthwise convolutions with wide kernels in the thresholding module increase the receptive field and lead to a one-to-one correspondence between the learnable threshold and the input feature elements of the NNGSBU, reducing the negative influence of the noise. Additionally, an adaptive Swish (ASwish) activation function module is developed, enabling adaptive nonlinear transformation of each feature channel. The experimental results on a public dataset and our laboratory dataset indicate that the NNGSN-AS is superior to the existing methods for rotating machinery fault diagnosis under noisy environment. Given that the performance gain of the NNGSN-AS stems from multiple components for deep feature extraction, the ablation experiments are conducted to demonstrate the improvement effect of each component.}, archive = {J_EAAI}, author = {Pengcheng Zhong and Zhenyu Liu and Rui Li and Hui Liu and Xiaoqi Yang and Zihan Dong and Jianrong Tan}, doi = {10.1016/j.engappai.2025.111845}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111845}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A non-negative garrote shrinkage network with adaptive swish for rotating machinery fault diagnosis under noisy environment}, volume = {160}, year = {2025}, }
• (2025). Multimodal feature fusion for human activity recognition using human centric temporal transformer. EAAI, 160, 111844. (www)

  @article{ , abstract = {In recent years, human activity recognition (HAR) has focused considerable interest due to its manifold monitoring applications. Mainstream HAR approaches often face challenges with the reliability of results when relying on a single data modality, especially when integrating heterogeneous data sources. A notable limitation of the implemented artificial intelligence (AI) models is their limited capability to handle dynamic scenarios, as they lack the necessary contextual information from multiple sources, which impedes the models’ adaptability and accuracy. This paper proposes a multi-modality framework for HAR that fuses human concern patterns using various spatiotemporal model flavors. In addition, to get the spatial features, a swin transformer with a dual attention concept is applied to process visual sensor data, while one-dimensional convolutional neural network leverages human skeleton information obtained from the detection model with numerous key points. Later, these multi-modality features are fused to improve the robust analysis and comprehension of activities. Next, these resulting features are passed to the human centric temporal transformer (HCTT), that has the capabilities to process multimodal sequence data for temporal learning. Moreover, the attention block of HCTT enables human-related attentive patterns followed by a dual fusion mechanism. The proposed model was evaluated on four open-access large-scale HAR datasets, where comprehensive ablation studies and comparative analyses demonstrated that our developed multimodal approach outperforms recent baseline HAR models. This underscores its potential for advancing AI applications and human activity analysis.}, archive = {J_EAAI}, author = {Samee Ullah Khan and Maryam Sultana and Sufyan Danish and Deepak Gupta and Norah Saleh Alghamdi and Suchang Woo and Dong-Gyu Lee and Sangtae Ahn}, doi = {10.1016/j.engappai.2025.111844}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111844}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodal feature fusion for human activity recognition using human centric temporal transformer}, volume = {160}, year = {2025}, }
• (2025). Forecasting biogas potential in türkiye’s central anatolia region with artificial neural networks and geographical information system-based analysis. EAAI, 160, 111843. (www)

  @article{ , abstract = {The global utilization of biogas energy generated via anaerobic digestion has been steadily increasing, emphasizing the importance of assessing biomass resources and estimating biogas generation potential before implementing large-scale production. This study focuses on Türkiye’s Central Anatolia Region (CAR), which had a bovine population of 3.95 million in 2021, accounting for over 5 % of Europe’s bovine population. A key novelty of this research lies in the estimation of the region’s bovine manure-based biogas potential (BMBP) using 2021 data, which revealed a remarkable increase to 3162 GW-hour (GWh) compared to 1370 GWh in 2004. This significant growth not only highlights the region’s untapped renewable energy potential but also underscores the critical role of advanced methodologies in accurately assessing and forecasting energy resources over time. To forecast future potential, artificial neural networks (ANNs) were employed to estimate the BMBP of all provinces in the CAR up to 2035. Among these, Konya is projected to have the highest BMBP in 2035, with 919 GWh, contributing approximately 50 % of the electricity consumed by its habitations and 4.5 times the electricity expended for lighting. Additionally, Arc Geographical Information System (ArcGIS) was utilized to perform geographical and temporal analyses of the region, providing a comprehensive spatial perspective. Both the ArcGIS findings and the BMBP findings highlight the significant contributions of BMBP to renewable energy resources in the CAR, offering critical insights for shaping regional energy policies. Future research should expand to include biogas potential from other agricultural wastes in the region.}, archive = {J_EAAI}, author = {Halil Şenol and Emre Çolak and Volkan Başer}, doi = {10.1016/j.engappai.2025.111843}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111843}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forecasting biogas potential in türkiye’s central anatolia region with artificial neural networks and geographical information system-based analysis}, volume = {160}, year = {2025}, }
• (2025). A machine learning-based inverse design method for frequency-selective surface microwave absorbers. EAAI, 160, 111842. (www)

  @article{ , abstract = {Electromagnetic compatibility (EMC) and electromagnetic interference (EMI) shielding in the electric vehicle have attracted more and more attention. As an effective solution for EMC and EMI in the electric vehicle, microwave absorbers have attracted more and more attention. In the design of microwave absorbers, more degrees of freedom (DoFs) can enhance performance, but the associated design cost increases. To address this issue, this work proposes a machine learning-based inverse topological design method. Different from the traditional methods, i.e. equivalent circuit methods, the proposed method designs the absorber using topological and decimal variables, offering more DoFs and higher efficiency. Furthermore, mixing on multi-layer perceptrons (MLP-Mixer), is employed to map the desired electromagnetic response to the corresponding absorber structure. To validate the proposed method’s design capability, a predicted design sample is fabricated and measured. Measurements indicate that under normal incidence, the absorber operates within a frequency range of 1.33–7.31 GHz, demonstrating a significant improvement in absolute bandwidth. The relative bandwidth is 138.2%, with a thickness of 0.114 λ L , where λ L is the wavelength at the lowest operating frequency. At the incidence angle of 45°, both transverse electric (TE) and transverse magnetic (TM) polarizations maintain a broad relative bandwidth.}, archive = {J_EAAI}, author = {Xin-Yue Qi and Li-Ye Xiao and Hao Lv and Yan-Fang Liu and Wei Shao}, doi = {10.1016/j.engappai.2025.111842}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111842}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A machine learning-based inverse design method for frequency-selective surface microwave absorbers}, volume = {160}, year = {2025}, }
• (2025). Application of artificial neural networks to model macroscopic spray characteristics of alcohol fuels. EAAI, 160, 111841. (www)

  @article{ , abstract = {In pursuit of global efforts to mitigate climate change and achieve net-zero carbon emissions by 2050, alcohol-based fuels are gaining attention as low-carbon alternatives for combustion engines. This study presents a novel application of Artificial Intelligence (AI) to predict the spray characteristics, specifically spray cone angle and penetration length, for a wide range of alcohol fuels and operating conditions. A robust Artificial Neural Network (ANN) model was developed using the Keras Application Programming Interface (API) on the TensorFlow platform, trained on a comprehensive dataset combining in-house experimental data for ethanol, octanol, and published data for methanol and butanol fuels. Spray behaviour was captured using shadowgraph technique under varied injection pressures (20–239 bar) and chamber pressures (9–21 bar). The ANN model demonstrated high predictive accuracy, with mean square error and coefficient of determination (R 2 ) of 1.7189 and 0.9841 for spray penetration, and 6.3734 and 0.9163 for spray angles, respectively. The unified modeling framework effectively captures the complex interactions of different alcohol fuels and operating conditions, enabling advanced, accurate simulations of spray behaviour. This AI-driven approach could be a valuable tool for predicting the spray behavior of alcohol fuels, facilitating advanced modeling of low-carbon fuel combustion processes.}, archive = {J_EAAI}, author = {Thangaraja Jeyaseelan and Abhijeet Gopalakrishnan and Sundararajan Rajkumar and Santosh Narayan V and Min Son and Lars Zigan}, doi = {10.1016/j.engappai.2025.111841}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111841}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Application of artificial neural networks to model macroscopic spray characteristics of alcohol fuels}, volume = {160}, year = {2025}, }
• (2025). Video saliency prediction via single feature enhancement and temporal recurrence. EAAI, 160, 111840. (www)

  @article{ , abstract = {Deep learning techniques have made significant advancements in Video Saliency Prediction (VSP), with most state-of-the-art models relying on complex multi-scale feature fusion. However, these models often encounter issues such as structural redundancy and the lack of effective temporal transmission mechanisms, which seriously affect prediction accuracy. Thus, we propose a Single Feature Enhancement and Temporal Recurrence Network (SETR-Net) to address these issues. First, we confirmed a single feature baseline with competitive accuracy and designed a subtle Saliency Context Enhancement (SCE) module to enhance the receptive field and expressive power of individual feature, making its performance superior to existing multi-scale models. Then, the Saliency Attention Transmission (SAT) mechanism is designed to introduce temporal recurrence into VSP transformer-based methods, thereby improving the capture capability of long-distance features. Experimental results on multiple challenging benchmark datasets demonstrate that SETR-Net achieved a gain of up to 3.7% in Normalized Scanpath Saliency (NSS) over state-of-the-art methods.}, archive = {J_EAAI}, author = {Yunzuo Zhang and Yaoge Xiao and Yuekui Zhang and Tian Zhang}, doi = {10.1016/j.engappai.2025.111840}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111840}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Video saliency prediction via single feature enhancement and temporal recurrence}, volume = {160}, year = {2025}, }
• (2025). Mixed-type micro-defect detection in semiconductor wafers: A dual-modal feature real-time detection approach via optical topography and lightweight classification network. EAAI, 160, 111838. (www)

  @article{ , abstract = {In the semiconductor silicon wafer manufacturing process, surface defects, particularly micro-scale imperfections, may arise during cutting, grinding, and polishing operations. To address the core challenges of difficult feature recognition and stringent real-time industrial detection requirements in micro-defect inspection of 12-inch semiconductor silicon wafers, this paper proposes an innovative tripartite solution integrating “dual-modal imaging, feature co-optimization, and edge computing adaptation.” First, a dual-modal imaging system combining annular optical arrays with line-scan cameras is developed to capture front/back surface and edge defect images with 0.1 μm (μm) pixel resolution, enabling micrometer-scale defects identification and feature extraction through edge detection and threshold segmentation. Second, a dataset containing six types of defects was produced, named the Wafer Defect Type Dataset (WDTDS-6). To overcome the limitations of conventional deep learning models in industrial scenarios, a lightweight residual network incorporating Efficient Multi-scale Attention for silicon wafer classification (SWC-ResEMA-Net) is proposed. The network integrates a heterogeneous collaborative design of depth-wise convolution, Grouped Shuffle Convolution, and Efficient Multi-Scale Attention (EMA) mechanisms, improving recall rates for critical defect categories, and reducing single-sample inference time through lightweight structural optimization. Experimental results demonstrate that the optimized model achieves 98.5 % classification accuracy on the test set, outperforming the baseline by 2.6 percentage points, with a comprehensive missed detection rate below 0.01 %. The deployed model occupies only 23.7 Megabyte (MB) of memory, supporting parallel processing on multi-station edge computing platforms. This work provides an artificial intelligence (AI) powered solution for high-precision and high-efficiency online quality monitoring in semiconductor wafer manufacturing processes.}, archive = {J_EAAI}, author = {Shangbin Jiao and Wei Yang and Chenyan Wu and Yujun Li and Bing Xue}, doi = {10.1016/j.engappai.2025.111838}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111838}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mixed-type micro-defect detection in semiconductor wafers: A dual-modal feature real-time detection approach via optical topography and lightweight classification network}, volume = {160}, year = {2025}, }
• (2025). Semantic awareness aggregation for salient object detection in remote sensing images. EAAI, 160, 111837. (www)

  @article{ , abstract = {In recent years, remote sensing image salient object detection (RSI-SOD) has garnered increasing attention from researchers. Due to the inherent complexity of remote sensing images (RSI), including multiple objects within images, significant variations in object sizes, and shadow interference, this research faces numerous challenges. To cope with these challenges, this paper proposes a semantic awareness aggregation network (SAANet), designed to address the limitations of existing methods in understanding global information and exploring feature relevance. SAANet consists of two main components: the multi-scale object extraction Mamba (MOEM) module and the adaptive feature fusion (AFF) module. Specifically, MOEM improves sensitivity to salient object locations through global information understanding. Meanwhile, AFF is utilized to progressively guide deep features to shallower layers and explore correlations between features at different levels, thus achieving accurate detection of salient regions and suppression of redundant features. Extensive experiments on three publicly available RSI-SOD datasets show that the proposed method outperforms 17 current state-of-the-art methods. Our code and prediction results are available on https://github.com/nzmtnh/SAANet .}, archive = {J_EAAI}, author = {Yanliang Ge and Taichuan Liang and Junchao Ren and Min He and Hongbo Bi and Qiao Zhang}, doi = {10.1016/j.engappai.2025.111837}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111837}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semantic awareness aggregation for salient object detection in remote sensing images}, volume = {160}, year = {2025}, }
• (2025). Advancements in collision avoidance techniques for internet-connected vehicles: A comprehensive review of methods and challenges. EAAI, 160, 111836. (www)

  @article{ , abstract = {Collision avoidance (CA) in internet-connected vehicles (ICVs) is critical for ensuring safety and efficiency in smart transportation systems. The ICV control system achieves CA through integrated features, including sensor-based perception, communication technologies, and data-driven artificial intelligence, enabling real-time optimization for smooth cruising. By using these adaptive architectures on one platform, ICVs enhances both individual-level vehicle performance and network-wide traffic efficiency. This review examines a wide range of CA methods and strategies, offering comprehensive insights into how ICV systems detect and avoid obstacles in dynamic environments. We critically assess existing research, evaluating the effectiveness, challenges, and future directions of CA techniques, with particular attention to static and dynamic obstacle handling and interactions with other road users. Furthermore, we systematically explore ICV control systems, emphasizing how integrated technologies improve safe mobility and accident prevention. Our research synthesizes findings from peer-reviewed journals and conference proceedings (primarily from the past decade) to support the development of robust CAsystems. These insights aim to advance reliable CA frameworks for ICVs, fostering safer and more efficient transportation networks.}, archive = {J_EAAI}, author = {Khurrum Jalil and Yuanqing Xia and Jing Zhao}, doi = {10.1016/j.engappai.2025.111836}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111836}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advancements in collision avoidance techniques for internet-connected vehicles: A comprehensive review of methods and challenges}, volume = {160}, year = {2025}, }
• (2025). Data expansion for industrial defect detection with diffusion prior. EAAI, 160, 111835. (www)

  @article{ , abstract = {Defect detection plays a vital role in industrial production. However, achieving high-precision detection requires substantial training data, which is inherently scarce due to the rare occurrence of defects in industrial settings. Although generative models present a potential solution, the limited defect data also hinders to train models from scratch. This particularly affects the ability to generate high-quality defect images that precisely follow specified layout conditions. As a solution, we propose a novel layout-to-defect generation method, which leverages the generative prior of established layout-conditioned diffusion models to guarantee the synthesis performance with limited defect data. The proposed method guides the diffusion model to synthesis with the specified defect category and region, also known as the layout. These generated images, paired with layout information as ground truth, can be directly used to train defect detection models. We conduct experiments on metal and cloth production datasets, and deploy it in a real-world round billet surface defect detection system. Experiments demonstrate that our method ensures high-quality defect samples that accurately match category and region annotations, boosting defect detection model performance.}, archive = {J_EAAI}, author = {Yichun Tai and Zhijiang Zhang}, doi = {10.1016/j.engappai.2025.111835}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111835}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data expansion for industrial defect detection with diffusion prior}, volume = {160}, year = {2025}, }
• (2025). A multimodal gait and ocular geometric representation to generate a parkinson progression report. EAAI, 160, 111834. (www)

  @article{ , abstract = {Parkinson’s disease (PD) is a progressive neurological condition, primarily associated with a deficiency in dopamine neurotransmitters, generating premotor, motor control, emotional, and executive dysfunctions. The characterization and PD diagnosis are principally based on the analysis of observed motion alterations, such as slowed movements (bradykinesia), wrong posture, and freezing of gait. Computational methods to support some of these observations remain limited to distinguishing between PD patients and control patients on the basis of protocols at advanced stages, according to current clinical PD guidelines. This work introduces a multi-item PD progression support statistically consistent with both the modified Hoehn and Yahr (H&Y) scale and the Movement Disorder Society-Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) part III, which is based on a multimodal geometric representation that combines gait and oculomotor markerless video sequences. The proposed representation evaluates gait autonomy, posture impairment, freezing, bradykinesia, bilateral gait issues, and ocular bradykinesia. To do so, a three dimensional (3D) convolutional neural network (CNN) first captures spatiotemporal PD patterns. Then, a Riemannian network learns second order relationships among observed patterns, which are further fused at early or intermediate stages to output multi-item PD predictions. In a retrospective study with 13 control subjects and 19 diagnosed Parkinson’s patients, the proposed approach (early fusion) achieved F1-scores of 95% for bilateral impairment, 94% for gait autonomy, 81% for freezing of gait, 92% for wrong posture, 91% for gait bradykinesia, and 94% for ocular bradykinesia in Parkinson’s patients. The proposed approach is a promising tool to support routine and standard clinical PD analysis.}, archive = {J_EAAI}, author = {John Archila and Ivan Peña and Luis Celis and Juan Olmos and Antoine Manzanera and Fabio Martínez}, doi = {10.1016/j.engappai.2025.111834}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111834}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multimodal gait and ocular geometric representation to generate a parkinson progression report}, volume = {160}, year = {2025}, }
• (2025). Deep image stitching for panoramic stereoscopic live broadcast system. EAAI, 160, 111832. (www)

  @article{ , abstract = {Image stitching is a crucial task in computer vision, enabling the creation of panoramic images for augmented reality/virtual reality (AR/VR) experiences. Image registration plays an important role in achieving spatial continuity and consistency in image stitching. Therefore, achieving accurate image registration is critical to high-quality panoramic image generation. Although recent learning-based methods have improved the optimization of stitched images, they still rely on simple associations of images or feature maps to search for feature correspondences, neglecting the capture of these correspondences during feature extraction. Additionally, directly warping the original image with the homography matrix extracted from the feature space ignores the network’s influence on image transformation relationships. To address these limitations and enhance the quality of panoramic images, we propose an artificial intelligence (AI) image registration network based on a cross-attention mechanism. Our approach incorporates a Local Transformer to help the network perceive image correspondences. Furthermore, we impose constraints on network training through warp-equivariance to mitigate the impact on image transformation relationships. These strategies increase the accuracy and generalization of the AI-based registration method and form a better image stitching algorithm. Finally, we apply the AI image stitching algorithm to construct a panoramic stereoscopic live broadcast system. Experimental results show that our method achieves competitive results and satisfies the requirements of panoramic systems.}, archive = {J_EAAI}, author = {Yun Tie and Zihao Zhao and Dalong Zhang and Yuning Gao}, doi = {10.1016/j.engappai.2025.111832}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111832}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep image stitching for panoramic stereoscopic live broadcast system}, volume = {160}, year = {2025}, }
• (2025). A cases and clusters framework for recording, retrieving, and reusing response plans in structured cybersecurity incident management. EAAI, 160, 111831. (www)

  @article{ , abstract = {The dynamic and increasing sophistication of cyberattacks and vulnerability exploitation creates a need for Explainable Artificial Intelligence (XAI) approaches that help maintain cyber resilience in organizations. In structured cybersecurity incident management, effective incident response demands explainable outputs from AI-based decision-support systems. To approach this problem, this work presents a framework for reusing concrete experiences of cybersecurity incident response, capturing problem-solving data and knowledge as cases for integrated Case-Based Reasoning (CBR) and Clustering. The contribution includes cluster-based query answer analysis, where cybersecurity analysts reuse clusters of retrieved incident response cases to build answers to new problems. Clustering helps analysts identify relevant groups from ranked lists of retrieved cases, making the reuse process more structured and understandable, especially when dealing with retrieval results for broad and ambiguous queries. Different clustering methods are applied to organize retrieved incident response cases from a case base, supporting the grouping of similar cases for a more straightforward interpretation. Multiple experiments, including cross-validation and real-world incident response testing, are conducted to demonstrate the effectiveness of the proposed framework in improving the decision-support system’s precision. The results indicate that exploring cases and clusters can enhance the selection of incident response procedures for reuse, mainly when analysts identify the most relevant clusters of retrieved cases for the given problem situations. The proposed framework contributes to the organization and understanding of responses to cybersecurity incidents, besides supporting more informed decision-making, ultimately improving cybersecurity incident management.}, archive = {J_EAAI}, author = {Patrick Andrei Caron Guerra and Raul Ceretta Nunes and Luis Alvaro de Lima Silva}, doi = {10.1016/j.engappai.2025.111831}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111831}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cases and clusters framework for recording, retrieving, and reusing response plans in structured cybersecurity incident management}, volume = {160}, year = {2025}, }
• (2025). Real-time individual subway destination prediction: An AdaBoost graph neural network. EAAI, 160, 111830. (www)

  @article{ , abstract = {Subways have long been an essential travel mode of urban public transportation. Real-time individual destination prediction for passengers is significant in realizing the personalized trip planning, location-based service (LBS) recommendations, and seamless public transportation transfer. Most existing studies for predicting the real-time individual destination rely on statistical models or deep learning models, but they ignore the influence of the bus network on the subway network and require millions of trip records for training. To fill this research gap, we propose an AdaBoost Graph Neural Network (AGNN) to predict the real-time subway destination. AGNN extracts four kinds of features: contextual features, historical trip features, instant trip features, and subway-bus integration features. Moreover, we employ a contextual module to process the semantic information. The graph convolution and attention-based module (GCAM) is utilized to capture spatiotemporal mobility features, including passenger movement patterns in hourly interval and origin–destination station pairs in space. Numerical experiments using the real-world subway dataset in Beijing are conducted to validate the effectiveness of our model. Our proposed model outperforms existing studies on the task of the real-time individual destination prediction, achieving the prediction accuracy of 98.48%.}, archive = {J_EAAI}, author = {Zhenhao Meng and Zhengli Wang and Xiang Liu and Shan Liu and Ya Zhang}, doi = {10.1016/j.engappai.2025.111830}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111830}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time individual subway destination prediction: An AdaBoost graph neural network}, volume = {160}, year = {2025}, }
• (2025). A cross-dimensional synergistic network for brain tumor segmentation. EAAI, 160, 111829. (www)

  @article{ , abstract = {Brain tumor segmentation is crucial in medical image processing but struggles to access and integrate information across dimensions, such as global–local, multi-scale, long-range dependencies information, and the receptive field, leading to suboptimal accuracy and robustness. This study proposes a novel attention network, the Cross-Dimensional Synergistic Network (CS-Net), for brain tumor segmentation, which incorporates three kinds of symmetrical relationships — also called cross-dimensional information, including global–local, vertical-horizontal axis, and large–small (multi-scale) relationships. The symmetrical relationships are translated into three attention mechanisms: Global–local Region Attention (GRA), Axis-aligned Translation Attention (ATA), and Multi-scale Fusion Attention (MFA). The GRA module enhances spatial perception by dividing the input feature map into non-overlapping sub-regions and computing attention weights between each sub-region (Local) and global features to capture finer spatial dependencies. The ATA module calculates the cross-directional attention between the entire original and shifted features, where the shifted features include horizontal and vertical shifts to learn the long-range dependencies and extend the receptive field in different directions. The MFA module executes self-attention interactions between feature maps at different scales to effectively integrate adjacent scale information. Experiments on the Brain Tumor Segmentation Challenge (BraTS) from the Medical Image Computing and Computer-Assisted Intervention (MICCAI) and the Medical Segmentation Decathlon (MSD) demonstrate significant performance improvements. Our model achieves 86.23%, 89.99%, and 1.2485 in the Dice coefficient, precision, and 95%Hausdorff Distance on the MICCAI BraTS dataset and achieves 81.57%, 86.70%, and 1.5547 in the Dice coefficient, precision, and 95%Hausdorff Distance on the MSD BraTS dataset, suppressing the state-of-the-art approaches.}, archive = {J_EAAI}, author = {Chih-Wei Lin and Ye Lin}, doi = {10.1016/j.engappai.2025.111829}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111829}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cross-dimensional synergistic network for brain tumor segmentation}, volume = {160}, year = {2025}, }
• (2025). State slow feature softmax Q-value regularization for offline reinforcement learning. EAAI, 160, 111828. (www)

  @article{ , abstract = {Offline reinforcement learning is constrained by its reliance on pre-collected datasets, without the opportunity for further interaction with the environment. This restriction often results in distribution shifts, which can exacerbate Q-value overestimation and degrade policy performance. To address these issues, we propose a method called state slow feature softmax Q-value regularization (SQR), which enhances the stability and accuracy of Q-value estimation in offline settings. SQR employs slow feature representation learning to extract dynamic information from state trajectories, promoting the stability and robustness of the state representations. Additionally, a softmax operator is incorporated into the Q-value update process to smooth Q-value estimation, reducing overestimation and improving policy optimization. Finally, we apply our approach to locomotion and navigation tasks and establish a comprehensive experimental analysis framework. Empirical results demonstrate that SQR outperforms state-of-the-art offline RL baselines, achieving performance improvements ranging from 2.5% to 44.6% on locomotion tasks and 2.0% to 71.1% on navigation tasks. Moreover, it achieves the highest score on 7 out of 15 locomotion datasets and 4 out of 6 navigation datasets. Detailed experimental results confirm the stabilizing effect of slow feature learning and the effectiveness of the softmax regularization in mitigating Q-value overestimation, demonstrating the superiority of SQR in addressing key challenges in offline reinforcement learning.}, archive = {J_EAAI}, author = {Lan Wu and Quan Liu and Renyang You}, doi = {10.1016/j.engappai.2025.111828}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111828}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {State slow feature softmax Q-value regularization for offline reinforcement learning}, volume = {160}, year = {2025}, }
• (2025). SFANet: Semantic feature aware for domain generalized semantic segmentation. EAAI, 160, 111827. (www)

  @article{ , abstract = {Domain generalized semantic segmentation requires that the model trained on the source data has strong generalization in the unseen target domain. Feature enhancement is an effective method to improve the generalization ability of the model. However, the role of Vision Foundation Models (VFMs) is often overlooked. In this paper, we propose a method that combines feature enhancement and model fine-tuning, which improves the generalization ability of the model by optimizing the fine-tuned feature maps through Semantic Feature Aware Normalization (SFAN) and Semantic Feature Aware Whitening (SFAW). First, Instance Compensation performs task compensation on the VFMs as the backbone through trainable low-rank tokens, narrowing the task gap between semantic segmentation and pre-training. Secondly, the pseudo segmentation head is used to segment the feature map of the backbone. SFAN performs instance-level normalization on the feature map based on the segmentation result to eliminate domain features. Finally, SFAW uses the weights of the pseudo segmentation head to calculate the influence of each channel on each instance. It decorrelates the channels with high influence to solve the problem of joint distribution mismatch. The datasets used are Grand Theft Auto V (GTAV), SYNTHetic Collection of Imagery and Annotations of Urban Scenarios (Synthia), Cityscapes, Berkeley DeepDrive 100K (BDD100K), and Mapillary Vistas Dataset (Mapillary). In the GTAV＋ Synthia → Cityscapes＋ BDD100K＋ Mapillary setting, the average mIoU (Mean Intersection over Union) reaches 67.7%. In the GTAV → Cityscapes＋ BDD100K＋ Mapillary setting, the average mIoU reaches 64.6%. In the Synthia → Cityscapes＋ BDD100K＋ Mapillary setting, the average mIoU reaches 64.2%. Code is available at https://github.com/xiaoxia0722/SFANet .}, archive = {J_EAAI}, author = {Yongchao Qiao and Ya’nan Guan and Wenyuan Yang and Jingmin Yang}, doi = {10.1016/j.engappai.2025.111827}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111827}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SFANet: Semantic feature aware for domain generalized semantic segmentation}, volume = {160}, year = {2025}, }
• (2025). A noise-robust framework for multi-step wind power forecasting via self-supervised joint optimization. EAAI, 160, 111826. (www)

  @article{ , abstract = {Short-term wind power forecasting remains a critical yet challenging task due to the inherent volatility and intermittency of wind. While deep learning has shown promise, it is still prone to noise and cumulative errors in multi-step prediction. This paper presents Time Self-Supervised Contrastive Learning (TimeSCL), a novel framework that integrates contrastive learning with self-supervised joint optimization to enhance robustness and accuracy. TimeSCL introduces an adaptive noise injection mechanism to generate clean—noisy sample pairs, enabling the model to learn noise-invariant representations. A dual forward propagation strategy contrasts these representations before and after denoising, improving temporal modeling. To further mitigate overfitting and error accumulation, TimeSCL incorporates dropout and a time-to-frequency domain loss function, jointly capturing temporal and spectral features. Additionally, a progressive contrastive loss weighting (PCLW) strategy is employed during training to dynamically balance the contributions of different loss components, ensuring stable and effective optimization. Extensive experiments conducted on diverse public benchmarks and real-world wind turbine datasets show that TimeSCL consistently outperforms state-of-the-art baselines, achieving up to a 13.6% reduction in mean squared error and a 5.4% reduction in mean absolute error. These results demonstrate the framework’s effectiveness and practicality for robust wind power forecasting in noisy and complex environments.}, archive = {J_EAAI}, author = {Xiaodong Huang and Gangliang Li and Chengfeng Chen and Kai Yang and Shouqiang Liu}, doi = {10.1016/j.engappai.2025.111826}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111826}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A noise-robust framework for multi-step wind power forecasting via self-supervised joint optimization}, volume = {160}, year = {2025}, }
• (2025). A novel lightweight semantic segmentation network for line-structured light measurement under strong reflection noise. EAAI, 160, 111825. (www)

  @article{ , abstract = {Line-structured light measurement is a critical technology for realizing industrial automation. However, in practical applications, the noise from multiple reflections between object surfaces can cause non-ideal distributions of light stripes, which poses challenges for three-dimensional (3-D) measurement accuracy and correctness. To alleviate this problem, a topology-based lightweight segmentation network based on the geometric features is proposed in this paper, and is integrated into the line-structured light 3-D measurement system under the premise of ensuring accuracy and low latency. Experimental results demonstrate that the proposed method can accurately extract laser stripes under strong reflection noise, with a processing speed of about 22.65 frames per second (FPS) on central processing unit (CPU) and 94.70 FPS on graphics processing unit (GPU).}, archive = {J_EAAI}, author = {Xinyu Chen and Chen Fang and Yan Ren and Ailing Hu}, doi = {10.1016/j.engappai.2025.111825}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111825}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel lightweight semantic segmentation network for line-structured light measurement under strong reflection noise}, volume = {160}, year = {2025}, }
• (2025). Reinforcement learning-based optimization for freeze-dried fruit production scheduling under time-of-use electricity tariffs. EAAI, 160, 111824. (www)

  @article{ , abstract = {This study addresses the batch scheduling challenges in fruit freeze-drying processes, characterized by high energy consumption during freezing and prolonged vacuum drying under time-of-use (TOU) electricity tariffs. To maximize profit while adhering to job deadlines and energy cost constraints, a mixed-integer linear programming (MILP) model is developed. For large-scale problems, a deep reinforcement learning-based batch scheduling (DRLB) framework is proposed, leveraging a deep Q-network to dynamically select optimal scheduling rules by integrating environmental states, including TOU tariffs, machine capacity, and job attributes. Experimental validation on small-scale cases confirms the MILP model's correctness. Comparison with benchmark algorithms using large-scale random data shows that DRLB outperforms heuristics in various scenarios. Except for situations where machine capacity is small and traffic intensity is high, DRLB yields better scheduling results than particle swarm optimization and genetic algorithms in all other cases. When the scheduling scale does not exceed 500 jobs, the computation time of DRLB is less than 1 s. The algorithm's adaptability to diverse job distributions and varying TOU price spreads highlights its practical value for energy-efficient production scheduling.}, archive = {J_EAAI}, author = {JinDian Huang and Chengfeng Peng and Jianjun Liu and Gengzhong Zheng}, doi = {10.1016/j.engappai.2025.111824}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111824}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reinforcement learning-based optimization for freeze-dried fruit production scheduling under time-of-use electricity tariffs}, volume = {160}, year = {2025}, }
• (2025). DRL-ED: A deep reinforcement learning with encoder–decoder method for traffic flow prediction. EAAI, 160, 111823. (www)

  @article{ , abstract = {Accurate traffic flow prediction assists transport authorities in addressing transportation problems. However, most forecasting methods encounter challenges in simultaneously modeling dynamic topologies and capturing long-range spatiotemporal dependencies. To this end, this paper proposes a D eep R einforcement L earning with E ncoder- D ecoder (DRL-ED) method for traffic flow prediction. The proposed method incorporates an encoder-decoder architecture, which is equipped with an attention mechanism, a gated recurrent unit (GRU) network, and a correlation graph convolutional network (CorrGCN). Specifically, the attention mechanism facilitates the capture of spatiotemporal relationships by the GRU and CorrGCN networks, improving the accuracy and robustness of traffic flow prediction. Moreover, the GRU network contains short- and long-term memory units, enabling the discovery of temporal correlations across different time scales. This enables the method to consider both short-term fluctuations and long-term trends, allowing for more accurate prediction of traffic flow changes. Finally, the CorrGCN network aggregates node features through multi-layer graph convolution, enabling step-by-step information propagation to extract both local and global spatial features, thereby comprehensively understanding inter-site traffic flow relationships. The deep deterministic policy gradient (DDPG) algorithm is employed, enabling the CorrGCN network to automatically learn and update the dynamic neighbor matrix by the actor-critic framework combined with the soft update and empirical playback mechanism. This enables the network to adaptively adjust the adjacency matrix according to traffic data in different periods or scenarios, capturing spatiotemporal correlations in graph data. Experimental results demonstrate that DRL-ED outperforms 15 baseline methods in prediction accuracy.}, archive = {J_EAAI}, author = {Dawen Xia and Yan Chen and Wenyong Zhang and Xiaoduo Wei and Yang Hu and Yantao Li and Huaqing Li}, doi = {10.1016/j.engappai.2025.111823}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111823}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DRL-ED: A deep reinforcement learning with encoder–decoder method for traffic flow prediction}, volume = {160}, year = {2025}, }
• (2025). Modal mimicking knowledge distillation for monocular three-dimensional object detection. EAAI, 160, 111821. (www)

  @article{ , abstract = {Monocular three-dimensional (3D) object detection has gained attention for its cost-effectiveness in autonomous driving systems. Nevertheless, the extraction of depth information from two-dimensional (2D) images is an ill-posed problem. To address this challenge, cross-modal knowledge distillation techniques is widely adopted. A prevalent approach involves projecting Light Detection and Ranging (LiDAR) data onto the image plane to train teacher networks that share homogeneous architectures with student networks. Nevertheless, the alignment of features between LiDAR-based teacher networks and image-based student networks remains challenging. In order to address the inherent misalignment between modalities, this paper proposes a Modal Mimicking Knowledge Distillation (MMKD) framework using deep convolutional neural networks for autonomous perception tasks. The purpose of the MMKD framework is to explicitly reinforce depth features in the image-based student network, by introducing a depth prediction branch on the foundation of homogeneous teacher and student networks. Specifically, we propose a Road Plane Discretization (RPD) strategy that transforms projected LiDAR information to construct depth supervision signals better suited to the image plane. Concurrently, we propose Dual-Kullback–Leibler divergence distillation (DualKL), which integrates a dynamic Kullback–Leibler divergence balancing mechanism with depth uncertainty weighting, to efficaciously extract and transfer knowledge from the teacher network. The experimental results demonstrate that the proposed method achieves significant performance improvements on Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) benchmarks. Specifically, our approach achieves 4.4% improvement on the easy level and 2.1% improvement on the difficult level compared to the baseline model. Our code will be released at https://github.com/yangmenghao9/MonoMMKD .}, archive = {J_EAAI}, author = {Menghao Yang and Yafei Qi and Bing Xiong and Zhaoning Zhang and Yaping Liu}, doi = {10.1016/j.engappai.2025.111821}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111821}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Modal mimicking knowledge distillation for monocular three-dimensional object detection}, volume = {160}, year = {2025}, }
• (2025). Thermal fall 66: A robust dataset for thermal imaging-based fall detection and eldercare. EAAI, 160, 111819. (www)

  @article{ , abstract = {Fall-related injuries pose a significant health risk, particularly among the elderly population, necessitating advancements in fall detection technologies. While traditional sensor-based methods offer some solutions, they are limited by issues, like user compliance and unreliability in high-stakes situations. Computer vision (CV) and artificial intelligence (AI)-based alternatives, notably thermal imaging, emerge as promising, privacy-preserving tools. However, the lack of standardized, generalizable benchmark datasets hampers the validation of these technologies’ effectiveness. The Thermal Fall 66 (TF-66) dataset introduced in this work addresses this crucial gap by providing an extensive, diverse collection of fall scenarios, recorded in various environments and featuring a wide range of participants. This dataset introduces sample subsets for tailored model training and a comprehensive data generator for customized access, setting a new benchmark in fall detection research. A baseline three-dimensional convolutional neural network (3D CNN) was trained to demonstrate the dataset’s suitability for supervised deep learning. As TF-66 continues to evolve, it aims to include even more diverse environments, such as bathrooms and staircases, further enhancing its applicability and serving as a robust resource for the research community. This work contributes significantly to the field by offering a dataset that not only improves the reliability of fall detection systems (FDS) but also facilitates a standardized approach to evaluating and advancing these technologies.}, archive = {J_EAAI}, author = {Christopher Silver and Thangarajah Akilan}, doi = {10.1016/j.engappai.2025.111819}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111819}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Thermal fall 66: A robust dataset for thermal imaging-based fall detection and eldercare}, volume = {160}, year = {2025}, }
• (2025). Causality thinking for large-scale long-tailed video action recognition. EAAI, 160, 111818. (www)

  @article{ , abstract = {Video action recognition aims to accurately classify actions in video data by leveraging spatial–temporal representations. However, this task faces two major challenges: (1) the intrinsic long-tailed data distribution in real-world scenarios, which skews model learning toward overrepresented classes, and (2) the presence of spurious correlations in the data, which undermines prediction reliability. In this paper, we propose a causal inference framework for long-tailed video action recognition. A novel debiasing approach with Causal Intervention and Counterfactual Reasoning (CICR) is proposed to yield more robust predictions. The Structural Causal Model (SCM) is constructed to identify and analyze the causal relationships among spatial–temporal variables and long-tailed class distributions. Specifically, we first diagnose momentum as the confounder and use the intervention to eliminate the spurious relationship existed among video frame appearance features. The counterfactual reasoning is then utilized during inference to calculate the direct causal effect more accurately. During experiments conducted on two large-scale datasets, the proposed approach improved video action recognition accuracy by 1.9% on EPIC-KITCHENS-100 and 3.65% on Something-Something-V2-LT, thereby showing the effectiveness of the CICR model in mitigating spurious correlation and long-tailed distribution issues. Codes are available at https://github.com/sandyzhang2021/Causality-for-long-tailed-video-action-recognition .}, archive = {J_EAAI}, author = {Zhengjin Zhang and Nannan Li and Wenmin Wang and Huiwen Guo and Wei Jin and Sudan Huang}, doi = {10.1016/j.engappai.2025.111818}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111818}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Causality thinking for large-scale long-tailed video action recognition}, volume = {160}, year = {2025}, }
• (2025). A decision-making model with new Dice–Jaccard similarity measure: Green hydrogen electrolyzer selection. EAAI, 160, 111817. (www)

  @article{ , abstract = {Decision-making is a fundamental and complex cognitive process that influences our daily lives. Moreover, as uncertainty looms over decision-making processes, effective management becomes imperative. Therefore, the objective of this study is to establish a novel hybrid multi-attribute decision-making methodology, with a specific focus on the following objectives: (i) addressing ambiguity and interrelationships inherent in decision-making problems, (ii) integrating uncertain attribute weights rather than relying solely on crisp values, and (iii) presenting ranking outcomes that incorporate conflict analysis among alternatives and accommodate preference relations within multiple conflicting elements. Firstly, q -rung orthopair fuzzy sets is initiated for handling uncertainties in expert opinions during evaluations process. Secondly, leveraging the flexibility and generality of Aczel–Alsina norms, weighted geometric Bonferroni mean is developed within q -rung orthopair fuzzy sets framework. Additionally, an advanced form of Dice and Jaccard similarity measures are devised, using new quadratic form to accurately measure similarity between q -rung orthopair fuzzy sets. Furthermore, a logarithmic methodology of additive weights method is proposed within the q -rung orthopair fuzzy sets environment to derive weights and minimize information loss. Subsequently, organísation, rangement et Synthèse de données relarionnelles (in French) method is extended to rank alternatives, employing a new generalized form of the developed similarity measures. Finally, harnessing these advancements, a hybrid multi-attribute decision-making methodology is developed and applied to determine the most suitable electrolysis for green hydrogen production in the eastern part of India.}, archive = {J_EAAI}, author = {Kaushik Debnath and Erfan Babaee Tirkolaee and Sankar Kumar Roy}, doi = {10.1016/j.engappai.2025.111817}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111817}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A decision-making model with new Dice–Jaccard similarity measure: Green hydrogen electrolyzer selection}, volume = {160}, year = {2025}, }
• (2025). A personalized human–machine cooperative approach with transformer-based recognition for longitudinal and lateral control of intelligent vehicles. EAAI, 160, 111816. (www)

  @article{ , abstract = {Human–machine interaction brings challenges for vehicle control design due to individual differences, a personalized cooperative approach with driving style recognition is proposed to achieve lateral and longitudinal control of intelligent vehicles in this paper. An improved Transformer-based method with an unsupervised pre-training and window-based multi-head self-attention is proposed to enhance the recognition accuracy and speed of driving styles, and thereby to capture the controller parameters under various driving styles. To achieve the lateral and longitudinal control of human-machine cooperative system, an integrated driver–vehicle model is established by considering driving styles and vehicle planar dynamics. Then, a Takagi–Sugeno fuzzy controller is developed to handle time-varying parameters and eliminate human-machine conflicts. Especially, stability conditions are exploited by Lyapunov arguments to achieve the control objective. Finally, simulation results show that the designed Transformer-based method has better classification accuracy and computational efficiency than other baselines on the same dataset. Based on recognition results, the designed controller can effectively improve the driving performance under various driving styles and time-varying parameters compared with other methods.}, archive = {J_EAAI}, author = {Yan Ma and Jingjing Xie and Liang He and Kailong Zhang and Xiongmei Zeng and Quan Ouyang and Danwei Wang}, doi = {10.1016/j.engappai.2025.111816}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111816}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A personalized human–machine cooperative approach with transformer-based recognition for longitudinal and lateral control of intelligent vehicles}, volume = {160}, year = {2025}, }
• (2025). Incremental approach: A workflow model for semi-supervised learning environment. EAAI, 160, 111815. (www)

  @article{ , abstract = {Semi-supervised learning plays an important role by combining both labeled and unlabeled data, bridging the gap between supervised and unsupervised learning. In today’s fast-changing data streams, an efficient classifier is essential to adapt to evolving patterns. It is also important to address challenges like concept drift and concept evolution to maintain consistent model performance. However, handling unlabeled data in semi-supervised data streams can be challenging, as it requires significant resources and can be costly. This paper introduces an Incremental Workflow Model (IWM) that effectively handles the underlying issues of the data stream in a semi-supervised environment while using minimal labeled information. The proposed model is designed to be highly adaptable, ensuring it can handle the concept drift and concept evolution challenges. One-Class Classifier (OC_Classifier) with a proposed confidence score is used to identify the novel data and to predict known data in the stream. Further, MicroClusters are employed to reduce the memory overhead in the Ensemble Model. Experiments are performed on benchmark datasets, and results are analyzed to demonstrate the effectiveness of the proposed workflow model.}, archive = {J_EAAI}, author = {Maneesha Gudapati and Audrika Borah and Swarupa Rani K.}, doi = {10.1016/j.engappai.2025.111815}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111815}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Incremental approach: A workflow model for semi-supervised learning environment}, volume = {160}, year = {2025}, }
• (2025). Cooperative traffic signal control for a partially observed vehicular network using multi-agent reinforcement learning. EAAI, 160, 111813. (www)

  @article{ , abstract = {Cooperative traffic signal control aims to alleviate congestion and reduce travel times in urban environments by coordinating traffic lights. However, accurately modeling urban traffic dynamics and managing multiple intersections remain challenging tasks. Although Reinforcement Learning (RL) offers adaptability without requiring predefined traffic models, it faces challenges such as high dimensionality and data dependency, especially in real-world conditions like adverse weather or device malfunctions. In this paper, we propose a novel multi-tiered, multi-agent RL approach, named Multi-Agent Partially Observed Light (MAPOLight), designed to operate in a Vehicle-to-Infrastructure (V2I) environment with a limited number of Connected and Automated Vehicles (CAVs). The upper tier emphasizes agent collaboration by aggregating states and actions into mean values, significantly reducing state dimensionality. The lower tier employs advanced deep RL algorithms for optimization, ensuring both flexibility and scalability. Additionally, we introduce the Observation Correlation Indicator to relate CAV penetration rates (CAV P-Rates) with RL convergence. Simulation results reveal that our method offers stability and superior performance across diverse scenarios compared to traditional approaches. MAPOLight demonstrates robustness with CAV P-Rates above 20% and achieves convergence with a minimum penetration rate of 5%, outperforming existing methods. Moreover, our approach promotes smoother vehicle trajectories and exhibits strong adaptability, enabling rapid congestion relief in respond to unexpected traffic accidents. These results highlight the effectiveness and adaptability of the proposed approach.}, archive = {J_EAAI}, author = {Chong Wang and Yueqi Li and Jiale Chen and Jian Zhang and Yu Xue}, doi = {10.1016/j.engappai.2025.111813}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111813}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cooperative traffic signal control for a partially observed vehicular network using multi-agent reinforcement learning}, volume = {160}, year = {2025}, }
• (2025). Picture fuzzy geometric heronian mean operator based novel ranking method for optimal plastic waste recycling technology selection. EAAI, 160, 111812. (www)

  @article{ , abstract = {Recycling is becoming a viable solution to manage plastic wastes, because burning and land filling of plastic waste can have detrimental effects on the environment and human health. Keeping the importance of plastic waste recycling in view, the present study aims to develop a mathematical model to identify the best plastic waste recycling technique (PWRT) considering social, environmental and economical parameters. For the first time, the problem of PWRT selection has been modeled as a multi criteria group decision making (MCGDM) problem under picture fuzzy environment (PFE). To do so, firstly, geometric heronian mean operator (GHMO) and weighted GHMO (WGHMO) have been introduced under PFE by analyzing their properties. The concept of traditional method based on the removal effects of criteria (MEREC) has been extended under PFE using WGHMO to handle group decision making issues. Secondly, a novel ranking method has been developed introducing novel score function, which is able to rank all types of picture fuzzy numbers. The prominent traits of score function have been illustrated theoretically. Thirdly, several aspects of result analyses have been performed assessing robustness and integrity of the suggested technique. According to the comprehensive strategy that has been presented here, Open Loop Recycling has been identified as the best PWRT concerning less toxic gas emissions, low operational costs and environmental consequences, etc. The proposed score function enriched ranking method is computationally friendly requiring minimum number of computing operations. Moreover, the proposed approach is a practical tool for managing uncertainty that occurs in practical decision-making.}, archive = {J_EAAI}, author = {Chayel Tripura and Sayanta Chakraborty and Baby Bhattacharya}, doi = {10.1016/j.engappai.2025.111812}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111812}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Picture fuzzy geometric heronian mean operator based novel ranking method for optimal plastic waste recycling technology selection}, volume = {160}, year = {2025}, }
• (2025). Construction methods for entropy measures of circular intuitionistic fuzzy sets and their application. EAAI, 160, 111809. (www)

  @article{ , abstract = {The circular intuitionistic fuzzy set (C-IFS) is a recent extension of the intuitionistic fuzzy set, whose elements are represented as circles instead of specific orthopairs. Entropy measures provide us with a quantitative measure of uncertainty. This research addresses the gap in entropy measures for C-IFSs by introducing innovative construction methods. It establishes robust theoretical frameworks for entropy measures, leveraging established mathematical concepts including t-norms, t-conorms, automorphisms, and aggregation operators. It provides several proven mathematical results and expressions for entropy measures. Moreover, it is worth mentioning that entropy measures can be generated not only using t-norms and t-conorms alone but also when coupled with automorphisms and aggregation operators. Additionally, the argument formulation of t-norms and t-conorms plays a significant role in entropy measure generation, and these formulations are not unique, thereby paving the way for future research avenues. In addition to these contributions, a novel transformation method from entropy to a similarity measure is proposed. This transformation method fully incorporates all aspects of C-IFSs. Lastly, the technique for order preference by similarity to ideal solution (TOPSIS) is extended for C-IFSs to deal with multi-criteria decision-making problems and overcome its previous extensions’ limitations. This method consists of a novel criteria weight generation method that is entropy-based and the modified relative closeness index, adding further depth to the approach. Various numerical examples are given to elaborate on our results.}, archive = {J_EAAI}, author = {Muhammad Jabir Khan and Shu Jiang and Weiping Ding and Muhammad Akram}, doi = {10.1016/j.engappai.2025.111809}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111809}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Construction methods for entropy measures of circular intuitionistic fuzzy sets and their application}, volume = {160}, year = {2025}, }
• (2025). Deep transferable model with multi-head attention for small sample fault diagnosis of hydraulic pumps. EAAI, 160, 111808. (www)

  @article{ , abstract = {The hydraulic pump is the crucial power element of the hydraulic transmission system, directly impacting system reliability. Due to the inherent hidden nature of typical faults and the scarcity of data, it is challenging to extract valid fault characteristics from the acquired signals. To address this issue, an intelligent diagnostic method that combines a residual multi-head model with transfer learning is proposed. The model utilizes sufficient fault data under specific load conditions as the source domain for pre-training. Experimental results demonstrate that the model achieves over 99 % diagnostic accuracy, with approximately a 28 % reduction in training time compared to traditional methods, exhibiting both high precision and robustness. Secondly, different transfer strategies are adopted to share parameters from pre-trained models to different target domains with varying load conditions, dynamically adjusting network parameters to extract cross-condition features and perform fault diagnosis. Transfer experiments reveal that the accuracy of the model remains above 99 % under the 49 % data condition in the target domain, effectively mitigating data scarcity issues while improving training efficiency by about 15 %. The method provides practical guidance for the intelligent operation and maintenance of hydraulic pumps and shows great potential for application in industrial production.}, archive = {J_EAAI}, author = {Shengnan Tang and Zhengqi Yan and Yuhang Xue and Kian Meng Lim and Yong Zhu}, doi = {10.1016/j.engappai.2025.111808}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111808}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep transferable model with multi-head attention for small sample fault diagnosis of hydraulic pumps}, volume = {160}, year = {2025}, }
• (2025). Data-driven models with physical interpretability for real-time cavity profile prediction in electrochemical machining processes. EAAI, 160, 111807. (www)

  @article{ , abstract = {The Electrochemical Machining (ECM) process can be effectively controlled by adjusting the process parameters, including electrolyte composition, current density, etc. However, this control strategy presents challenges associated with modeling the complex relationships between processing parameters and the resulting process outcomes. Data-driven approaches hold promise for real-time cavity profile prediction, while their black-box nature limits the interpretability. To address these issues, this study initiated by developing several Machine Learning (ML) models to predict cavity profiles using process parameters, in-process data, or a combination of both. Subsequently, these models were inspected by explainable artificial intelligence (XAI) methods. Using process parameters for model inspection provides preliminary guidance for achieving desired outcomes, while inspections with in-process data shed light on process dynamics and enable diagnostics by detecting anomalies through atypical model focus. Linear regression (LR), with high interpretability, offers moderate predictive accuracy, whereas neural networks (NN) and convolutional neural networks (CNN) perform better yet require XAI methods to interpret their decision-making mechanisms. Model inspection was conducted on global and local levels. Global inspection via SHapley Additive exPlanations (SHAP) for LR and NN models identified factors affecting cavity size, aligning predictions with ECM knowledge. Local inspection through Gradient-weighted Class Activation Mapping (Grad-CAM) examined CNN predictions, revealing temporal process dynamics and model focus during specific processing stages. The impact of unexpected events identified by the ML models was validated using a physics-based model. A web application was developed to integrate these insights, allowing for real-time cavity prediction and visualization based on process parameters and in-process data.}, archive = {J_EAAI}, author = {Ming Wu and Zequan Yao and Mathias Verbeke and Peter Karsmakers and Benjamin Gorissen and Dominiek Reynaerts}, doi = {10.1016/j.engappai.2025.111807}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111807}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven models with physical interpretability for real-time cavity profile prediction in electrochemical machining processes}, volume = {160}, year = {2025}, }
• (2025). Secure industrial federated learning: Label encryption for model protection. EAAI, 160, 111806. (www)

  @article{ , abstract = {Amid the widespread adoption of deep learning techniques in the Industrial Internet of Things, unauthorized extraction of trained deep learning models has emerged as a significant threat to model intellectual property protection. Existing intellectual property protection methods, such as neural network watermarking and fingerprinting, primarily focus on passive tracing, lacking proactive prevention capabilities. In this paper, we propose an encryption-based intellectual property protection framework specifically designed for federated learning scenarios in industrial settings. The primary artificial intelligence-related contribution of this framework is the design of an efficient label encryption scheme, which selectively encrypts only label information rather than entire datasets or model parameters, significantly reducing computational and communication overhead while preserving model accuracy. The primary engineering application involves integrating encryption mechanisms into the federated learning training and deployment processes to ensure proactive access control and robust passive traceability. The proposed framework employs a hierarchical access control protocol leveraging client-specific encryption keys, providing active prevention against unauthorized model use and facilitating forensic evidence collection. Additionally, the encryption mechanism protects client data privacy and clearly establishes model ownership. Through comprehensive experiments and analyses, we demonstrate that the proposed encryption-based framework effectively safeguards model intellectual property, preserves model performance, and achieves robustness suitable for resource-constrained industrial environments.}, archive = {J_EAAI}, author = {Xuemei Yuan and Hewang Nie}, doi = {10.1016/j.engappai.2025.111806}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111806}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Secure industrial federated learning: Label encryption for model protection}, volume = {160}, year = {2025}, }
• (2025). Enhanced knowledge graph cascade learning model for cyber–physical systems. EAAI, 160, 111802. (www)

  @article{ , abstract = {Recently, the application prospects of knowledge graph technology in cyber–physical systems (CPS) have attracted considerable attention. However, knowledge graph data in various CPS domains are typically collected from sensors or through manual efforts, which inevitably results in incomplete and unreliable data, thereby impacting the performance of downstream task models. This issue is often overlooked in existing studies. This paper proposes an enhanced knowledge graph cascade learning model for CPS. The model performs cascaded and iterative learning of both graph structure and graph representation. By optimizing the graph structure and incorporating hierarchical learning of graph-structured information, the proposed model enhances feature propagation and aggregation during representation learning. Experiments show that our model achieves outstanding results: compared to the baseline models, our approach achieves an average improvement of 2.7% in accuracy on the node classification task and 1.35% in MRR on the link prediction task.}, archive = {J_EAAI}, author = {Shumao Zhang and Jie Xu and Haodiao Xie and Qiuru Fu and Ke Miao and Shixue Cheng and Zelei Wu}, doi = {10.1016/j.engappai.2025.111802}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111802}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced knowledge graph cascade learning model for cyber–physical systems}, volume = {160}, year = {2025}, }
• (2025). Advanced deep learning-based methodology for multi-class diagnosis of wind turbine blade faults at the wind farm level. EAAI, 160, 111801. (www)

  @article{ , abstract = {Accurately diagnosing wind turbine blade faults is crucial for the efficient operation of wind farms. Recent advancements in deep learning have significantly improved fault detection and diagnosis techniques. However, a major challenge in current research is the limited generalizability of models trained on data from a single turbine, which often proves ineffective when applied to the broader context of an entire wind farm. To addressing this limitation, this work introduces a novel data-driven multi-fault diagnosis system tailored for application at the wind farm level. Our approach proposes a dropout-enhanced bidirectional long short-term (MBiLSTM-D) model that enhances generalizability across various turbines within a wind farm. Moreover, a focal loss function is employed to address data imbalance. To address the incompleteness and redundancy of the raw data, we also propose a series of innovative data preprocessing methods to handle missing values and useless information in the data. Extensive validation using supervisory control and data acquisition (SCADA) data from a real-world wind farm demonstrates that our system effectively overcomes the generalization shortcomings of previous models.}, archive = {J_EAAI}, author = {Chunchen Wei and Shimin Cai and Han Yang and Yanru Zhang}, doi = {10.1016/j.engappai.2025.111801}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111801}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advanced deep learning-based methodology for multi-class diagnosis of wind turbine blade faults at the wind farm level}, volume = {160}, year = {2025}, }
• (2025). EnTAIL: Evolutional temporal-aware interaction learning for motion forecasting. EAAI, 160, 111800. (www)

  @article{ , abstract = {Accurately predicting the future trajectories of traffic agents in real-world scenarios is critical for advancing intelligent cyber–physical systems (CPS), such as autonomous driving systems and smart cities. A fundamental challenge lies in mining the evolving interaction patterns among multiple agents from their past trajectories, as traffic scenarios often exhibit complex interactions and continuously evolve along the timeline. However, existing methods fail to fully exploit the temporality inherent in sequential interactions. In the process of modeling interactions, they lack a comprehensive understanding of static interactions that occur at constant timestamps and the evolving patterns of interactions across timestamps. To tackle these challenges, we propose E volutio n al T emporal- A ware I nteraction L earning ( EnTAIL ), a novel temporal-aware interaction learning framework to model and reason the interactions among agents. EnTAIL captures both static interaction patterns at individual timestamps and temporal-aware interaction patterns across timestamps through a unified framework. Specifically, we introduce a trainable constant time encoding to integrate with the interaction modeling in each individual timestamp, which aims to capture the static interaction information. We propose a dynamic evolution encoder to model temporal-aware interaction features, enabling learning both short-term and long-term interactions within multiscaled observation windows. Besides, EnTAIL also considers the temporal feature in the prediction stage and models the long-range interactions ignored during the encoding phase. Extensive experiments conducted on the challenging real-world Argoverse dataset demonstrate that our proposed model achieves substantial performance improvement and outperforms the baseline methods up to 2.5% in minimum Average Displacement Error (minADE) and 1.2% in minimum Final Displacement Error (minFDE).}, archive = {J_EAAI}, author = {Chunyu Liu and Hao Dong and Pengyang Wang and Jianjun Yu}, doi = {10.1016/j.engappai.2025.111800}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111800}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {EnTAIL: Evolutional temporal-aware interaction learning for motion forecasting}, volume = {160}, year = {2025}, }
• (2025). A review of dynamic flexible regulation strategies for multi-energy coupled steelmaking-continuous casting production. EAAI, 160, 111799. (www)

  @article{ , abstract = {The Steelmaking-Continuous Casting (SCC) process is a critical bottleneck in steel production, involving complex energy couplings and flexible constraints, requiring adaptation to dynamic production conditions. This paper explores dynamic regulation strategies for SCC under multi-energy coupling, aiming to optimize and innovate steel production regulation. It addresses the challenges of regulating SCC in multi-energy environments, focusing on energy flow principles and the conditions needed for stable and efficient operation in disturbed environments. Unlike existing reviews, this paper distinguishes itself explicitly by emphasizing the research gap in integrating the energy coupling characteristics of the steelmaking-continuous casting process with dynamic, flexible regulation strategies and discusses the potential of this integration in improving production efficiency and energy management. The paper further explores specific multi-energy coupling patterns and regulation goals, highlighting the applicability of operations research, heuristic approaches, and machine learning methods in solving optimization challenges. These insights provide valuable references for future research in this field. Lastly, this paper identifies limitations in existing methods, such as insufficient model accuracy and adaptability, and proposes solutions for overcoming these bottlenecks through technological and theoretical advancements, enabling more innovative and automated regulation in SCC. In conclusion, the paper links energy coupling characteristics with dynamic regulation strategies, advancing understanding of SCC mechanisms and highlighting the need for future improvements in modeling and optimization.}, archive = {J_EAAI}, author = {Liangliang Sun and Hao Li and Zhenghao Song and Ge Guo and Changyu Wang and Natalia M. Matsveichuk and Yuri N. Sotskov}, doi = {10.1016/j.engappai.2025.111799}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111799}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A review of dynamic flexible regulation strategies for multi-energy coupled steelmaking-continuous casting production}, volume = {160}, year = {2025}, }
• (2025). An invulnerable leader–follower collision-free unmanned aerial vehicle flocking system with attention-based multi-agent reinforcement learning. EAAI, 160, 111797. (www)

  @article{ , abstract = {Deep reinforcement learning has been proved useful for the flocking control of Unmanned Aerial Vehicle (UAV) swarm with the leader–followers topology. However, it remains unclear how to fully utilize the spatial information among the followers to alleviate the problems of sparse reward and policy convergence. In this article, we propose a novel multi-agent reinforcement learning-based fixed-wing UAV flocking approach named Attention Based Cucker–Smale (ABCS) Flocking to learn collision-free leader–follower flocking by utilizing the information among followers with the attention mechanism. Specifically, we design an explainable Cucker–Smale criterion-based flocking reward named ABCS reward to associate the followers with high efficiency in flocking. Then, a leader-guide attention mechanism is proposed by transferring the difference between leader and follower as the weights to support the follower selectively utilizing the followers’ information. As a result, we prove that an optimal state can be achieved so that each follower can keep an optimal distance from the other followers when the ABCS reward is maximized. In addition, we prove the ABCS reward is bounded, which can be used to indicate learning convergence. To improve the invulnerability, we proposes a leader selection method based on ABCS flocking, which can effectively select a new leader when the old leader is destroyed. Finally, we demonstrate the effectiveness of ABCS Flocking over the Multi-Agent Deep Deterministic Policy Gradient approach using various reward functions with various numbers of followers in the obstacles and leader-destroyed scenarios. The code is published on github https://github.com/YunxiaoGuo/ABCS-Flocking .}, archive = {J_EAAI}, author = {Yunxiao Guo and Dan Xu and Chang Wang and Jinxi Li and Han Long}, doi = {10.1016/j.engappai.2025.111797}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111797}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An invulnerable leader–follower collision-free unmanned aerial vehicle flocking system with attention-based multi-agent reinforcement learning}, volume = {160}, year = {2025}, }
• (2025). Enhancing fundus image diabetic retinopathy classification through modified conformer with sparse attention. EAAI, 160, 111795. (www)

  @article{ , abstract = {As a prevalent ocular disorder, diabetic retinopathy poses significant challenges in accurate and timely diagnosis through fundus image classification. Current classification approaches frequently encounter obstacles in attaining high accuracy and efficiency. Therefore, the primary purpose of this study is to address these limitations and enhance the classification performance. The methodology encompasses an elaborate design of a modified Conformer architecture. Specifically, within the modified Conformer, meticulous adjustments are made to incorporate the sparse attention layer. This layer is engineered to selectively concentrate on pertinent features within the fundus images, thereby enhancing the model’s discriminatory power. For the proposed model, the training process employs advanced optimization algorithms and judiciously selected hyperparameters. The model is evaluated using a set of well-established public datasets of fundus images, with an extensive array of performance metrics such as accuracy, recall, precision, and F1-score. The experimental outcomes demonstrate that the proposed model surpasses conventional classification methods in terms of various evaluation metrics. Accordingly, this research offers a novel and efficacious approach for diabetic retinopathy fundus image classification.}, archive = {J_EAAI}, author = {Jian Lian and Jiafu Ji and Yawen Niu and Wanzhen Jiao}, doi = {10.1016/j.engappai.2025.111795}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111795}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing fundus image diabetic retinopathy classification through modified conformer with sparse attention}, volume = {160}, year = {2025}, }
• (2025). A graph-based state representation learning for episodic reinforcement learning in task-oriented dialogue systems. EAAI, 160, 111793. (www)

  @article{ , abstract = {Recent research in dialogue state tracking has made significant progress in tracking user goals using pretrained language models and context-driven approaches. However, existing work has primarily focused on contextual representations, often overlooking the structural complexity and topological properties of state transitions in episodic reinforcement learning tasks. In this study, we introduce a cutting-edge, dual-perspective state representation approach that provides a dynamic and inductive method for topological state representation learning in episodic reinforcement learning within task-oriented dialogue systems. The proposed model extracts inherent topological information from state transitions in the Markov Decision Process graph by employing a modified clustering technique to address the limitations of transductive graph representation learning. It inductively captures structural relationships and enables generalization to unseen states. Another key innovation of this approach is the incorporation of dynamic graph representation learning with task-specific rewards using Temporal Difference error. This captures topological features of state transitions, allowing the system to adapt to evolving goals and enhance decision-making in task-oriented dialogue systems. Experiments, including ablation studies, comparisons with existing approaches, and interpretability analysis, reveal that the proposed model significantly outperforms traditional contextual state representations, improving task success rates by 9%–13% across multiple domains. It also surpasses state-of-the-art Q-network-based methods, enhancing adaptability and decision-making in domains such as movie-ticket booking, restaurant reservations, and taxi ordering.}, archive = {J_EAAI}, author = {Yasaman Saffari and Javad Salimi Sartakhti}, doi = {10.1016/j.engappai.2025.111793}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111793}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A graph-based state representation learning for episodic reinforcement learning in task-oriented dialogue systems}, volume = {160}, year = {2025}, }
• (2025). Artificial intelligence assisted non-destructive testing of welding joints: A review of techniques, X-ray image processing and applications. EAAI, 160, 111792. (www)

  @article{ , abstract = {Non-Destructive Testing (NDT) is an appealing technique for verifying welding quality and ensuring structural integrity without causing damage. It assists in regulatory compliance, risk mitigation, and process optimization for improved industrial reliability. This study reviews potential NDT approaches. Through a rigorous analysis of existing surveys, we aim to decipher the current landscape and highlight the significant advancements in the field. Because of the potential of artificial intelligence (AI) assistive X-ray imaging-based NDT, we particularly examine the integration of AI algorithms and X-ray imaging in the NDT of welds. This convergence represents a paradigm shift, redefining traditional methods and ushering in a new era of precision, automation, and efficiency. As we navigate through potential X-ray imaging datasets, delve into crucial image processing techniques, examine feature extraction methods, and explore AI algorithms, our survey reveals the intricate interplay of technologies that drive automated weld defect categorization, size detection, and segmentation. The review broadens its focus by including various practical applications, highlighting the adaptable utility of AI-assistive X-ray imaging for weld defect detection in potential industrial applications. Moreover, the promising future developments, opportunities, and nuanced challenges associated with integrating X-ray imaging and AI in weld integrity assessment are highlighted. It provides a comprehensive perspective on this rapidly evolving field.}, archive = {J_EAAI}, author = {Dalila Say and Saeed Mian Qaisar and Moez Krichen and Salah Zidi}, doi = {10.1016/j.engappai.2025.111792}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111792}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence assisted non-destructive testing of welding joints: A review of techniques, X-ray image processing and applications}, volume = {160}, year = {2025}, }
• (2025). Advancing automatic photovoltaic defect detection using semi-supervised semantic segmentation of electroluminescence images. EAAI, 160, 111790. (www)

  @article{ , abstract = {Photovoltaic (PV) systems allow us to tap into all abundant solar energy, however they require regular maintenance for high efficiency and to prevent degradation. Traditional manual health check, using Electroluminescence (EL) imaging, is expensive and logistically challenging which makes automated defect detection essential. Current automation approaches require extensive manual expert labeling, which is time-consuming, expensive, and prone to errors. We propose PV-S3 ( P hoto v oltaic- S emi-supervised S emantic S egmentation), a Semi-Supervised Learning approach for semantic segmentation of defects in EL images that reduces reliance on extensive labeling. PV-S3 is an artificial intelligence (AI) model trained using a few labeled images along with numerous unlabeled images. We introduce a novel Semi Cross-Entropy loss function to deal with class imbalance. We evaluate PV-S3 on multiple datasets and demonstrate its effectiveness and adaptability. With merely 20% labeled samples, we achieve an absolute improvement of 9.7% in mean Intersection-over-Union (mIoU), 13.5% in Precision, 29.15% in Recall, and 20.42% in F1-Score over prior state-of-the-art supervised method (which uses 100% labeled samples) on University of Central Florida-Electroluminescence (UCF-EL) dataset (largest dataset available for semantic segmentation of EL images) showing improvement in performance while reducing the annotation costs by 80%. For more details, visit our GitHub repository: https://github.com/abj247/PV-S3 .}, archive = {J_EAAI}, author = {Abhishek Jha and Yogesh Rawat and Shruti Vyas}, doi = {10.1016/j.engappai.2025.111790}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111790}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advancing automatic photovoltaic defect detection using semi-supervised semantic segmentation of electroluminescence images}, volume = {160}, year = {2025}, }
• (2025). A communicability-driven expert influence model for large-scale group decision-making based on complex network theory. EAAI, 160, 111789. (www)

  @article{ , abstract = {This paper proposes a novel large-scale group decision-making (LSGDM) framework based on complex network theory. First, by enhancing the clustering efficiency of the Louvain algorithm through complex network theory, this framework enables community classification of a large-scale expert group, thereby reduce the complexity of the expert network and improve decision-making efficiency. Second, a novel communicability-driven expert influence identification model within expert social networks is designed, and a new consensus feedback mechanism is proposed. This approach comprehensively considers both the individual influence of experts and the impact of their interactions to guide consensus-reaching, which can accurately reflect the actual flow of information within the network. Third, by extending the Best–Worst Method and Shannon Entropy Method to the context of fuzzy information, a balance between subjectivity and objectivity is obtained, which can make more accurate and interpretable decision results. The proposed method is validated through a case study on green material selection. Moreover, its superiority is further demonstrated through comparative experiments. The results show that our method demonstrates a 9.89% reduction in opinion revision costs with accelerated consensus-reaching efficiency.}, archive = {J_EAAI}, author = {Fenglan Sun and Yuteng Yi and Wei Zhu and Jürgen Kurths}, doi = {10.1016/j.engappai.2025.111789}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111789}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A communicability-driven expert influence model for large-scale group decision-making based on complex network theory}, volume = {160}, year = {2025}, }
• (2025). Finding the underlying viscoelastic constitutive equation via universal differential equations and differentiable physics. EAAI, 160, 111788. (www)

  @article{ , abstract = {Determining the appropriate constitutive model to describe the behavior of a given material is a fundamental, yet challenging, aspect of rheology. While data-driven methods present a promising path for refining these models, a more in-depth investigation into the capabilities and limitations of emerging techniques is required. This research addresses this gap by employing Universal Differential Equations (UDEs) and differentiable physics to model viscoelastic fluids, merging conventional differential equations with neural networks to reconstruct missing terms in constitutive models. This study focuses on analyzing four viscoelastic models, Upper Convected Maxwell (UCM), Johnson–Segalman, Giesekus, and Exponential Phan–Thien–Tanner (ePTT) using synthetic datasets. The methodology was tested across different experimental conditions, including oscillatory and startup flows. Relative error analyses revealed that the UDEs framework maintains low and stable errors (below 0.3%) for the UCM, Johnson–Segalman, and Giesekus models under various conditions, while exhibiting higher but consistent errors (4%) for the ePTT model due to its strong nonlinearity. These findings highlight the potential of UDEs in fluid mechanics while also identifying critical areas for methodological improvement. Additionally, a model distillation approach was employed to extract simplified models from complex ones, emphasizing the versatility and robustness of UDEs in rheological modeling.}, archive = {J_EAAI}, author = {Elias C. Rodrigues and Roney L. Thompson and Dário A.B. Oliveira and Roberto F. Ausas}, doi = {10.1016/j.engappai.2025.111788}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111788}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Finding the underlying viscoelastic constitutive equation via universal differential equations and differentiable physics}, volume = {160}, year = {2025}, }
• (2025). Factors analysis of commercial bank digital transformation with random neural network algorithm. EAAI, 160, 111787. (www)

  @article{ , abstract = {With the deepening of global economic and technological integration, digital transformation has emerged as a key factor for enhancing operational efficiency, improving customer experience, and reshaping the strategic and organizational frameworks of commercial banks. Existing literature often treats digital transformation as a derivative process influenced by external technological trends. This paper argues that digital transformation is an original and strategic operation that requires user-centric design and adaptive technological integration. By leveraging an independently developed digital platform, and employing artificial intelligence algorithms and optimization models, this study constructs a comprehensive model to analyze the key factors influencing a bank’s digital transformation from the perspective of considering user-platform interaction. Experimental results show that, in terms of interpretability, the proposed model demonstrates a 9%/7% improvement in R 2 and adjusted R 2 , respectively. In terms of prediction, the proposed model shows a clear advantage, providing more accurate and stable results. The findings from this study offer a scientifically grounded reference for commercial banks, enabling them to better understand and address the challenges and opportunities associated with digital transformation, ultimately leading to more efficient and sustainable development.}, archive = {J_EAAI}, author = {Yixi Ding and Ju Wei and Bo Pang and Ling Tong}, doi = {10.1016/j.engappai.2025.111787}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111787}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Factors analysis of commercial bank digital transformation with random neural network algorithm}, volume = {160}, year = {2025}, }
• (2025). Attention based network for real-time road drivable area, lane line detection and scene identification. EAAI, 160, 111781. (www)

  @article{ , abstract = {The detection of road drivable areas and lane lines is considered a fundamental component of autonomous driving systems. However, most existing approaches handle these tasks independently, and multi-task networks frequently neglect the inherent correlation between them while failing to differentiate various lane line types. In practice, the delineation of drivable regions is strongly influenced by both lane line characteristics and contextual street scenes. To address these limitations, a novel multi-task network—Real-time Road Drivable Area, Lane Line Detection, and Scene Identification Network (RLSNet)—is proposed. This network is designed to perform simultaneous segmentation of drivable areas, detection of lane lines, and classification of road scenes. Drivable area estimation is optimized through the integration of lane and scene cues, guided by traffic regulations. A Residual Network (ResNet)-based backbone is employed, enhanced with Bidirectional Fusion Attention (BFA) for feature encoding. This is followed by a decoder incorporating a Feature Aggregation Module (FAM) to enable effective semantic–spatial fusion. Lane line detection is further refined using a Bilateral Up-Sampling Decoder (BUSD), while scene understanding is enhanced via a Scene Classification Module (SCM). Extensive experiments conducted on the challenging Berkeley DeepDrive 100K(BDD100K) dataset have demonstrated that RLSNet achieves high accuracy in both drivable area and lane line detection by leveraging the mutual guidance of lane and scene information. Furthermore, the network maintains real-time inference speed at 93 frames per second (FPS), striking a practical balance between semantic fidelity and computational efficiency for real-world deployment. The implementation code has been made publicly available at: https://github.com/033186ZSY/RLSNet-master .}, archive = {J_EAAI}, author = {Feng You and Yi Xie and Siyi Zhang and Hao Chen and Haiwei Wang and Wei Zhang and Jianrong Liu}, doi = {10.1016/j.engappai.2025.111781}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111781}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Attention based network for real-time road drivable area, lane line detection and scene identification}, volume = {160}, year = {2025}, }
• (2025). BiKD2: Bidirectional knowledge distillation-enhanced explicit graph disentangling network for review-based recommendation. EAAI, 160, 111779. (www)

  @article{ , abstract = {Reviews, serving as textual vessels for expressing subjective user preferences and actual item attributes, are pivotal in review-based recommender systems. However, prior research has overlooked the fact that reviews contain a significant portion of content irrelevant to user interests (referred to as review noise). This oversight allows review noise to propagate unhindered through the information processing pipeline of existing methodologies, resulting in contaminated representations. Notably, this issue will be exacerbated in graph methods where message propagate more frequently. To fill this gap, this paper proposes a Bi directional K nowledge D istillation-enhanced explicit graph D isentangling network (BiKD2) for review-based recommendation. First, an explicit disentanglement-based graph convolution network is developed as a representation generator to block the cross-rating propagation of review noise. Next, a bidirectional distillation strategy is proposed to transfer preference-related pure knowledge from reviews, where the reverse distillation direction aims to extract finer-grained generic features and differential features. Extensive experiments demonstrate the superiority of the proposed BiKD2 and the effectiveness of its individual components. The complete codes are available at https://github.com/kangliu1225/BiKD2 .}, archive = {J_EAAI}, author = {Kang Liu and Tongtong Yu and Han Huang and Yueli Song and Yuling Li}, doi = {10.1016/j.engappai.2025.111779}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111779}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {BiKD2: Bidirectional knowledge distillation-enhanced explicit graph disentangling network for review-based recommendation}, volume = {160}, year = {2025}, }
• (2025). Adaptive image-based visual servoing gain using reinforcement learning for robotic arm. EAAI, 160, 111778. (www)

  @article{ , abstract = {Traditional image-based visual servoing (IBVS) methods face significant challenges when the current camera pose differs substantially from the desired pose, particularly in the presence of rotational errors along or around the optical axis. Singular points in the image Jacobian matrix can lead to servoing failures during robot movement, limiting the effectiveness of conventional IBVS approaches. To address these issues, this study proposes a novel reinforcement learning framework for adaptively adjusting the image Jacobian matrix in real-time. The image plane is discretized into a state space based on the locations of feature points, while the action space consists of linear combinations of the image Jacobian matrix. The learning agent selects actions using an ϵ -greedy policy based on the current state and receives rewards from the environment to update its policy, gradually converging to an optimal solution through iterative interactions. The proposed method is validated using a seven-axis manipulator equipped with a single-lens camera, and its performance is compared to a conventional IBVS system with a fixed image Jacobian matrix. The experimental results demonstrate the superiority of the adaptive approach in terms of robustness, convergence speed, and overall servoing accuracy.}, archive = {J_EAAI}, author = {Maxwell Hwang and Yu-Jen Chen and Wei-Cheng Jiang and Jia-Lin Lee}, doi = {10.1016/j.engappai.2025.111778}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111778}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive image-based visual servoing gain using reinforcement learning for robotic arm}, volume = {160}, year = {2025}, }
• (2025). A lightweight detection and recognition framework for cigarette laser code. EAAI, 160, 111777. (www)

  @article{ , abstract = {Automatic detection and recognition of laser codes on cigarette case is important in distinguishing the authenticity of cigarettes. However, detecting and recognizing cigarette laser codes presents a challenging industrial problem due to the intricate background of cigarette images. In this paper, a Lightweight Detection and Recognition Framework (LDRF) is proposed to detect and recognize cigarette laser code. The LDRF model consists of the Lightweight Detection Network (LDNet) and the Lightweight Recognition Network (LRNet). In the LDNet stage, a lightweight feature extraction network is proposed to extract features of the cigarette code area. Furthermore, a bidirectional feature pyramid network (BiFPN) feature fusion structure is introduced to tackle the multi-scale feature fusion challenge in cigarette code detection scenarios. Notably, alignment and normalization of all features channels are conducted to reduce computational requirements in the post-processing stage, enabling precise detection performance while significantly reducing parameters and computational complexity. In the LRNet stage, an integrated network architecture is designed to enhance the fusion of visual and temporal features. Furthermore, a combination of bidirectional temporal convolutional network (BiTCN) and Transformer is employed in the feature extraction stage to differentiate between background and characters, as well as capture the interdependence among different characters. Specifically, DownSampling is utilized to adjust the size of input images and Merging or Combining methods are applied at each stage to capture multi-level features. Experimental results demonstrate that the proposed LDRF method provides better performance than state-of-the-art models, and achieves trade-off between accuracy and speed.}, archive = {J_EAAI}, author = {Wei Zhou and Honggang Li and Li Zheng and Shan Xiao and Jun Yi}, doi = {10.1016/j.engappai.2025.111777}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111777}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight detection and recognition framework for cigarette laser code}, volume = {160}, year = {2025}, }
• (2025). Collaborative strategy for hybrid actions of radar modes and maneuver decisions under observation errors. EAAI, 160, 111774. (www)

  @article{ , abstract = {The rapid advancement of airborne avionics has driven modern air combat to rely heavily on information-centric operations, with radar serving as a primary tool for information acquisition and playing a critical role in air combat. However, existing research on air combat strategies often overlooks the impact of different radar operating modes on maneuvering strategies, as well as the challenges posed by learning strategies under observational disturbances. To address these gaps, this study investigates the problem of hybrid actions decision-making for radar modes and maneuver decisions in the presence of observational errors. Specifically, the characteristics of various radar operating modes are analyzed and modeled, followed by an exploration of the convergence process of reinforcement learning strategies under observational disturbances. To mitigate the instability and volatility in strategy learning caused by observation errors, Entropy-Decoupling-Noisy-net Proximal Policy Optimization-Advanced (EDN-PPOA) algorithm is proposed, which significantly enhances the robustness and exploratory capability of the model. Simulation results demonstrate that the proposed algorithm effectively achieves coordinated tactical integration of radar modes and maneuvers in complex hybrid action spaces, producing flexible tactical strategies that outperform expert-designed heuristics. Furthermore, compared to the existing algorithms, the proposed method exhibits superior stability and robustness in noisy observational environments, providing a reliable technical foundation for intelligent decision-making in complex adversarial scenarios.}, archive = {J_EAAI}, author = {Xingyu Wang and Zhen Yang and Jichuan Huang and Bao Zhang and Yuhe Zhang and Deyun Zhou}, doi = {10.1016/j.engappai.2025.111774}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111774}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Collaborative strategy for hybrid actions of radar modes and maneuver decisions under observation errors}, volume = {160}, year = {2025}, }
• (2025). Text-based image retrieval system using semantic visual content for re-ranking. EAAI, 160, 111770. (www)

  @article{ , abstract = {We address the problem of retrieving relevant images from the web in response to a text query. Text-based image retrieval is a challenging multimodal problem that requires understanding of the user intent from a few keywords and a semantic understanding of the images to predict relevance. In this paper, we present the hybrid search approach of the Turkcell-Yaani Search Engine, which was the leading search engine for queries in the Turkish language in 2017–2022. With over 1 million monthly users and processing millions of requests per month, its hybrid image search approach demonstrates the practical relevance and effectiveness of integrating text and visual information for enhanced image retrieval. The initial stage of the system focuses on retrieving images by comparing the query text with textual information associated with previously crawled and indexed images. The second stage aims to address the limitations of relying solely on text-based information by incorporating the post-processing steps based on semantic analysis of retrieved images. This stage, which is the main contribution of this paper, involves clustering, filtering and reordering the retrieved images using their embedded representations that are obtained from pretrained deep neural networks. The experimental results show that incorporating visual content improves the performance of the text-based retrieval engine significantly.}, archive = {J_EAAI}, author = {Berkay Topcu and Alper Mıtıncık and Merve Gülnaz Erdem and Berrin Yanikoglu}, doi = {10.1016/j.engappai.2025.111770}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111770}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Text-based image retrieval system using semantic visual content for re-ranking}, volume = {160}, year = {2025}, }
• (2025). Energy-derivative attention enhanced deep learning for multi-phase segmentation of mesoscale heterogeneous material using X-ray computed tomography images. EAAI, 160, 111768. (www)

  @article{ , abstract = {The precise and autonomous segmentation of mesoscale phases in heterogeneous materials remains challenging due to the similarity in characterization between holes and fractures. To address this issue, a neural network for computed tomography of multi-phase composites (MCCTNet) is established with an innovative architecture of adaptable configurations to recognize pixel-level mesoscale phases using X-ray Computed Tomography (X-CT) images of composites. Based on the original U-Net-like encoder-decoder structure, M to N layers are integrated with a novel attention module, termed the energy-derivative attention module (EDAM), which is designed to learn explicit feature representations for regional energy and boundary geometry . A pixel-level labeled dataset with 600 X-CT images covering diverse phases was established. The effectiveness of the proposed method and its superiority over existing methods were validated through comparative studies and ablation tests. EDAM significantly improves the recognition of small region-of-interests (RoIs), achieving in an improvement of 2.29 %, 1.16 %, 0.92 %, and 0.66 % for fracture, hole, cement paste, and aggregate, respectively. In addition, the proposed MCCTNet-1-4 embedded with EDAM demonstrated robust and consistent segmentation accuracy under Gaussian, salt-and-pepper, and speckle noise. Finally, practical applications including two-dimensional analysis, uniaxial compression simulations, and three-dimensional reconstruction based on the multi-phase segmentation results were conducted to verify the proposed method.}, archive = {J_EAAI}, author = {Xin Jing and Yu Wang and Yixuan Huan and Kaiyu Guo and Jiaqi Dong and Zhanxiong Ma and Yang Xu and Qiangqiang Zhang}, doi = {10.1016/j.engappai.2025.111768}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111768}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Energy-derivative attention enhanced deep learning for multi-phase segmentation of mesoscale heterogeneous material using X-ray computed tomography images}, volume = {160}, year = {2025}, }
• (2025). Multimodal data fusion-based intelligent fault diagnosis for ship rotating machinery: Status quo and perspectives. EAAI, 160, 111767. (www)

  @article{ , abstract = {Ships are indispensable to global transportation and play a critical role in fostering economic growth and cultural exchange. Central to ship operations is the rotating machinery, which, due to the complex and demanding conditions at sea, is susceptible to faults posing a serious threat to the safety of ship navigation. Therefore, timely and effective fault detection and diagnosis are essential for minimizing operational interruptions and maintaining the safety of ship navigation. Fault diagnosis methods based on multimodal data fusion (MDF) strategies can fully utilize the monitoring capabilities of different modal signals for different types of faults, significantly improving diagnostic accuracy, and have attracted widespread attention from researchers in recent years, yielding promising results. Nevertheless, there is currently a notable lack of a comprehensive review of multimodal data fusion approaches. This paper addresses this gap by conducting an exhaustive review of existing literature on multimodal data fusion technology, focusing on fusion methods. It discusses their practical applications in diagnosing faults in ship rotating machinery, analyzes current challenges, and outlines future research directions. This comprehensive synthesis aims to serve as a key reference for researchers in the field, guiding future developments in fault diagnosis technologies.}, archive = {J_EAAI}, author = {Yaqiong Lv and Jian Hao and Min Tang and Jun Wu}, doi = {10.1016/j.engappai.2025.111767}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111767}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodal data fusion-based intelligent fault diagnosis for ship rotating machinery: Status quo and perspectives}, volume = {160}, year = {2025}, }
• (2025). Parameter-efficient fine-tuning for citrus fruit defect segmentation model based on text prompt. EAAI, 160, 111765. (www)

  @article{ , abstract = {Citrus fruits are a diverse and nutritious group, and their detection of cosmetic defects is an important part of quality grading in the sorting line. The detection of fruit defects has become more efficient with the use of deep learning, which is a customized technology due to the different appearance characteristics of citrus fruit defects. The urgent problem to be solved is how to make it adapt quickly to different tasks. Our approach builds on the Low-Rank adaptation (LoRA) approach by incorporating a textual prompt module to introduce greater nonlinearity. Additionally, our method proposes modules tailored to convolutional neural networks (CNNs) for downstream segmentation tasks. The utilization of toggle text prompts facilitates the switching between models for disparate datasets. This provides a Parameter Efficient Fine-Tuning (PEFT) method for fast transfer of segmentation models. Our method adds only 0.7% additional parameters to the base segmentation model, reducing training costs and eliminating the need to update pre-training model parameters while incorporating more nonlinearities, achieving significantly higher mean intersection over union (mIoU) values of 77.25%, 64.32%, and 68.69% on three different citrus fruit defect segmentation datasets, outperforming other PEFT methods and approaching the performance of full fine-tuning. This is the first time the PEFT method has been applied to the field of fruit defect detection, reducing the process and cost of model development and providing a unified and efficient framework for rapid model transfer. Data and code are available at https://github.com/caixiongjiang/TP-LoRA .}, archive = {J_EAAI}, author = {Yun Zhu and Xiongjiang Cai and Shuwen Liu and Zhiyue Yu and Lianfeng Gao and Youyun Xu}, doi = {10.1016/j.engappai.2025.111765}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111765}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Parameter-efficient fine-tuning for citrus fruit defect segmentation model based on text prompt}, volume = {160}, year = {2025}, }
• (2025). Weighted C-type random 2 satisfiability in discrete hopfield neural network. EAAI, 160, 111760. (www)

  @article{ , abstract = {Implementing satisfiability logic rules in the Discrete Hopfield Neural Network provides a promising approach for modeling intelligent behavior in artificial intelligence. However, existing methods often exhibit limited flexibility and insufficient dynamic adaptability, thereby restricting their practical performance and generalization capabilities. To address these limitations, this study proposes a novel logical rule named Weighted C-Type Random 2 Satisfiability in Discrete Hopfield Neural Network, which dynamically generates the distribution of negative literals and randomly varies the order of clauses. Furthermore, a new learning strategy based on the Binary Artificial Bee Colony algorithm is introduced to optimize logic structure generation. Extensive experimental evaluations demonstrate that the proposed logical rule achieves high dynamism and flexibility by regulating the negative literal distribution and varying both first-order and second-order clauses. Comparative experiments against state-of-the-art logical rules show that the model improves the efficiency of attaining optimal synaptic weights by approximately 32.95 %, enhances the identification of global minimum solutions by around 23.06 %, and increases the diversity of final neuron states by about 57.87 %. In terms of artificial intelligence, the model integrates systematic and non-systematic clause representations while dynamically controlling negative literal distribution, contributing to a more adaptive logic rule. In engineering applications, the proposed method enhances practical performance in tasks requiring flexible, dynamic, and robust solution strategies, particularly in complex data analysis.}, archive = {J_EAAI}, author = {Yunjie Chang and Mohd Shareduwan Mohd Kasihmuddin and Wan Nur Aqlili Ruzai and Yueling Guo and Ju Chen}, doi = {10.1016/j.engappai.2025.111760}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111760}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weighted C-type random 2 satisfiability in discrete hopfield neural network}, volume = {160}, year = {2025}, }
• (2025). Multi-variable false data injection attack detection and classification in solar photovoltaic grids with ensemble learning. EAAI, 160, 111755. (www)

  @article{ , abstract = {As the world moves toward more sustainable energy options to reduce carbon footprints, the demand for grid-connected solar photovoltaic(PV) power plants is getting a lot of attention. To enhance the efficiency of these power plants, there is a need for extensive data analysis in the backend. This integration of physical infrastructure with the network and Internet of Things(IoT) devices enhances the system’s efficiency but also increases the risk of cyber intrusion. One of the most serious attacks includes the False Data Injection Attack(FDIA), which can affect working and suboptimal operations and has the potential to cause financial losses. These attacks can target single and multiple sensor datasets at the same time. In this study, we used multiple machine learning(ML) models along with ensemble learning techniques to compare the efficacy of each model for detecting FDIA. We used the ML models Random Forest (RF), Multi-Layer Perceptrons(MLP), and Convolutional Neural Network-Long Short-Term Memory(CNN-LSTM) networks, along with their combination in pairwise and triple ensembles, to identify and categorize FDIA in PV systems. Using data from grid-connected PV sites with simulated attacks, pairwise ensemble models consistently outperformed standalone techniques across falsification scenarios. Notably, RF+MLP and CNN-LSTM+RF ensembles achieved F1-scores of 0.95–0.97, precision of 0.92–0.99, and recall of 0.94–1.0 for uncorrelated and correlated multi-value injections. CNN-LSTM+RF ensembles excelled in Class T(Time Targeted Attack) with F1-scores of 0.69–0.80. While standalone models performed similarly for single-value injections, ensemble models combining RF, MLP, and CNN-LSTM demonstrated a 1% improvement in overall accuracy.}, archive = {J_EAAI}, author = {Shahid Tufail and Hasan Iqbal and Mohd Tariq and Arif Sarwat}, doi = {10.1016/j.engappai.2025.111755}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111755}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-variable false data injection attack detection and classification in solar photovoltaic grids with ensemble learning}, volume = {160}, year = {2025}, }
• (2025). Computationally efficient multi-objective optimization of an interior permanent magnet synchronous machine using neural networks. EAAI, 160, 111753. (www)

  @article{ , abstract = {Improving the power density of an interior permanent magnet synchronous machine requires a complex and comprehensive approach that includes electromagnetic and thermal aspects. To achieve that, a multi-objective optimization of the machine’s geometry was performed according to selected key performance indicators by using numerical and analytical models. The primary objective of this research was to create a computationally efficient and accurate alternative to a direct finite element method-based optimization. By integrating artificial neural networks as meta-models, we aimed to demonstrate their performance in comparison to existing State-of-the-Art approaches. The artificial neural network approach achieved a nearly 20-fold reduction compared with the finite element method-based approach in computation time while maintaining accuracy, demonstrating its effectiveness as a computationally efficient alternative. The obtained artificial neural network can also be reused for different optimization scenarios and for iterative fine-tuning, further reducing the computation time. To highlight the advantages and limitations of the proposed approach, a multi-objective optimization scenario was performed, which increased the power-to-mass ratio by 16.5%.}, archive = {J_EAAI}, author = {Mitja Garmut and Simon Steentjes and Martin Petrun}, doi = {10.1016/j.engappai.2025.111753}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111753}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Computationally efficient multi-objective optimization of an interior permanent magnet synchronous machine using neural networks}, volume = {160}, year = {2025}, }
• (2025). An expert features enhanced temporal and contextual contrasting learning model for detecting wind turbine blade icing. EAAI, 160, 111745. (www)

  @article{ , abstract = {With global carbon neutrality goals, wind power has rapidly developed, but blade icing remains a major challenge. AI(artificial intelligence) methods show great promise for detecting icing on wind turbine blades. However, early icing data overlap, difficulty obtaining continuous labeled data, and small variations between samples due to short sampling intervals complicate the task. This study proposes an expert feature-enhanced temporal and contextual contrastive learning model for detecting blade icing. This approach efficiently extracts data features and combines self-supervised contrastive learning, maximizing data utilization without requiring extensive labeled data. To validate the effectiveness of this method, extensive experiments were conducted on two public datasets. The results achieved the best performance across multiple metrics, with F1-Score and AUC exceeding 98%, significantly enhancing wind power generation efficiency.}, archive = {J_EAAI}, author = {Jiamei Zhou and Feng Xiao and Xiaoying Zhang and Xu Cheng and Jianhua Zhang}, doi = {10.1016/j.engappai.2025.111745}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111745}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An expert features enhanced temporal and contextual contrasting learning model for detecting wind turbine blade icing}, volume = {160}, year = {2025}, }
• (2025). PKDF-net: Anticancer peptide prediction via a prior-knowledge-aware dual-path feature-entangled network. EAAI, 160, 111743. (www)

  @article{ , abstract = {Anticancer peptides (ACPs) have emerged as promising therapeutic agents due to their ability to selectively target cancer cells. In recent years, various computational methods have been developed to identify ACPs. However, these approaches often struggle to effectively handle the complexity and variability inherent in peptide sequences. To address these challenges, we propose PKDF-Net, a prior-knowledge-aware dual-path feature-entangled network, designed to provide accurate ACP prediction. PKDF-Net consists of three key modules: the prior-knowledge-aware residue interaction representation (PKRI), the Mamba-enhanced amino acid residue representation (MER), and the dual-path feature-entangled prediction (DPFE). The PKRI module integrates prior biological knowledge, the MER module enhances sparse word embeddings within the sequence state space, and the DPFE module entangles dual-path features for precise prediction. Extensive experiments on benchmark datasets demonstrate that PKDF-Net outperforms state-of-the-art models, offering superior performance in ACP prediction. This study presents a promising tool to accelerate ACP development and provides valuable insights for the broader field of biological sequence analysis.}, archive = {J_EAAI}, author = {Qiangguo Jin and Ankang Wu and Leyi Wei and Hui Cui and Ping Xuan and Xikang Feng and Ran Su}, doi = {10.1016/j.engappai.2025.111743}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111743}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {PKDF-net: Anticancer peptide prediction via a prior-knowledge-aware dual-path feature-entangled network}, volume = {160}, year = {2025}, }
• (2025). Automated machine learning for text classification: Creating ensembles of classifiers through evolutionary hyper-heuristics. EAAI, 160, 111738. (www)

  @article{ , abstract = {This work presents a new approach in automated machine learning using hyper-heuristics to optimize classifier selection in text classification problems. Instead of choosing a single classifier, the approach generates a tailored classifier ensemble, enhancing overall performance. Hyper-heuristics are represented as sets of blocks, where a block comprises a 16-dimensional meta-feature vector and is labeled with a classifier. When presented with a dataset, they select the k-nearest blocks to the dataset in the 16-dimensional space, using their classifiers to build the corresponding ensemble. This space is defined by meta-features that measure the data distribution at the document, category, and dataset levels. The hyper-heuristics are learned through an evolutionary process, using a genetic algorithm, which allows the modification of their components and leads them to better solutions by applying crossover and mutation operators during a certain number of generations. During each generation, hyper-heuristics are evaluated with a training group of datasets to calculate their average classification performance. In the last generation, the best hyper-heuristic is returned and evaluated using a test group of unseen datasets to discover its generalization capabilities. The results are compared against two state-of-the-art automated machine learning systems and a previous hyper-heuristics-based approach. The best hyper-heuristic obtained presents an average classification performance superior to those of the other methods. This approach advances automated machine learning in text classification and significantly reduces the complexity of current systems, making them more accessible and efficient for non-expert users.}, archive = {J_EAAI}, author = {Jonathan Estrella-Ramírez and Carlos Hugo Garcia-Capulin and Oscar Almanza-Conejo and Juan Carlos Gómez Carranza}, doi = {10.1016/j.engappai.2025.111738}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111738}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated machine learning for text classification: Creating ensembles of classifiers through evolutionary hyper-heuristics}, volume = {160}, year = {2025}, }
• (2025). Optimized methods for basic probability assignments in evidence theory: Applications to fault diagnosis. EAAI, 160, 111737. (www)

  @article{ , abstract = {Evidence theory provides a robust framework for addressing classification problems under high uncertainty, with the generation of basic probability assignments (BPA) being a critical and challenging task. This study introduces novel methods for BPA generation, including a modified non-parametric (MNP) method, a sample-based direct (SD) method, and the modified sample-based direct (MSD) method. The MNP method introduces three strategies - plausibility function generation, integral optimization, and attribute discounting - to enhance BPA generation within the original framework. The SD method eliminates the necessity of estimating attribute probability density functions by directly constructing BPAs from individual samples and integrating them through Dempster's combination rule, thereby enabling asynchronous processing of sample data. Additionally, the MSD method incorporates sample discounting and attribute discounting for further optimization. Experimental evaluations conducted across multiple datasets from the University of California, Irvine Machine Learning Repository (UCI datasets) and a motor rotor fault diagnosis case study demonstrate that the proposed methodologies significantly enhance BPA generation, thereby improving classification accuracy. Furthermore, dynamic sample update experiments reveal that the SD method effectively facilitates sequential processing frameworks, enabling adaptive data adjustments with reduced computational and storage costs. In addition, we performed computational time analysis and parameter sensitivity analysis, providing a comprehensive guideline for method selection. These findings highlight the potential of artificial intelligence to address the needs of various intelligent-driven scenarios, such as real-time systems and resource-constrained environments, predictive maintenance in smart manufacturing, and condition monitoring of industrial machinery.}, archive = {J_EAAI}, author = {Xin Guan and Haotian Yu and Xiao Yi}, doi = {10.1016/j.engappai.2025.111737}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111737}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimized methods for basic probability assignments in evidence theory: Applications to fault diagnosis}, volume = {160}, year = {2025}, }
• (2025). Cardiometabolic biomarker prediction based on retinal fundus image. EAAI, 160, 111734. (www)

  @article{ , abstract = {Diagnosing common noncommunicable diseases, such as cardiovascular disease and diabetes, typically relies on blood sample analysis for biomarker measurement. This process is invasive, time-consuming, and relatively expensive. To address these limitations, deep learning methods were leveraged to estimate common cardiometabolic biomarkers using retinal fundus (RF) images. The study utilized 15,802 RF images from 5,653 participants in the Qatar Biobank (QBB), leading to the development of 19 deep-learning models to estimate biomarkers across seven categories: demographics and body composition, blood pressure, lipid profile, blood profile, hormones, kidney function, and metabolites. The proposed model outperformed existing models for the QBB-specific cohort across all biomarkers, achieving higher R-squared ( R 2 ) values, lower mean absolute error (MAE), and higher area under the curve (AUC). The proposed model achieved excellent performance in demographic predictions with age (MAE: 2.56, R 2 : 0.93) and gender (Accuracy: 96%, AUC: 0.94). For cardiovascular markers, it showed moderate predictability with systolic blood pressure (MAE: 8.02, R 2 : 0.49) and diastolic blood pressure (MAE: 6.06, R 2 : 0.45). For metabolic markers, the model demonstrated varying performance, with hemoglobin showing strong prediction (MAE: 0.79, R 2 : 0.60) while lipid markers showed moderate performance (total cholesterol MAE: 0.63, R 2 : 0.29). For creatinine, a kidney function marker, we achieved the best results with MAE: 9.00, R 2 : 0.33. Stratified analyses revealed systematic performance variations across gender, age, and disease-specific subgroups, with better predictions in males, young-agers, and non-diabetic participants. External validation of the CAD group confirms the effect of age, gender, and disease on prediction results, suggesting the need for personalized background in consideration for developing AI models. This study presents a promising approach for non-invasive biomarker estimation using retinal images, potentially revolutionizing early intervention and treatment planning in healthcare.}, archive = {J_EAAI}, author = {Syed Abdullah Basit and Hamada R.H. Al-Absi and Saleh Musleh and Tanvir Alam}, doi = {10.1016/j.engappai.2025.111734}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111734}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cardiometabolic biomarker prediction based on retinal fundus image}, volume = {160}, year = {2025}, }
• (2025). A self-adaptive energy-based learning rate for stochastic gradient descent via vector auxiliary variable method. EAAI, 160, 111731. (www)

  @article{ , abstract = {Optimizing the learning rate remains a critical challenge in machine learning, essential for achieving model stability and efficient convergence. This paper introduces the Vector Auxiliary Variable (VAV) algorithm, an energy-based self-adaptive learning rate optimization method for stochastic gradient descent(SGD). Unlike traditional methods with fixed learning rates, the VAV method dynamically adjusts learning rates using an auxiliary variable linked directly to the training loss, ensuring significantly larger stable learning rates, rapid convergence, and substantial reductions in training time. Empirical results from regression (Physics-Informed Neural Networks for Burgers’ equation), image classification (on CIFAR-10, CIFAR-100 datasets), and natural language processing (SST-2 sentiment classification) tasks demonstrate improved stability, faster convergence rates (e.g., achieving 88% accuracy on CIFAR-10 in 30 epochs vs. 80 epochs for SGD), and competitive accuracy performance. Theoretical analysis confirms the unconditional energy dissipation property and convergence guarantees under reasonable assumptions. The VAV method offers practical advantages, particularly beneficial for large-scale real-time machine learning applications in scientific computing and deep learning.}, archive = {J_EAAI}, author = {Jiahao Zhang and Christian Moya and Guang Lin}, doi = {10.1016/j.engappai.2025.111731}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111731}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A self-adaptive energy-based learning rate for stochastic gradient descent via vector auxiliary variable method}, volume = {160}, year = {2025}, }
• (2025). Convolutional neural networks for accurate and robust noninvasive pressure measurements in sealed systems. EAAI, 160, 111729. (www)

  @article{ , abstract = {Accurate noninvasive pressure measurement in sealed systems is essential for many applications in process control, chemical transport, and chemical safety assessment. This paper introduces a pioneering approach to enhance pressure estimation using Acoustic Resonance Spectroscopy (ARS), a promising technique that analyzes the system's exterior vibrations to infer internal pressure. Despite its potential, extracting precise measurements from noisy ARS data remains challenging. We present a novel method employing a convolutional neural network (CNN) to improve the accuracy and generalizability of the pressure prediction. The CNN performance is compared against benchmark ARS processing methods such as k-nearest neighbors (kNN) and a peak tracking method. The models were trained and tested using labeled data from laboratory-controlled tests on vacuum vessels subjected to varying internal pressures. Evaluation under various signal processing and field measurement scenarios was conducted to assess the accuracy and generalizability of the proposed models. The CNN outperformed other models by significant margins, achieving a mean absolute error (MAE) of approximately 15 Torr, compared to approximately 30 Torr for kNN and 60 Torr for peak tracking. Furthermore, we tested the generalizability of the models by introducing synthetic data augmentations such as spectrum shift and dropout, which approximate real-world measurement errors in ARS measurements. We found that the CNN maintained high accuracy under shifting and dropout data scenarios, showcasing its robustness, while the other models showed larger increases in error. This suggests CNN as a strong candidate for noninvasive pressure measurements, as well as a wide range of other applications detectible via spectral measurements, such as corrosion detection, structural integrity monitoring, or reaction tracking in closed vessels, offering high accuracy and resilience to environmental fluctuations.}, archive = {J_EAAI}, author = {D. Pereira and M. Prisbrey and E.S. Davis and P. Vakhlamov and A. Saini and C. Chavez and C. Pantea and J. Greenhall}, doi = {10.1016/j.engappai.2025.111729}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111729}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Convolutional neural networks for accurate and robust noninvasive pressure measurements in sealed systems}, volume = {160}, year = {2025}, }
• (2025). Bogie key components fault diagnosis utilizing multi-sensor tensor graphs and dual-attribute feature selection. EAAI, 160, 111727. (www)

  @article{ , abstract = {Fault diagnosis of the bogie is pivotal for the stable and efficient operation of the rail transit system. Although multi-sensor systems can reflect equipment states from multiple dimensions, handling high-dimensional sensor data with complex structural correlations remains challenging. This study proposes a novel intelligent fault diagnosis framework integrating tensor graph modeling and dual-attribute feature selection. Specifically, the contribution of artificial intelligence methodology includes developing a tensor graph structure to preserve spatial–temporal relationships, designing a dual-attribute selection method that simultaneously optimizes feature sensitivity and redundancy, and proposing a tensor graph embedding algorithm. In terms of engineering application, the framework is validated using traction motor bearing datasets, gearbox datasets, and wheelset bearing datasets. The proposed algorithm achieves dimensionality reduction with 0.982 clustering accuracy and 0.874 Silhouette score, outperforming conventional methods by 25.6% to 34.9%. In application results, the gear fault diagnosis accuracy reaches 98.5%, and the wheelset bearing recognition accuracy attains 99.5%. This study demonstrates a feasible path for efficiently utilizing multi-sensor data in industrial intelligent monitoring systems.}, archive = {J_EAAI}, author = {Zexian Wei and Deqiang He and Zhenzhen Jin and Haimeng Sun and Jinxin Wu and Sheng Shan and Jian Miao and Cai Yi}, doi = {10.1016/j.engappai.2025.111727}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111727}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bogie key components fault diagnosis utilizing multi-sensor tensor graphs and dual-attribute feature selection}, volume = {160}, year = {2025}, }
• (2025). Fuzzy granular computing for evaluating average uncertainty in machine learning models. EAAI, 160, 111723. (www)

  @article{ , abstract = {Realistic evaluation of uncertainty is crucial for informed decision-making in machine learning (ML) models. Our study introduces a novel approach to quantify uncertainty in model predictions using fuzzy numbers. Fuzzy numbers capture the inherent vagueness and imprecision in real-world data by providing a flexible framework for expressing uncertainties. In contrast to the conventional approach of constructing prediction intervals by assuming Gaussian probability distributions, which results in symmetric confidence intervals, our method can develop prediction intervals that are asymmetrically centered around the numeric prediction results. The approach involves forming triangular fuzzy intervals around ML predictions, where interval shape parameters are optimized using Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) to maximize coverage while maintaining specificity. To the best of our knowledge, this is the first model-agnostic approach using fuzzy numbers to generate asymmetric prediction intervals. Validation on benchmark datasets demonstrates the method’s effectiveness and suggests that asymmetric intervals better capture predictive uncertainty. Results across three public datasets demonstrate that our method produces competitive or superior uncertainty estimates compared to standard techniques, particularly under data degradation scenarios such as added noise or reduced output resolution (‘coarsening’).” A case study on industrial construction project duration estimation illustrates the method’s applicability. This case study highlights the practical value of our method in construction project duration forecasting, where distinguishing between over- and under-estimation is essential for budgeting, scheduling, and risk management.}, archive = {J_EAAI}, author = {Naimeh Sadeghi and Nima Gerami-Seresht and Witold Pedrycz and Aminah Robinson Fayek}, doi = {10.1016/j.engappai.2025.111723}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111723}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzy granular computing for evaluating average uncertainty in machine learning models}, volume = {160}, year = {2025}, }
• (2025). Weighted nonlinear information extension based time series Kolmogorov–Arnold network for industrial application with soft sensing. EAAI, 160, 111719. (www)

  @article{ , abstract = {Due to the complex characteristics of industrial process data, such as nonlinearity and variable coupling, accurately extracting important features for predicting key process variables using conventional soft sensing models remains a significant challenge. To address these issues, we propose a Mutual Information-guided Nonlinear Extension Attention-LSTM Kolmogorov–Arnold Network (NELKAN). Unlike conventional nonlinear soft sensor models that merely reconstruct data, NELKAN incorporates mutual information analysis to evaluate input–output correlations and assigns greater weights to highly relevant variables. These weighted variables are then expanded into a higher-dimensional feature space through a sparse autoencoder-based nonlinear transformation, which effectively mitigates variable coupling and captures complex inter-variable relationships. To further address the temporal dependencies in process data, an Attention-enhanced Long Short-Term Memory (LSTM) network is integrated with the Kolmogorov–Arnold Network (KAN) to enable accurate modeling of dynamic, nonlinear behaviors through enhanced sequential feature extraction. Experimental validation on the purified terephthalic acid (PTA) solvent system and gas turbine-based power plant (GP) case are conducted to demonstrate that the proposed NELKAN significantly improves the prediction accuracy of soft sensors.}, archive = {J_EAAI}, author = {Guo-Yu Liu and Qun-Xiong Zhu and Ning Zhang and Yan-Lin He and Ming-Qing Zhang and Yuan Xu}, doi = {10.1016/j.engappai.2025.111719}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111719}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weighted nonlinear information extension based time series Kolmogorov–Arnold network for industrial application with soft sensing}, volume = {160}, year = {2025}, }
• (2025). A general graph learning based framework for job offer recommender system. EAAI, 160, 111713. (www)

  @article{ , abstract = {In the digital age, job markets have undergone a transformation, becoming supply and demand for jobs through e-recruitment platforms. These e-recruitment platforms generally contain recommender systems to ensure easy filtering of large quantities of job offers and/or jobseeker profiles. Recommender systems play a major role in the fight against unemployment due to their strong presence in job recommendation platforms, by reducing the time taken to search for jobs and profiles. The use of graphs in recommender systems offers a more natural and intuitive technique for representing the interactions between jobseekers and job offers and thus providing good recommendations. In order to do this, these recommender systems rely on different types of information, such as the characteristics of the description of the offers/applicants, the history of interactions between profiles and offers, location, time, profile level and the number of experiences of the profiles. Existing studies that have tackled this subject generally use only one or two of these pieces of information, which limits their performance. They therefore have a common procedure which consists of constructing the graph from the job offers or jobseekers and computing recommendations on the graph, but with the limitation that the graph construction techniques are very specific to the dataset and therefore difficult to reproduce on other data. In additional, these graphs are mainly exploited using random walks algorithms and very few have explored graph learning technologies. In this paper, we propose a general framework based on graph learning for job recommender systems. This framework can be used to construct graphs (job offer-oriented graph, job seeker-oriented graph, bipartite graph of job offers and job seekers) from characteristics based on the description of the offer or profile, on interactions between job offers and job seekers, on temporal dynamics, on location, on the level of job seekers and on the number of experiences of job seekers; with the possibility of varying the weight of each characteristic to adapt to different data sets. It can then be used to learn graphs for embedding the various graphs constructed and finally to compute the recommendations based on graph learning or graph embeddings. We experimented our framework on Nigham and LinkedIn datasets, using Node2Vec, Graph Sample and Aggregation (GraphSage) and Graph Convolutional Networks (GCN) as graph embedding techniques, Decision Trees, eXtreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Bidirectional Encoder Representations from Transformers for Recommendation (BERT4Rec) and GCN as top-N recommendation algorithm and Accuracy, Precision, F1-score and Area Under the Curve (AUC) as evaluation metrics. The results obtained show that compared with standard graph-based recommendation techniques (PageRank and K-Nearest Neighbors) we have improvements of 7%, 5%, 11% and 11% respectively for accuracy, precision, F1-score and AUC, and compared with previous work an increase of up to 54% for F1-score.}, archive = {J_EAAI}, author = {Ghislain Wabo Tatchum and Armel Jacques Nzekon Nzeko’o and Thomas Messi Nguele and Xaveria Youh Djam}, doi = {10.1016/j.engappai.2025.111713}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111713}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A general graph learning based framework for job offer recommender system}, volume = {160}, year = {2025}, }
• (2025). A multi-topology quantum convolutional neural network with qubit-measurement attention for image classification. EAAI, 160, 111705. (www)

  @article{ , abstract = {With the increasing scale of data and complexity of problems, some researchers have explored the integration of parameterized quantum circuits (PQCs) within convolutional neural networks (CNNs) as a means to enhance algorithmic performance. However, in most current quantum convolutional neural networks (QCNN) models, a single topological structure for quantum kernels (qkernels) is adopted and only one qubit of qkernels is measured, both of which may limit the model’s performance. To solve these problems, a novel multi-topology quantum convolutional neural networks with qubit-measurement attention for image classification is proposed. In order to enhance the capability of feature extraction, a multi-topology PQCs strategy is proposed, i.e., we adopt the different topology PQCs to construct quantum convolutional layers. In addition, a qubit-measurement attention mechanism is designed to mitigate the significant loss of entanglement information during the measurement phase. Specifically, each qubit in the qkernel is measured to generate a local feature map, and the weight of each local feature map is then calculated, resulting in the final feature map. Nine image classification experiments conducted on CIFAR-10 demonstrate that our model outperforms the state-of-the-art QCNN model, achieving an improvement of 11.7% on ten categories classification. Our model not only introduces a new approach for constructing QCNNs but also provides valuable reference for designing attention mechanisms tailored to quantum computing.}, archive = {J_EAAI}, author = {QingShan Wu and WenJie Liu and XinRan Li and Zhaofeng Su and Jian Lei}, doi = {10.1016/j.engappai.2025.111705}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111705}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-topology quantum convolutional neural network with qubit-measurement attention for image classification}, volume = {160}, year = {2025}, }
• (2025). Ship anomalous behavior detection based on interval prediction of multiple vessel trajectories. EAAI, 160, 111693. (www)

  @article{ , abstract = {In response to the increasing complexity of maritime traffic situations, effective maritime surveillance is essential to ensure the safety of maritime activities. This paper leverages Automatic Identification Systems (AIS) data in conjunction with deep learning techniques and statistical methods to achieve adaptive matching and prediction of multiple vessel trajectories, as well as the identification of anomalous vessel behaviors. Initially, an improved Hausdorff distance is employed to accommodate preprocessed trajectory data, and an enhanced Density-Based Spatial Clustering of Applications with Noise(DBSCAN) algorithm is utilized for cluster analysis, facilitating adaptive multi-vessel trajectory matching. Subsequently, considering the inter-vessel influence zones, a multi-vessel behavior feature prediction is realized based on the completed adaptive vessel trajectory matching. Furthermore, distribution characteristics of trajectory prediction errors are analyzed, and an adapted kernel density estimation method is applied for confidence testing, leading to the proposal of the Unknown Behavioral Expansion- Adaptive Kernel Density Estimate(UBE-AKDE) model for identifying anomalous vessel behaviors. Finally, the algorithms presented are validated using trajectory data from all vessels within and near of the Tianjin Port over a period of five months.}, archive = {J_EAAI}, author = {Wentao Liu and Chen Chen and Yaowu Peng and Maohan Liang}, doi = {10.1016/j.engappai.2025.111693}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111693}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ship anomalous behavior detection based on interval prediction of multiple vessel trajectories}, volume = {160}, year = {2025}, }
• (2025). Mamba adaptive anomaly transformer with association discrepancy for time series. EAAI, 160, 111685. (www)

  @article{ , abstract = {Anomaly detection in time series poses a critical challenge in industrial monitoring, environmental sensing, and infrastructure reliability, where accurately distinguishing anomalies from complex temporal patterns remains an open problem. While existing methods, such as the Anomaly Transformer leveraging multi-layer association discrepancy between prior and series distributions and Dual Attention Contrastive Representation Learning architecture (DCdetector) employing dual-attention contrastive learning, have advanced the field, critical limitations persist. These include sensitivity to short-term context windows, computational inefficiency, and degraded performance under noisy and non-stationary real-world conditions. To address these challenges, we present MAAT (Mamba Adaptive Anomaly Transformer), an enhanced architecture that refines association discrepancy modeling and reconstruction quality for more robust anomaly detection. Our work introduces two key contributions to the existing Anomaly transformer architecture: Sparse Attention, which computes association discrepancy more efficiently by selectively focusing on the most relevant time steps. This reduces computational redundancy while effectively capturing long-range dependencies critical for discerning subtle anomalies. A Mamba-Selective State Space Model (Mamba-SSM) is also integrated into the reconstruction module. A skip connection bridges the original reconstruction and the Mamba-SSM output, while a Gated Attention mechanism adaptively fuses features from both pathways. This design balances fidelity and contextual enhancement dynamically, improving anomaly localization and overall detection performance. Extensive experiments on benchmark datasets demonstrate that MAAT significantly outperforms prior methods, achieving superior anomaly distinguishability and generalization across diverse time series applications. By addressing the limitations of existing approaches, MAAT sets a new standard for unsupervised time series anomaly detection in real-world scenarios. Code available at https://github.com/ilyesbenaissa/MAAT .}, archive = {J_EAAI}, author = {Abdellah Zakaria Sellam and Ilyes Benaissa and Abdelmalik Taleb-Ahmed and Luigi Patrono and Cosimo Distante}, doi = {10.1016/j.engappai.2025.111685}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111685}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mamba adaptive anomaly transformer with association discrepancy for time series}, volume = {160}, year = {2025}, }
• (2025). DeclareAligner: A leap towards efficient optimal alignments for declarative process model conformance checking. EAAI, 160, 111683. (www)

  @article{ , abstract = {Conformance checking is a crucial aspect of process mining, enabling organizations to identify deviations between actual process behavior and modeled expectations. At the heart of conformance checking lies the concept of optimal alignments, which provide a detailed, cost-minimized mapping of observed behavior to expected behavior. Optimal alignments facilitate the identification of root causes of non-conformity and guide corrective actions. This is a critical area where Artificial Intelligence (AI) plays a pivotal role in driving effective process improvement. However, computing optimal alignments poses significant computational challenges due to the vast search space inherent in declarative process models. Consequently, existing approaches often struggle with scalability and efficiency, limiting their applicability in real-world settings. This paper introduces DeclareAligner , a novel algorithm that uses the A* search algorithm, an established AI pathfinding technique, to tackle the problem from a fresh perspective leveraging the flexibility of declarative models. Key features of DeclareAligner include only performing actions that actively contribute to fixing constraint violations, utilizing a tailored heuristic to navigate towards optimal solutions, and employing early pruning to eliminate unproductive branches, while also streamlining the process through preprocessing and consolidating multiple fixes into unified actions. The proposed method is evaluated using 8054 synthetic and real-life alignment problems, demonstrating its ability to efficiently compute optimal alignments by significantly outperforming the current state of the art. By enabling process analysts to more effectively identify and understand conformance issues, DeclareAligner has the potential to drive meaningful process improvement and management.}, archive = {J_EAAI}, author = {Jacobo Casas-Ramos and Manuel Lama and Manuel Mucientes}, doi = {10.1016/j.engappai.2025.111683}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111683}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DeclareAligner: A leap towards efficient optimal alignments for declarative process model conformance checking}, volume = {160}, year = {2025}, }
• (2025). Underwater acoustic target detection with imbalanced dataset based on deep Q-network with dual termination criterion. EAAI, 160, 111661. (www)

  @article{ , abstract = {Underwater acoustic target detection often encounters the challenge of severe data imbalance. This work proposes an improved deep reinforcement learning algorithm for imbalanced classification, termed DQNiDTC(Deep Q-Network Based Algorithm for Imbalanced Classification with Dual Termination Condition). To fully utilize prior knowledge of acoustic signals, we propose an underwater acoustic feature acquisition strategy for imbalanced dataset with methods of feature extraction and feature selection, which helps to simplify the structure of Q-network from CNNs(Convolutional Neural Networks) to FCNNs(Fully Convolutional Networks) for faster convergence. To correctly classify samples of target echoes and radiated noises with serious imbalanced distribution, we propose a dual termination criterion for the simulation of designed environment, where misjudgments of both positive and negative samples will lead to the termination of that round of training with a reward mechanism for imbalanced classification, which enhances imbalanced classification by effectively utilizing rewards for both positive and negative samples, minimizing bias toward the majority class. Compared with existing method of DQNimb(Deep Q-Network Based Algorithm for Imbalanced Classification), experimental results show that F-measure of DQNiDTC is improved by 4.59%–10.23% with the imbalanced level of 1:100, 1:150, 1:200 and 1:250, thereby achieving efficient imbalanced binary classification for underwater acoustic target detection. Additionally, we provide the design and testing of a quasi-real-time model prediction based on the DQNiDTC algorithm, which can meet the quasi-real-time requirements of the model prediction end in imbalanced underwater acoustic target detection systems.}, archive = {J_EAAI}, author = {Zhifeng Liu and Chenrui Hu and Ben Xu and Min Huang and Zhongzhe Xiao}, doi = {10.1016/j.engappai.2025.111661}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111661}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Underwater acoustic target detection with imbalanced dataset based on deep Q-network with dual termination criterion}, volume = {160}, year = {2025}, }
• (2025). An efficient point network for light detection and ranging point cloud perception in large-scale scene. EAAI, 160, 111617. (www)

  @article{ , abstract = {With the ongoing advancement and wide application of Light Detection and Ranging (LiDAR) technology, the efficient deployment of deep learning models in various offline edge scenes, such as autonomous vehicles and drones, has become an important research hotspot in industry and academia. How to achieve lightweight and efficient network model (light and fast) on edge devices with limited resources is the key point. Model quantization methods, benefiting from lightweight parameters, have become one of the most promising approaches for model compression and acceleration. However, due to the unstructured characteristic of the point cloud, quantization may lead to aggregation-induced feature homogenization and structural scale distortion, resulting in performance bottlenecks for quantization models. Moreover, unstructured point cloud is difficult to adapt to existing image processing quantification methods and require a new network architecture. To this end, we constructed a specialized point cloud quantization network and designed an effective feature extraction architecture, proposing an Efficient Point Network (EPoNET). Specifically, to address the issues of feature homogenization and limited feature discriminability after quantization, we designed the dual refining kernel and the local feature reshaping module to refine and expand the features. To tackle the class imbalance within the dataset, we designed the weighted loss function to enhance the effectiveness of the network training and alleviate the interference caused by local optima during training. Finally, we evaluated the proposed EPoNET using the large-scale, real-world multispectral LiDAR point cloud dataset. This dataset covers about 25 square kilometers and includes 8.52 million points of different land features like roads, trees, and buildings. Experimental results demonstrate that EPoNET achieves an overall accuracy surpassing 90%, with an average accuracy of 91.31% across the testing areas. Compared with the lightweight benchmark method PointNet++, EPoNET brings 5.7x inference acceleration and 58.4% usage savings, achieving a balance between model inference accuracy, speed, and computational consumption.}, archive = {J_EAAI}, author = {Kai Xiao and Jialin Gui and Yuanxi Peng and Teng Li}, doi = {10.1016/j.engappai.2025.111617}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111617}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient point network for light detection and ranging point cloud perception in large-scale scene}, volume = {160}, year = {2025}, }
• (2025). Risk preference evaluation approach based on the choquet integral in q-rung orthopair fuzzy environment and its automotive performance assessment. EAAI, 160, 111611. (www)

  @article{ , abstract = {A q -rung orthopair fuzzy number ( q -ROFN) is a highly regarded and effective tool in intelligent computing and decision science, its unique representation ability and theoretical methods are widely used in many engineering fields, especially, they can handle some group decision-making problems with multi-attribute fuzzy information on a wider region. The main objective of this study is to overcome the shortcomings of existing ranking methods related to q -ROFNs and seek a more reasonable ranking criterion, and then establish a novel risk evaluation approach through the aggregation operator of discrete Choquet integrals in a q -ROFN environment. In this article, some related definitions for q -rung orthopair fuzzy numbers are introduced, a few defects of its traditional score functions are pointed out, and then a new risk preference score function with its ranking rule are given by introducing a q -level compression mapping and risk factor. Secondly, it is demonstrated that the novel risk score function satisfies monotonicity and the rationality of ranking criterion, two types of Choquet integral average (or geometric) aggregation operators are proposed in a q -rung orthopair fuzzy environment, and several calculation examples are provided. Finally, the novel aggregation operator and risk preference evaluation approach are established by the Choquet integral average operator and risk preference score function, and apply them to the assessment of automotive performances, and then their validity and superiority are verified by an example.}, archive = {J_EAAI}, author = {Deli Zhang and Guijun Wang}, doi = {10.1016/j.engappai.2025.111611}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111611}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Risk preference evaluation approach based on the choquet integral in q-rung orthopair fuzzy environment and its automotive performance assessment}, volume = {160}, year = {2025}, }
• (2025). Label distribution learning with correlation information. EAAI, 160, 111591. (www)

  @article{ , abstract = {Label distribution learning quantifies the label space for each instance and has broad applicability in various fields. However, most existing works primarily focus on label correlation, but they have a deficiency in capturing instance correlation. Meanwhile, traditional label distribution learning operates on individual labels sequentially, which restricts the potential application of common features. Therefore, in this paper, we propose a novel approach for label distribution learning with correlation information, i.e. , instance correlation and label correlation. Specifically, the instance correlation and label correlation are identified by an optimization function and Pearson correlation coefficient, respectively. Besides, we conduct ℓ 2 , 1 regularization to exploit common features. Comprehensive experiments conducted on twelve publicly datasets demonstrate the effectiveness of our proposed approach against other well-established label distribution learning algorithms.}, archive = {J_EAAI}, author = {Yilin Wu and Yaojin Lin and Wenzhong Guo and Weiping Ding}, doi = {10.1016/j.engappai.2025.111591}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111591}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Label distribution learning with correlation information}, volume = {160}, year = {2025}, }
• (2025). Intelligent long-term planning with interaction for crop management. EAAI, 160, 111506. (www)

  @article{ , abstract = {Utilizing artificial intelligence technology for crop management has garnered researchers’ attention in recent years. However, the extremely long cycles and unpredictable factors in practice have made it challenging to implement effective intelligent management policies. This work focuses on long-term planning and interaction issues in crop management. We construct a reinforcement learning framework and state prediction module, explicitly modeling the crop growth cycle in the process of learning single-step decisions. Through chain prediction, our method guides the learned policy, enabling the generation of long-term crop management plans at any stage. Additionally, these plans allow for user interaction to address unforeseen risks. Using irrigation and fertilization tasks as examples, we conducted numerous simulation experiments on the DSSAT (Decision Support System for Agrotechnology Transfer) platform. The results demonstrate that our method achieves satisfactory outcomes in both single-step decision-making and long-term planning. To our knowledge, this is also the first work to realize long-term planning in crop management through reinforcement learning.}, archive = {J_EAAI}, author = {Jingchen Li and Yusen Yang and Yisheng Miao and Wang Guo and Jingqiu Gu and Huarui Wu}, doi = {10.1016/j.engappai.2025.111506}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111506}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent long-term planning with interaction for crop management}, volume = {160}, year = {2025}, }
• (2025). Research on weld start point detection and localization using deep learning techniques. EAAI, 159, 111833. (www)

  @article{ , abstract = {In industrial automated welding, precise positioning of the weld start point is crucial for ensuring weld stability and quality. To address this challenge, this paper proposes an improved weld start point recognition algorithm based on the Real-Time Detection Transformer (RT-DETR), enabling the welding robot to adaptively recognize and initially position the weld start point. Images of the weld start point are captured using a vision system based on the D435i camera, and a dataset is constructed for deep learning training. The algorithm is based on the RT-DETR model, utilizing a Residual Network architecture with 18 layers (RT-DETR-R18), integrating the Context Guided Block (CGBlock) from the Context Guided Network (CGNet) with the BasicBlock of Residual Network 18 (ResNet18) to form a new architecture, Context Guided Residual Network 18 (CG-ResNet18), enhancing the model's global perception and accuracy. Additionally, a Large Separable Kernel Attention (LSKA) mechanism is incorporated into the backbone network to improve applicability and robustness. The hybrid encoder of RT-DETR is further optimized using Group Separable Convolution (GSConv) and Efficient Cross Stage Partial Network Block (VoV-GSCSP) modules, enhancing feature fusion and reducing computational load. Experimental results show that the improved model achieves a mean Average Precision (mAP) of 99.4 %, a 5 % increase from the original, with a 32 % reduction in parameters and a 31 % reduction in computational load. The detection error of the weld start point is below 0.4 mm, meeting the real-time detection and positioning requirements of the welding robot.}, archive = {J_EAAI}, author = {Mingda Huang and Linyuxuan Li and Yixing Meng and Guanhua Zhu and Hu Wu and Xianhai Yang}, doi = {10.1016/j.engappai.2025.111833}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111833}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on weld start point detection and localization using deep learning techniques}, volume = {159}, year = {2025}, }
• (2025). Unsupervised fault diagnosis and novel fault recognition for rotating machinery based on enhanced clustered autoencoder. EAAI, 159, 111822. (www)

  @article{ , abstract = {In recent years, deep learning has demonstrated significant potential in diagnosing faults in rotating machinery. However, its reliance on large volumes of labelled data and its limited capacity to identify new fault states continue to hinder practical applications. To overcome these challenges, this paper presents a novel unsupervised fault diagnosis method based on an enhanced clustering autoencoder. Unlike traditional unsupervised fault diagnosis methods that primarily depend on data dimensionality reduction or a single clustering technique, this approach incorporates a feature transformation mechanism and dynamic clustering optimisation during the dimensionality reduction process of the autoencoder. This effectively minimizes the internal distance between samples of the same category while expanding the discriminative space between different faults. Furthermore, by integrating deep clustering and anomaly detection techniques, the model establishes stable decision boundaries without the need for labelled data, significantly enhancing its ability to identify new faults that were not encountered during the training phase. Extensive experiments demonstrate that this method outperforms existing unsupervised diagnostic techniques on both public and gearbox datasets, particularly in the task of detecting new faults, thereby validating its efficiency and robustness in complex industrial environments.}, archive = {J_EAAI}, author = {Hangfeng Mo and Anzheng Huang and Shenxiao Chen and Zijia Wang and Xiangxin Kong and Zhinong Jiang and Zhiwei Mao}, doi = {10.1016/j.engappai.2025.111822}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111822}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised fault diagnosis and novel fault recognition for rotating machinery based on enhanced clustered autoencoder}, volume = {159}, year = {2025}, }
• (2025). A reliable enhanced learning algorithm for ship detection in inland water maritime surveillance system. EAAI, 159, 111820. (www)

  @article{ , abstract = {Accurate ship detection is crucial for intelligent maritime surveillance systems. Given the complicated navigation environment of the inland waterway and the low accuracy and efficiency of model detection, this work introduces an enhanced detection algorithm. Specifically, the learning and detection capabilities of the model are enhanced by optimizing the network structure, introducing an attention mechanism, and designing a non-maximal value suppression method based on bounding box loss. In addition, based on the physical principle of imaging in severe weather conditions, a data enhancement strategy that simulates low-quality images is proposed to enrich the dataset further and train the detection model using a learning approach, which can enable the algorithm to exhibit strong detection performance in severe weather conditions, such as rainy, hazy, and low-light. As demonstrated in the experiments, the enhanced learning algorithm achieves optimal detection performance in several state-of-the-art comparison algorithms, showcasing the effectiveness and superiority of the proposed algorithm.}, archive = {J_EAAI}, author = {Yuanzhou Zheng and Long Qian and Jingxin Cao and Haichao Huang and Wenbo Hou and Yeyu Liu}, doi = {10.1016/j.engappai.2025.111820}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111820}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A reliable enhanced learning algorithm for ship detection in inland water maritime surveillance system}, volume = {159}, year = {2025}, }
• (2025). A 6-dimensional pose estimation method combining sparse viewpoint classification initialization and optical flow-guided iterative refinement. EAAI, 159, 111814. (www)

  @article{ , abstract = {Electronic equipment is typically a complex and high-precision electromechanical system, where the routing and bundling of Radio Frequency (RF) cables are crucial to equipment performance. Traditional assembly methods require workers to assemble according to the assembly process card, which can easily lead to incorrect or missing assembly, poor assembly consistency, and low efficiency. Augmented Reality (AR) assembly guidance can effectively improve efficiency and reduce errors. 6-dimensional (6D) pose estimation is a key technology for AR assembly guidance. In the assembly process of complex electronic products, existing deep learning methods suffer from poor tracking and localization robustness and real-time performance due to factors such as arm occlusion, resulting in slow tracking recovery. This article proposes a two-stage real-time 6D pose estimation method from coarse to fine, which can estimate the pose of target objects in complex backgrounds at a speed of about 20 frames per second and quickly recover after tracking target loss. The real-time and effectiveness were verified through experiments on the red squirrel and electronic chassis.}, archive = {J_EAAI}, author = {Huan Yang and Yue Wang and Xinghang Yin and Yongxu Liu and Peng Wang}, doi = {10.1016/j.engappai.2025.111814}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111814}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A 6-dimensional pose estimation method combining sparse viewpoint classification initialization and optical flow-guided iterative refinement}, volume = {159}, year = {2025}, }
• (2025). A multi-layered neural control framework: Combining central pattern generators utilizing muscle synergy theory and incorporating proprioceptors. EAAI, 159, 111811. (www)

  @article{ , abstract = {Understanding human control strategy can promote the application of neuroscience in rehabilitation medicine, and drive the development of bionic robotics. The current research is to develop a bionic human control framework including both upper-level control signal from the central nervous system and lower feedback control loop. The upper-level control strategy is a combination of Central pattern generators (CPGs) signals based on the principle of muscle synergy, the weight of which is adaptively adjusted by reinforcement learning. The lower-level control involves feedback signals from primary proprioceptors. A neuromuscular arm model was established by integrating the bionic control strategy in an upper limb musculoskeletal dynamics model. The model feasibility was evaluated by elbow flexion experiments in the human sagittal plane and trajectory planning experiments in three-dimensional (3D) space. The indices such as joint rotation angles, muscle activation levels, and co-correlation index (CCI) were adopted. The simulated joint angles are well correlated with the experimental results of the two types, while the acceptable results for muscle activation levels and CCI values can only be obtained in the elbow flexion experiment. The proprioceptive feedback loop is also indicated to be necessary for accurate and stable human movement prediction. In summary, the current control strategy can well accommodate complex joint kinematics, and be further improved by addressing coordination issues of multiple muscle activations.}, archive = {J_EAAI}, author = {Fuhao Mo and Zhi Yang and Yiran Cui and Dongxi Ma and Haining Wang}, doi = {10.1016/j.engappai.2025.111811}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111811}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-layered neural control framework: Combining central pattern generators utilizing muscle synergy theory and incorporating proprioceptors}, volume = {159}, year = {2025}, }
• (2025). An optimized grey prediction model with second-order derivatives for wind power generation prediction in china. EAAI, 159, 111810. (www)

  @article{ , abstract = {The development of wind energy is vital to China's sustainable development, given its economic and environmental significance. However, wind power generation is affected by many uncertain factors, which poses a challenge to its accurate prediction. To this end, an optimized second-order-derivative grey prediction model is proposed for accurate prediction of China's wind power generation. Firstly, the grey buffer operator is employed to preprocess historical data, reducing the interference of impact disturbance factors and improving the reliability of prediction results. Secondly, a new three-parameter whitening grey prediction model with second-order derivatives is constructed by extending the model structure. Thirdly, a combinatorial optimization approach, utilizing Particle Swarm Optimization, is applied to fine-tune the parameters of the new model. Lastly, the proposed model is applied to predict China's wind power generation. The result shows that China's wind power generation will reach 2425.44 billion kilowatt-hour in 2030, which is 2.6 times higher than that of 2024. Based on the prediction data, several suggestions have been proposed to accelerate the development of China's wind power industry. The research findings of this paper provide an important predictive modeling method for the wind power generation with uncertain characteristics and also have positive significance for the improvement of the grey prediction model methodology.}, archive = {J_EAAI}, author = {Cuiwei Mao and Chengxiang He and Bo Zeng and Yingjie Yang}, doi = {10.1016/j.engappai.2025.111810}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111810}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An optimized grey prediction model with second-order derivatives for wind power generation prediction in china}, volume = {159}, year = {2025}, }
• (2025). An integrated exergy efficiency and machine learning method for optimizing organic solid waste gasification process. EAAI, 159, 111805. (www)

  @article{ , abstract = {Organic solid waste (OSW) gasification is a critical pathway toward sustainable energy utilization. This study develops an integrated prediction model by combining exergy efficiency-based analytic hierarchy process-fuzzy comprehensive evaluation (AHP-FCE) with machine learning techniques. The model aims to select the optimal gasifier type and operational parameters based on OSW characteristics and processing capacities. Exergy efficiency derived from experimental data is used to construct AHP-FCE scores, which are then predicted using eight machine learning algorithms. Gradient boosting decision tree (GBDT) achieves the best performance. The prediction model is applied to three practical cases. For a project with an annual processing capacity of 2000 tons of refuse-derived fuel (RDF), the model consistently recommends the downdraft fixed-bed gasifier (DBG). In a corn straw gasification project processing 11,000 tons per year, the bubbling fluidized-bed gasifier (BBG) is identified as the optimal choice. For a bamboo chip gasification project with an annual capacity of 150,000 tons, the model suggests using the circulating fluidized-bed gasifier (CFBG) for reduction objectives and the dual fluidized-bed gasifier (DFBG) for hydrogen production goals. Additionally, the model shows significant potential. It can also be applied to optimize other complex systems that require balancing multiple influencing factors.}, archive = {J_EAAI}, author = {Wenni Chen and Xianan Xiang and Sha Liu and Jun Guo and Tao Li and Xuehua Zhou and Deyong Peng and Zhiya Deng and Bo Wang and Hong Wang and Chunhui He}, doi = {10.1016/j.engappai.2025.111805}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111805}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated exergy efficiency and machine learning method for optimizing organic solid waste gasification process}, volume = {159}, year = {2025}, }
• (2025). Vehicle trajectory prediction model based on graph attention kolmogorov-arnold networks and multiple attention. EAAI, 159, 111804. (www)

  @article{ , abstract = {As the core of autonomous driving technology, trajectory prediction methods can analyze the topological structure of high-precision maps in-depth and combine the motion characteristics and interaction patterns of target agents. This method can not only provide reliable decision support for the safe driving of autonomous agents but also effectively predict potential risks. The existing research paradigm employs fixed activation functions, resulting in a limited nonlinear expression ability of the model and inflexibility. In addition, most methods employ fixed historical observation windows for research while overlooking the continuous characteristics of the dynamic evolution of traffic scenes. Inspired by this, we propose a vehicle trajectory prediction method based on Graph Attention Kolmogorov-Arnold networks and Multiple Attention module (GAKMA). In GAKMA, Graph Attention Kolmogorov-Arnold Networks (GAKAN) with learnable activation functions are employed for network construction, allowing for a more flexible capture of nonlinear relationships in the data. In addition, GAKMA designs a spatio-temporal encoder to perform multi-source heterogeneous fusion of high-precision maps and agent data, extract spatio-temporal features, and design multiple attention modules to absorb the impact of different factors on the future trajectory of the target agent from the agent, time, and modal dimensions. The cross-temporal attention mechanism can extract the dependency between historical prediction embeddings and future prediction embeddings, expand the attention window from the historical observation range to the time range containing the prediction embedding, and model the continuous correlation in the dynamic evolution of trajectory data in traffic scenarios. We evaluated GAKMA on two public real agent trajectory datasets. Experimental results show that GAKMA's performance in both single-agent trajectory prediction and multi-agent collaborative prediction scenarios has reached a relatively advanced level.}, archive = {J_EAAI}, author = {Jie Yang and Jian-wei Liu}, doi = {10.1016/j.engappai.2025.111804}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111804}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Vehicle trajectory prediction model based on graph attention kolmogorov-arnold networks and multiple attention}, volume = {159}, year = {2025}, }
• (2025). A physically informed neural network approach for modeling wave transformation in vegetated waters. EAAI, 159, 111803. (www)

  @article{ , abstract = {Prediction of wave propagation in vegetated waters is crucial for the design and maintenance of coastal ecological protection systems. In this study, we propose a Physics-Informed Neural Network (PINN) model that incorporates physical constraints from the Boussinesq equations for modeling wave propagation processes in vegetated waters. The results demonstrate that the PINN model effectively captures the evolution of regular wave propagation in rigid, non-submerged vegetated waters. Compared to conventional numerical models, the PINN approach offers a more efficient preprocessing framework while maintaining comparable simulation accuracy with an average Coefficient of Determination (R 2 ) of 0.942, an average Root Mean Square Error (RMSE) of 1.84 × 10 −3 m and an average Mean Absolute Error (MAE) of 1.19 × 10 −3 m. Moreover, the parametric inference framework embedded within PINN enables precise determination of the optimal drag coefficient ( C d ) through systematic assimilation of experimental measurements. Additionally, the accuracy of both the simulation and the inferred C d improves as more external data are integrated into the model.}, archive = {J_EAAI}, author = {Xinyu Huang and Jun Tang and Yongming Shen and Yanlong Zhao and Shuai Hao}, doi = {10.1016/j.engappai.2025.111803}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111803}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A physically informed neural network approach for modeling wave transformation in vegetated waters}, volume = {159}, year = {2025}, }
• (2025). Correcting update anomalies in spatiotemporal knowledge graphs with uncertainties. EAAI, 159, 111798. (www)

  @article{ , abstract = {As a knowledge description framework certified by the W3C (World Wide Web Consortium), Resource Description Framework (RDF) has a strict yet flexible description syntax and is widely accepted as a carrier of knowledge graphs (KGs). Currently, a large amount of research work has been done on static knowledge graphs. However, dealing with dynamic knowledge graphs, especially spatiotemporal knowledge, remains an important research topic. Knowledge in the real world is not always deterministic. Consequently, some research on uncertain knowledge modeling and management has been proposed. This paper focuses on fuzzy spatiotemporal knowledge modeling within the context of the RDF model. We formally propose an extended RDF model to represent fuzzy spatiotemporal knowledge. This model can represent the continuous motion trajectory of fuzzy spatiotemporal entities and avoid repeated time intervals and location information in RDF. Specifically, we systematically identify the consistency constraints in the fuzzy spatiotemporal RDF model. Based on these constraints, we further propose a method for correcting timely inconsistencies when updating fuzzy RDF graph data. Experimental results demonstrate the effectiveness of the fuzzy spatiotemporal RDF model and the inconsistency correction method proposed in this paper.}, archive = {J_EAAI}, author = {Xiaowen Zhang and Li Yan and Beijing Zhou and Zongmin Ma}, doi = {10.1016/j.engappai.2025.111798}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111798}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Correcting update anomalies in spatiotemporal knowledge graphs with uncertainties}, volume = {159}, year = {2025}, }
• (2025). A gated recurrent unit considering spatial correlation for short-term traffic volume forecasting in highway tunnels. EAAI, 159, 111796. (www)

  @article{ , abstract = {Growing emphasis on low-carbon targets and energy conservation has significantly increased demand for efficient ventilation systems and energy optimization in highway tunnels, necessitating more accurate traffic forecasting. This study investigated spatiotemporal patterns of traffic flow within tunnels and at adjacent monitoring points, specifically considering unique tunnel ventilation requirements. Recognizing distinct traffic characteristics during different periods, we proposed enriching the dataset with comprehensive public holiday records to improve forecasting accuracy during these critical intervals. Through spatial correlation analysis, we established the quantitative relationship between inter-node distances and optimal sliding time steps that maximize spatial dependency, along with a methodological framework for estimating these temporal parameters. A comparative evaluation of prevalent deep learning architectures identified the Gated Recurrent Unit (GRU) as the most suitable baseline. Building on spatial correlation insights, we developed the Considering Spatial Correlation (CSC) method, which employs adaptive attention mechanisms to dynamically weight critical sliding time steps, enabling focused extraction of spatially influential features. By integrating CSC with GRU, we constructed the CSC-GRU model. Experimental comparisons with state-of-the-art baselines, including the Graph Neural Networks-Long Short-Term Memory (GNN-LSTM) model, demonstrate CSC-GRU model's superior performance, achieving a 9.9 % reduction in Mean Absolute Error (MAE) and a 13.5 % improvement in Symmetric Mean Absolute Percentage Error (SMAPE) over the best-performing baseline. Notably, the forecast curves exhibit enhanced responsiveness to abrupt traffic fluctuations during peak periods. This research provides a robust framework for optimizing traffic forecasting and energy management in tunnel ventilation systems, offering practical implications for sustainable infrastructure development.}, archive = {J_EAAI}, author = {Hongrui Zeng and Chen Dong and Rui Fu and Kaichun Su and Xiqiao Leng and Chun Guo}, doi = {10.1016/j.engappai.2025.111796}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111796}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A gated recurrent unit considering spatial correlation for short-term traffic volume forecasting in highway tunnels}, volume = {159}, year = {2025}, }
• (2025). An effective convolutional and transformer cooperation network for underwater acoustic target recognition. EAAI, 159, 111791. (www)

  @article{ , abstract = {Underwater acoustic target recognition (UATR) is a key technology in the field of underwater acoustic information processing. In recent years, models based on convolutional neural networks (CNN) have shown excellent performance in the domain of UATR. However, CNN have limitations in capturing the global information of underwater acoustic features. Due to its advantages in modeling global dependencies, the Transformer model is gradually gaining attention from researchers. In order to capture the time-frequency dependencies in acoustic spectrograms more effectively, this paper proposes a recognition model based on the Mel spectrogram that combines CNN with the Transformer, named the underwater acoustics CNN-Transformer cooperation network (UACTC). Compared to the Transformer alone, this model is more efficient in extracting local features. The CNN module uses a residual network based on the efficient channel attention (ECA) module for efficient deep feature extraction. Additionally, the ECA module is introduced into the Transformer block to enhance the channel feature extraction of the Transformer. Experiments prove that the ECA module effectively improves the performance of the recognition system. The effectiveness of the proposed model has been validated on two public datasets, achieving 98.05 % and 96.96 % on the ShipsEar and DeepShip datasets, respectively.}, archive = {J_EAAI}, author = {Anqi Jin and Shuang Yang and Menghui Lei and Xiangyang Zeng and Haitao Wang}, doi = {10.1016/j.engappai.2025.111791}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111791}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An effective convolutional and transformer cooperation network for underwater acoustic target recognition}, volume = {159}, year = {2025}, }
• (2025). Optimization and prioritization of electric vehicle charging locations for InterCity transport: A dual-phase approach. EAAI, 159, 111786. (www)

  @article{ , abstract = {Electric vehicles (EVs) are the key concepts for advancing sustainable transportation by reducing dependence on fossil fuels and increasing energy efficiency. So, determining optimal locations for electric vehicle charging stations (EVCSs) is critical. This study proposes a two-phase methodology that integrates both qualitative and quantitative data to optimize and prioritize the EVCS locations. In Phase I, an extended Set Covering Model (SCM) is applied to optimize potential alternatives. In Phase II, a prioritization framework is developed for cases where infrastructure cannot be implemented simultaneously. To address uncertainty and subjectivity in expert evaluations, one of critical methods on Artificial Intelligence (AI) named fuzzy logic is used. For this aim, Z-numbers, which is an extension of fuzzy sets is employed, capturing both data vagueness and judgment hesitation. Accordingly, Interpretive Structural Modeling (ISM), Fuzzy Cognitive Mapping (FCM), and Technique for Order Preference by Similarity (TOPSIS) methods are extended with Z-numbers to evaluate and rank the optimized alternatives. The methodology is applied to a real-world case along the İstanbul–Ankara route in Türkiye. Results indicate that the criterion “needs to meet electric vehicle charging station demand” is the most influential, and “Erkanoğlu Resting Facility” is identified as the most suitable location. Sensitivity analysis confirms the robustness of the results under varying conditions. Overall, the proposed approach offers a comprehensive and reliable decision-making framework for intercity EVCS planning by effectively integrating optimization, expert judgment, and uncertainty modeling.}, archive = {J_EAAI}, author = {Kübra Yazır and Ali Karasan and İhsan Kaya}, doi = {10.1016/j.engappai.2025.111786}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111786}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimization and prioritization of electric vehicle charging locations for InterCity transport: A dual-phase approach}, volume = {159}, year = {2025}, }
• (2025). Research on optimization strategy for steel strip temper rolling elongation based on model predictive control. EAAI, 159, 111785. (www)

  @article{ , abstract = {This study addresses the limitations of traditional Proportional-Integral-Derivative (PID) control in strip steel temper rolling elongation rate regulation, such as weak anti-interference capability and insufficient dynamic response, by proposing an intelligent elongation rate optimization strategy based on a hybrid Model Predictive Control (MPC) and Long Short-Term Memory (LSTM)-Transformer model. A data-driven closed-loop control framework is constructed by integrating the temporal modeling capability of deep learning with the dynamic optimization characteristics of MPC. First, an innovative LSTM-Transformer hybrid prediction model is applied, combining LSTM's local temporal feature extraction capability with Transformer's global dependency capturing mechanism, achieving high-precision prediction of real elongation rate (test set Mean Absolute Error <0.02, R 2 > 0.97). Second, a dynamic weighted sampling method is proposed to address the insufficient model generalization caused by imbalanced industrial data distribution by enhancing learning of key dynamic conditions such as weld seam transitions and rolling mill speed changes. On this basis, a Cross-Entropy Method (CEM)-based MPC controller is developed, integrating the prediction model with multi-objective optimization to dynamically adjust rolling force, tension, and other parameters while minimizing elongation rate fluctuations under equipment constraints. Finally, experiments on an offline simulation platform demonstrate that elongation rate fluctuations are substantially reduced, achieving precise control and robust disturbance rejection under complex conditions like acceleration/deceleration and weld seams. This study provides theoretical support and a technical paradigm for the intelligent transformation of the steel industry, demonstrating significant engineering application value.}, archive = {J_EAAI}, author = {Zhixuan Wang and Ji Zhang and Zhuo Wang and Hao Wang and Zhenhua Bai}, doi = {10.1016/j.engappai.2025.111785}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111785}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on optimization strategy for steel strip temper rolling elongation based on model predictive control}, volume = {159}, year = {2025}, }
• (2025). A strongly supervised hyperspectral unmixing framework for precise mineral composition and coal ash content estimation. EAAI, 159, 111784. (www)

  @article{ , abstract = {Accurate coal ash content detection is essential for advancing intelligent clean coal processing and holds significant practical value across mining, washing, combustion, and conversion technologies. This paper introduces a strongly supervised hyperspectral unmixing (SSHU) framework designed to estimate mineral composition proportions and ash content. We conducted systematic ablation experiments on concentrated coal and tailings coal datasets to evaluate the method's effectiveness and analyze the mechanisms of proportional prior information and reconstruction decoders. Results demonstrate that proportional prior information effectively constrains the proportional encoder, making estimated mineral and pure coal distributions closer to actual material distributions. The reconstruction decoder enhances the proportional encoder's feature extraction ability, guides model convergence, and improves both proportion and ash content estimation accuracy. Compared to existing hyperspectral unmixing methods, our approach incorporates pure substance spectral information during model training and combines proportional prior constraints. This provides a robust solution for complex mixture analysis and demonstrates significant potential in hyperspectral unmixing applications.}, archive = {J_EAAI}, author = {Yao Cui and Ziqi Lv and Ying Gao and Yuxin Wu and Xuan Zhao and Qingxuan Meng and Jun Dong and Zhiqiang Xu and Weidong Wang}, doi = {10.1016/j.engappai.2025.111784}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111784}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A strongly supervised hyperspectral unmixing framework for precise mineral composition and coal ash content estimation}, volume = {159}, year = {2025}, }
• (2025). A group decision making scheduling method for flexible job shop integrating blockchain and rule mining. EAAI, 159, 111783. (www)

  @article{ , abstract = {The future workshop will be an intelligent one based on the Cyber-Physical System paradigm, where each device functions as an agent. Each agent will possess independent data perception and reasoning capabilities, enabling it to operate autonomously and freely join or leave the agent network. Production scheduling will be the result of multi-agent collective decision-making. Traditional centralized scheduling methods are no longer applicable to such workshops, and existing distributed scheduling approaches remain incomplete. Specifically, current distributed scheduling methods still exhibit traces of centralized scheduling and have not fully realized decentralized scheduling in a strict sense. To address this deficiency, we propose a multi-objective flexible job shop distributed scheduling method. In this method, we improve the neighborhood search algorithm by generating a high-quality initial solution, selecting effective critical operation blocks, and using rule extraction to search for the optimal solution. To achieve complete decentralization of scheduling, blockchain is introduced and redesigned for distributed processing of data. Some numerical experiments, based on well-known benchmark instances, are carried out. The results verify the feasibility and competitiveness of the scheduling method. The solution of this problem has important academic significance and engineering value for intelligent factory design.}, archive = {J_EAAI}, author = {Yingli Li and Ying Zhao}, doi = {10.1016/j.engappai.2025.111783}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111783}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A group decision making scheduling method for flexible job shop integrating blockchain and rule mining}, volume = {159}, year = {2025}, }
• (2025). A stochastic multi-criteria acceptability analysis-based group consensus model considering overlapping decision-makers. EAAI, 159, 111782. (www)

  @article{ , abstract = {Social network group decision-making (SNGDM) provides support for obtaining consistent decision results based on social trust relationships between individuals. However, excessive internal conflict level within clustered subgroups may affect the consensus reaching process (CRP). To solve this problem, it may be necessary to assign some decision-makers (DMs) to two or more subgroups simultaneously during the clustering process. This paper investigates the CRP in the context of SNGDM problems with overlapping DMs. Firstly, taking into account the trust degree and evaluation conflict degree between DMs, we propose a definition of comprehensive evaluation conflict network in SNGDM. Then, we introduce a DM overlapping clustering method based on the Floyd-Warshall algorithm. The method takes into account the comprehensive evaluation conflict degrees among DMs and is capable of identifying overlapping DMs belonging to multiple clustered subgroups. Furthermore, we propose a two-stage opinion adjustment process that takes into account overlapping DMs. This process provides an opinion adjustment strategy for DMs when the consensus states across the multiple subgroups they belong to differ. Moreover, we design a consensus determination mechanism based on stochastic multi-criteria acceptability analysis method, which can reduce the impact of subjectively given criteria weights on the accuracy of SNGDM results. Finally, we use an elderly care service provider evaluation case to illustrate the effectiveness of our proposed method. Results indicate that the proposed method can effectively identify overlapping subgroups and overlapping DMs, and the proposed method can solve the SNGDM problems when DMs can not provide preferences for criteria weights.}, archive = {J_EAAI}, author = {Zhiwei Xu and Peide Liu and Peng Li and Cuiping Wei}, doi = {10.1016/j.engappai.2025.111782}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111782}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A stochastic multi-criteria acceptability analysis-based group consensus model considering overlapping decision-makers}, volume = {159}, year = {2025}, }
• (2025). A high-accuracy deep learning framework for digital twin model development of actual chemical processes. EAAI, 159, 111780. (www)

  @article{ , abstract = {The development of intelligent chemical industry urgently calls for high-accuracy digital twin modeling methods of actual chemical processes. However, although intelligent technologies such as deep learning are utilized, digital twin modeling is challenged by the complexity and nonlinearity of actual chemical processes, leading to unexpected deviations and poor generalization. To provide a valuable modeling methodology in this field, a novel deep learning framework based on chemical process mechanisms is developed. According to the properties of different neural networks, the framework is designed in a hierarchical way to learn the underlying process mechanisms and deal with the nonlinearity of actual chemical processes. Thus, a digital twin model of the actual chemical process can be developed, which is high-accuracy and good-generalization. Based on the sensor data from the distributed control system of an actual industrial distillation process, the proposed deep learning digital twin modeling framework is verified by thoroughgoing and careful analyses, including the modeling test, ablation experiment and residue distribution analysis of model results. Compared with wide-spread baseline models, the new framework decreases the predicting mean absolute error by 45.6 % and increases the coefficient of determination by 13.8 % on average. This intelligent framework reveals desirable application potentials that it can create a high-fidelity digital twin of actual chemical processes. Such digital representation acts as a basic model, which can support advanced manufacturing tasks such as real-time optimization by detailed digital inference. This work provides an intelligent digital twin modeling methodology in chemical engineering and facilitates the development of intelligent chemical industry.}, archive = {J_EAAI}, author = {Yue Li and Zhongmei Li and Jingzheng Ren and Wenli Du and Weifeng Shen}, doi = {10.1016/j.engappai.2025.111780}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111780}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A high-accuracy deep learning framework for digital twin model development of actual chemical processes}, volume = {159}, year = {2025}, }
• (2025). User feedback information analysis based on collaborative filtering and revised rough numbers: A study on product green design elements extraction. EAAI, 159, 111776. (www)

  @article{ , abstract = {In the era of the knowledge economy, user requirements (URs) have become one of the core driving forces that promote the development of enterprise product collaborative design. However, since user interests often drive URs and they are not professional designers, expecting them to translate requirements into design information is unrealistic. This paper presents an Artificial Intelligence (AI)-driven method to extract the product green design elements (PGDEs) from the user feedback preference information (UFPI). Firstly, sentiment analysis and Latent Dirichlet Allocation (LDA) topic modeling are implemented to determine the URs in online comments. Secondly, based on Large Language models (LLMs), specifically SparkDesk, the product green design requirements (PGDRs) that designers must consider during the product life cycle are generated. Third, the product green design elements (PGDEs) set composed of URs and PGDRs is constructed. Then, based on the collaborative filtering (CF) algorithm, the UFPI is obtained, and the essential elements of product green design are determined. Finally, based on the revised rough number theory, the decision and optimization of green design elements are completed. This study demonstrates the practical application of AI technologies in green design decision-making and provides theoretical and methodological support for the decision to PGDEs.}, archive = {J_EAAI}, author = {Yan Xuan and Lei Zhang}, doi = {10.1016/j.engappai.2025.111776}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111776}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {User feedback information analysis based on collaborative filtering and revised rough numbers: A study on product green design elements extraction}, volume = {159}, year = {2025}, }
• (2025). Performance evaluation of low-power and lightweight object detectors for real-time monitoring in resource-constrained drone systems. EAAI, 159, 111775. (www)

  @article{ , abstract = {The integration of drones in intelligent systems has gained significant traction, particularly in real-time aerial monitoring and object detection applications. However, deploying advanced artificial intelligence models on edge devices such as drones platforms remains challenging due to limited computing resources and power constraints. This study evaluates the suitability of several cutting-edge You Only Look Once (YOLO) variants, specifically versions 9-tiny, 10-nano, and 10-small, alongside version 7-tiny for real-time object detection on resource-limited platforms. Using a Jetson Xavier-equipped drone, the study focuses on achieving a balance between speed, accuracy, and efficient resource utilization in dynamic environments. The low-power consumption and minimal latency properties of the models were investigated considering tiny machine-learning (TinyML) principles, ensuring their suitability for intelligent edge computing systems. Key performance metrics, including inference time, average precision, graphics processing unit (GPU) usage, and power consumption, were analyzed in detail. YOLO version 10-nano, optimized for TinyML deployment, with an inference time of 10.34 ms, a mean average precision (mAP) of 0.657, and moderate GPU usage, achieved an optimal balance of detection speed, accuracy, and energy efficiency, making it highly suitable for drone-based autonomous systems requiring rapid decision-making. YOLO version 9-tiny, with an inference time of 13.55 ms and a mAP of 0.688, also demonstrated strong potential for real-time, time-sensitive applications, though with slightly higher GPU usage and power consumption. In contrast, YOLO version 10-small, while achieving the highest mAP of 0.77, showed elevated power requirements and GPU utilization, indicating its use in scenarios prioritizing accuracy over power efficiency.}, archive = {J_EAAI}, author = {Murat Bakirci}, doi = {10.1016/j.engappai.2025.111775}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111775}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Performance evaluation of low-power and lightweight object detectors for real-time monitoring in resource-constrained drone systems}, volume = {159}, year = {2025}, }
• (2025). A computer vision-based crop and weed segmentation using federated ensemble learning in diverse multi-crop fields. EAAI, 159, 111773. (www)

  @article{ , abstract = {An intelligent weed segmentation system is a suitable Artificial Intelligence-based application for modern agricultural activities to control weeds in crop fields. Building such a system that segments weed portions in multiple crop fields is a challenging task since the features of these weeds vary based on the nature of the crop field. It is even more difficult to build a scalable crop and weed segmentation system capable of incorporating new crop field images. To attain this, federated learning and ensemble learning techniques are incorporated in this research work. Three pre-trained models, DeepLabV3+, MobileNetV3, and Xception, serve as core models for building the underlying encoder-decoder models for the crop and weed segmentation system. Images of carrot, sugar beet, sunflower, potato, and maize crop fields are employed in this study. During federated learning, the local model of each client is trained using distinct crop field images, and later the learned features are aggregated into the global model on the server side using two different methods, namely; Federated Global Model with Weight Averaging (FGMWA) and Federated Global Model with Ensemble Weights (FGMEW). The global model built using the Federated Global Model with Ensemble Weights (FGMEW) approach with the DeepLabV3+ backbone outperformed others in segmenting crop and weed portions across all five crops, achieving overall Mean Intersection over Union (MIoU), Accuracy, and F1-scores of 79.1 %, 94.2 %, and 86.8 %, respectively. This approach of building the global model enables the addition of new crops without the need for complete retraining of the backbone models.}, archive = {J_EAAI}, author = {Sankeshwar Sivakumar and Abel Joshy Payyappilly and Akhil Rajan and Arumuga Arun R. and R.S. Rampriya}, doi = {10.1016/j.engappai.2025.111773}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111773}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A computer vision-based crop and weed segmentation using federated ensemble learning in diverse multi-crop fields}, volume = {159}, year = {2025}, }
• (2025). An efficient multimodal electroencephalography and functional near-infrared spectroscopy approach using stroop task to evaluate internet gaming disorder with machine learning methods: A comprehensive study. EAAI, 159, 111772. (www)

  @article{ , abstract = {Internet Gaming Disorder (IGD) has emerged globally in recent decades with increasing severity. Despite its growing risk to public health, the neurobiological mechanisms underlying IGD remain unclear. Thus, underpinning the diagnosis of IGD based on neurophysiological data using machine learning (ML) techniques is highly valuable. 102 male university students, including 51 IGD and 51 healthy controls aged 18–23, participated in the study. A classic Stroop task with congruent, incongruent, and neutral stimuli was presented during simultaneous electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS). Features extracted from EEG sub-bands included statistical, complexity, and frequency domain features, amplitude, and latency information from event related potential (ERP) components, as well as oxygenation changes from fNIRS. The features were selected based on statistical significance using a t -test; subsequent classification was performed over the well-known ML. EEG, ERP, and fNIRS features were compared between the groups and provided valuable insight into IGD, the best classification performance obtained with Support Vector Machines. The accuracy was 79.4 % for EEG and 87.25 % for a combination of EEG and fNIRS systems. The results indicate that combining EEG and fNIRS modalities provides better and more robust performance. Examining linear and nonlinear EEG features, ERP component features, and fNIRS hemodynamic activation offers comprehensive analysis and potential biomarkers. This study aimed to develop an ML based diagnosis of IGD in male university students using multimodal EEG and fNIRS data recorded during a Stroop task. This study investigated IGD for the first time using multimodal EEG-fNIRS responses elicited by the Stroop task.}, archive = {J_EAAI}, author = {Demet Yeşi̇lbaş and Turgay Batbat and Miray Altınkaynak and Elif Uğurgöl and Ayşegül Güven and Nazan Dolu and Esra Demi̇rci̇ and Meltem İzzetoğlu}, doi = {10.1016/j.engappai.2025.111772}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111772}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient multimodal electroencephalography and functional near-infrared spectroscopy approach using stroop task to evaluate internet gaming disorder with machine learning methods: A comprehensive study}, volume = {159}, year = {2025}, }
• (2025). Knowledge-enhanced multi-objective memetic algorithm for energy-efficient flexible job shop scheduling with limited multi-load automated guided vehicles. EAAI, 159, 111771. (www)

  @article{ , abstract = {In alignment with the national call for energy conservation and emission reduction, energy-efficient scheduling in manufacturing, especially intelligent workshops, has become a key research area. Automated guided vehicles (AGVs), as the core component of intelligent logistics systems, especially in applying multi-load AGVs, play a vital role in improving green manufacturing and optimizing logistics efficiency. While AGV transportation is considered in traditional energy-saving scheduling, most studies assume unlimited AGVs, and each can only load one job. This paper is the first to study the energy-efficient flexible job shop scheduling with limited multi-load AGVs (EFJSP-LMA), which integrates the sequencing of pickup and delivery tasks, and the allocation strategy of machines and AGVs. To address this problem effectively, the multi-objective mixed-integer programming (MMIP) model is developed to optimize the makespan and total energy consumption. To solve the MMIP model, a knowledge-enhanced multi-objective memetic algorithm (KMMA) is proposed. In the proposed KMMA, problem-specific heuristics are designed to generate a high-quality initial population with strong convergence and diversity. Subsequently, five knowledge-enhanced variable neighborhood structures are designed to enhance the quality and diversity of solutions. Additionally, an energy-saving strategy is incorporated to further optimize energy consumption. The effect of AGV quantity and load modes on the performance of the production system is studied and analyzed. Furthermore, experiment results of 60 test instances indicate that KMMA outperforms comparison algorithms, demonstrating its effectiveness in addressing the EFJSP-LMA. Finally, Real-world case studies further support our research, offering valuable insights for managing manufacturing environments.}, archive = {J_EAAI}, author = {Lianghua Fan and Qi Lei and Yuchuan Song and Yang Liu and Yunfan Yang}, doi = {10.1016/j.engappai.2025.111771}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111771}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Knowledge-enhanced multi-objective memetic algorithm for energy-efficient flexible job shop scheduling with limited multi-load automated guided vehicles}, volume = {159}, year = {2025}, }
• (2025). Self-information and prediction mask enhanced blind inpainting network for dunhuang murals. EAAI, 159, 111769. (www)

  @article{ , abstract = {Blind image inpainting methods based on deep learning have shown promising results in digital image inpainting of dunhuang mural images in recent years. However, current blind inpainting methods still suffer from color patches and structural confusion in the repair results caused by contamination of damaged features and sub-network interference. To address the above problems, a self-information and prediction mask enhanced blind inpainting network (SIME-BINet) for dunhuang mural images is proposed. SIME-BINet redesigns blind inpainting method of phased guidance paradigm into information enhance paradigm, which continuously optimizes enhanced information in dynamic form during training process and provides guidance for encoding process. Meanwhile, an information enhanced transformer block is designed to overcome the problem of damaged feature contamination by introducing enhanced information. Experiments show that SIME-BINet outperforms recent state-of-the-art blind inpainting methods on DhMurals1714 dataset and real damage mask. SIME-BINet offers a new paradigm for blind image inpainting based on deep learning and provides an innovative approach for inpainting of dunhuang mural images. The code, data, and pre-trained models will be made available at https://github.com/IPCSRG/SIME-BINet after the paper is published.}, archive = {J_EAAI}, author = {Jiahao Meng and Weirong Liu and Changhong Shi and Zhijun Li and Jie Liu}, doi = {10.1016/j.engappai.2025.111769}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111769}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-information and prediction mask enhanced blind inpainting network for dunhuang murals}, volume = {159}, year = {2025}, }
• (2025). Inter-country trade similarity graph-based long short-term memory for port throughput prediction. EAAI, 159, 111766. (www)

  @article{ , abstract = {In this study, we propose a new port throughput prediction method called inter-country trade similarity graph-based long-short term memory, designed to enhance the efficiency and reliability of port operations. We introduce the inter-country trade similarity graph, a novel representation that not only captures port throughput trends but also integrates trade transaction information between trading countries and the target port, providing a comprehensive perspective for throughput prediction. For graph generation, an inter-country trade similarity is determined by calculating the similarity of trade transaction information between trading countries via text mining techniques. Then, we develop an end-to-end prediction model based on graph-based sequential learning for the inter-country trade similarity graph. By leveraging both the port throughput trends and the inter-country trade relationships encapsulated in the graph, our model delivers precise throughput predictions. We implemented an empirical analysis and model validation in terms of liquid bulk cargo and dry bulk cargo. Experimental results demonstrate that our model outperforms state-of-the-art baseline methods in terms of root mean squared error and unscaled mean bounded relative absolute error, achieving a remarkable accuracy improvement of 6.9% and 13.0% for liquid bulk cargo and 4.7% and 14.5% for dry bulk cargo, highlighting its effectiveness in real-world applications. We believe that our model is a useful tool for strategic decision-making in port-related industries.}, archive = {J_EAAI}, author = {Wonho Sohn and Dongcheol Lim and Suhyeon Kim and Junghye Lee}, doi = {10.1016/j.engappai.2025.111766}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111766}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Inter-country trade similarity graph-based long short-term memory for port throughput prediction}, volume = {159}, year = {2025}, }
• (2025). A lightweight and efficient model for photovoltaic panel defect detection in complex environments. EAAI, 159, 111764. (www)

  @article{ , abstract = {Traditional photovoltaic panel defect detection tasks rely primarily on manual visual inspections, which are inefficient, have low accuracy, and incur high costs. Within this research, we introduce a streamlined yet effective model founded on the “You Only Look Once” algorithm to detect photovoltaic panel defects in intricate settings. Our approach involves enhancing the network structure to effectively extract features from multiple targets, maintaining high accuracy while reducing computational resource requirements. The improvements include replacing the Concatenate-to-Fuse feature fusion module in the backbone network with a new structure and performing cross-stage partial hierarchical enhancement, which reduces computational complexity and model parameter count, thus improving the model's feature representation ability. In the neck part, the slim-neck structure mitigates unnecessary features and reduces the model intricacy by efficiently recycling shallow features to acquire more complex information. We utilize lightweight convolutional networks, such as grouped spatial convolution, to implement convolution operations. Finally, we propose an inner-weighted intersection-over-union metric to optimize the loss function, facilitating model convergence and elevating detection precision. The empirical results demonstrate that, compared with the state-of-the-art, the proposed framework provides a 3.4 % enhancement in mean average precision across intersection-over-union thresholds spanning 0.5 to 0.95 while reducing model computational load by 25 %. Thus, this approach integrates the previously mentioned improvements, aiming to achieve both efficiency and pinpoint accuracy in detection. This technique has demonstrated remarkable success in identifying flaws within solar panels.}, archive = {J_EAAI}, author = {Lixiong Gong and Yuanyuan Wang and Yaowang Lv and Quanlong Wu and Yongping Zhu}, doi = {10.1016/j.engappai.2025.111764}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111764}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight and efficient model for photovoltaic panel defect detection in complex environments}, volume = {159}, year = {2025}, }
• (2025). Generative adversarial network-based prediction of microhole profile drilled with high-energy electron beam on silicon wafer. EAAI, 159, 111763. (www)

  @article{ , abstract = {Drilling with a high-energy electron beam on a semiconductive ceramic substrate is emerging as an effective solution for creating high-aspect-ratio microholes. This method effortlessly surpasses band gaps and facilitates machining with continuous irradiation. However, the inherent brittleness and crystallinity of the semiconductive ceramic substrate hinder handling for quality analysis of the drilled substrate. From experimental drilling, the unique deformation history of the microhole with a high-energy electron beam hints at the potential for predictive modeling for non-destructive analysis of the microhole. In this study, we proposed a conditional generative adversarial network model for the non-destructive analysis of drilled microholes. We collected a limited number of images of hole inlets and cross-sectional holes for training and testing the network model. To effectively build the predictive model with our limited dataset, we introduced a generative adversarial network architecture with embedding a self-attention mechanism with multiple parallel heads. This architecture combines the advantages of the convolutional neural networks and the self-attention mechanism. The proposed architecture showed improvements in training loss and evaluation for image generation compared to the original convolutional neural network. The predictive precision for the inlet diameter, hole straightness, and drilled depth was enhanced by 11.6 %, 8.3 %, and 15 %, respectively. The maximum improvement in predictive accuracy for the inlet diameter, hole straightness, and drilled depth was 36.2 %, 22.98 %, and 58.6 %, respectively. These results indicate that the proposed model not only generated the geometrical profile of the cross-sectional hole but also accurately predicted geometrical dimensions.}, archive = {J_EAAI}, author = {Hyunmin Park and Jun Goo Kang and Jin Seok Kim and Eun Goo Kang and Seung-Kyum Choi and Hyung Wook Park}, doi = {10.1016/j.engappai.2025.111763}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111763}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generative adversarial network-based prediction of microhole profile drilled with high-energy electron beam on silicon wafer}, volume = {159}, year = {2025}, }
• (2025). Hybrid modelling and identification of mechanical systems using physics-enhanced machine learning. EAAI, 159, 111762. (www)

  @article{ , abstract = {Obtaining mathematical models for mechanical systems is a key subject in engineering. These models are essential for calculation, simulation and design tasks, and they are usually obtained from physical principles or by fitting a black-box parametric input–output model to experimental data. However, both methodologies have some limitations: physics based models may not take some phenomena into account and black-box models are complicated to interpretate. In this work, we develop a novel methodology based on discrepancy modelling , which combines physical principles with neural networks to model mechanical systems with partially unknown or unmodelled physics. Two different mechanical systems with partially unknown dynamics are successfully modelled and the values of their physical parameters are obtained. Furthermore, the obtained models enable numerical integration for future state prediction, linearization and the possibility of varying the values of the physical parameters. The results show how a hybrid methodology provides accurate and interpretable models for mechanical systems when some physical information is missing. In essence, the presented methodology is a tool to obtain better mathematical models, which could be used for analysis, simulation and design tasks.}, archive = {J_EAAI}, author = {Mikel Merino-Olagüe and Xabier Iriarte and Carlos Castellano-Aldave and Aitor Plaza}, doi = {10.1016/j.engappai.2025.111762}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111762}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid modelling and identification of mechanical systems using physics-enhanced machine learning}, volume = {159}, year = {2025}, }
• (2025). Enhancing occupational safety in construction: Predictive analytics using advanced ensemble machine learning algorithms. EAAI, 159, 111761. (www)

  @article{ , abstract = {In the construction industry, persistent occupational injury risks demand robust predictive methodologies to identify and mitigate hazards effectively. This study leverages advanced ensemble machine learning algorithms (Ml), Extreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), and a Modified Decision Tree (MDT) that evaluated against a comprehensive Occupational Safety and Health Administration (OSHA) dataset. Our findings reveal that the MDT, with tailored split criteria and advanced pruning techniques, achieved a prediction accuracy of 98.13 %, surpassing traditional models and other algorithms. XGBoost and LightGBM also showed robust performance with prediction accuracies of 92.9 % and 95.97 %, respectively. The efficacy of these models is demonstrated through k-fold cross-validation, with the MDT consistently outperforming others, recording a Mean Absolute Error (MAE) of 6.37, Mean Squared Error (MSE) of 59.57, and Mean Absolute Percentage Error (MAPE) of 7.97. These advancements underscore the potential of these models to enhance predictive accuracy in construction safety, enabling precise safety interventions to reduce risks and improve worker protection.}, archive = {J_EAAI}, author = {Ali Shehadeh and Odey Alshboul}, doi = {10.1016/j.engappai.2025.111761}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111761}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing occupational safety in construction: Predictive analytics using advanced ensemble machine learning algorithms}, volume = {159}, year = {2025}, }
• (2025). Dual-domain contrastive learning for three-dimensional multi-parametric magnetic resonance imaging to end-to-end predict kidney cancer subtypes. EAAI, 159, 111759. (www)

  @article{ , abstract = {Prediction of subtypes is important for clinical decision-making in kidney cancer. Multi-parametric magnetic resonance imaging (mp-MRI) provides a non-invasive way to evaluate tumor characteristics. However, due to the heterogeneity of pixel, modality, and objective representation, the computer-aided diagnosis of subtypes is challenging. In this study, we propose a novel diagnosis framework for kidney cancer subtypes based on mp-MRI, dual-domain contrastive learning network (DCLNet), which has two innovations: (i) the dual-domain contrastive learning scheme based on intra-case consistency and inter-case specificity that mines the correlation and diversity of dual-domain (T1-weighted and T2-weighted) image information, and (ii) the linear diffusion augmentation strategy that enriches training data in three-dimensional image sparse representation and increases the robustness of features. In experiments, a real-world dataset from multiple centers is established for the development and validation of DCLNet. The proposed method yields multiple classification accuracy of 75.49 % for kidney cancer subtypes. The area under the curve for the aggressive malignant tumor clear cell renal cell carcinoma and the benign tumor angiomyolipoma is 89.53 % and 88.95 %, respectively. Significantly, our proposed method demonstrates significant improvement over state-of-the-art methods (p < 0.01). This study offers a reliable model for non-invasive prediction of kidney cancer subtypes. It also shows potential to overcome multi-source heterogeneity and improve performance in cancer classification. Our code is available at https://github.com/xiaojidream/DCLNet .}, archive = {J_EAAI}, author = {Guoying Ji and Lizhi Shao and Yihao Zhu and Xuwen Li and Tianwang Xun and Junxian Wu and Yabo Zhai and Yuan Yuan and Jie lv and Xiaoming Jiang and Xiongjun Ye}, doi = {10.1016/j.engappai.2025.111759}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111759}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-domain contrastive learning for three-dimensional multi-parametric magnetic resonance imaging to end-to-end predict kidney cancer subtypes}, volume = {159}, year = {2025}, }
• (2025). A hierarchical network model for the estimate of the energy expenditure in individuals with type 1 diabetes. EAAI, 159, 111758. (www)

  @article{ , abstract = {Total daily energy expenditure (TDEE) is impacted by many medical conditions, such as diabetes. In the case of type 1 diabetes (T1D), individuals need to have an accurate assessment of the energy expenditure in real-time to avoid dietary imbalance, and improve glycemic control. This work proposes a hierarchical Long Short-Term Memory (LSTM)-based modeling approach to predict real-time continuous energy expenditure, expressed as metabolic equivalents (METs), for individuals with T1D on a 24-hour basis by leveraging the step count and heart rate data from a wrist-band smartwatch. To deal with the inter- and intra-individual variability, the proposed model uses three different LSTMs to capture population, activity-type and subject scale information. To evaluate the impact of the components of the hierarchy, the performance of the proposed hierarchical model was assessed at each level. The results show that the combination of population data, such as heart rate and step counts, with individual data in a hierarchical architecture helps to achieve superior prediction performance, than using only individual heart rate and step counts data.Additionally, compared to non-hierarchical modeling, the hierarchical modeling can provide precise and individualized prediction of the METs categories, as it allows the integration of the variation at different levels of the hierarchy. This model can be used to augment current automated insulin delivery (AID) systems to adapt insulin infusion according to the predicted activity intensity and compensate for glycemic perturbations due to exercise.}, archive = {J_EAAI}, author = {Eleonora M. Aiello and Chiara Toffanin and Michael C. Riddell and Corby K. Martin and Susana R. Patton and Robin L. Gal and Francis J. Doyle III}, doi = {10.1016/j.engappai.2025.111758}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111758}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hierarchical network model for the estimate of the energy expenditure in individuals with type 1 diabetes}, volume = {159}, year = {2025}, }
• (2025). DynaKey-GNN: An efficient dynamic key-node multi-graph neural network for spatio-temporal traffic flow forecasting. EAAI, 159, 111757. (www)

  @article{ , abstract = {Accurate traffic flow prediction is crucial for effective management of urban transportation but remains challenging due to rapidly changing network conditions. We introduce a dynamic key-node multi-graph neural network (DynaKey-GNN) that identifies both long-term vital nodes and short-term critical nodes, enabling real-time adaptation to traffic shifts. An incremental update strategy selectively processes network changes, boosting computational efficiency without sacrificing accuracy. We also propose a dual-stream architecture that fuses global patterns with targeted key-node processing, capturing both stable and fast-evolving dependencies. Experiments on four real-world traffic datasets show that our approach achieves up to 12.37% higher accuracy than state-of-the-art baselines under dynamic scenarios. Case studies on volatile nodes further confirm the model’s ability to handle abrupt fluctuations in traffic flow, providing consistent and reliable forecasts.}, archive = {J_EAAI}, author = {Guangrui Fan and Aznul Qalid Md Sabri and Siti Soraya Abdul Rahman and Lihu Pan}, doi = {10.1016/j.engappai.2025.111757}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111757}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DynaKey-GNN: An efficient dynamic key-node multi-graph neural network for spatio-temporal traffic flow forecasting}, volume = {159}, year = {2025}, }
• (2025). Predicting flood energy reduction in vegetated open channel: Comparative assessment of hybrid artificial intelligence techniques. EAAI, 159, 111756. (www)

  @article{ , abstract = {Flooding is the most devastating natural disaster, causing significant damage to the civil engineering infrastructure. Previous studies reported the impact of vegetation on flow dynamics without providing a precise prediction of flood energy reduction. Therefore, the present study utilized a hybrid artificial intelligence model for the accurate prediction of flood energy reduction. Data series including the hydraulic and geometric conditions of vegetation were collected from the existing literature. Hybrid models, including artificial neural networks, extreme gradient boosting, and support vector regression with Particle Swarm Optimization and random forest, were used. The performance of each artificial intelligence model was evaluated considering the correlation coefficient and mean square error. The result demonstrated superior performance of the extreme gradient boosting with Particle Swarm Optimization model with correlation coefficient and mean square error values of 0.88 and 7.01, respectively. In contrast, an artificial neural network with Particle Swarm Optimization has less predictive power, with correlation coefficient and mean square error values of 0.83 and 7.89. The superior performance of the best model was also confirmed through Analysis of Variance and T-test with P-values of 0.0054 and 0.0062. Furthermore, the best performance of the extreme gradient boosting with Particle Swarm Optimization model was also achieved utilizing Taylor's diagram, in which predicted standard deviation (13.75) matched with measured standard deviation (13.98) compared to other models. The findings of the current research provide insight into the significance of hybrid models, especially the extreme gradient boosting with Particle Swarm Optimization model, while predicting flood energy reduction through vegetation.}, archive = {J_EAAI}, author = {Aïssa Rezzoug and Atif Khan and Nadir Murtaza and Syed Abdul Sami Rizvi}, doi = {10.1016/j.engappai.2025.111756}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111756}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predicting flood energy reduction in vegetated open channel: Comparative assessment of hybrid artificial intelligence techniques}, volume = {159}, year = {2025}, }
• (2025). Digital twin-driven reinforcement learning-based operational management for customized manufacturing. EAAI, 159, 111754. (www)

  @article{ , abstract = {Due to the increasing complexity of customer demands for different batches and types of products, manufacturing operations management has been facing the challenge of uncertain product arrival times and resource processing times in customized manufacturing (CM). This paper proposes a dynamic scheduling method to solve the uncertainty in CM via the integration of the digital twin and fuzzy reinforcement learning methods. In this study, a digital twin-driven framework is first designed to describe the operation management system (OMS) hierarchies. Then a semi-Markov decision process (MDP) model with fuzzy definition is built by abstracting the stochastic scheduling process. To solve the semi-MDP model, an asynchronous multi-edge co-training method is presented to train a fuzzy deep neural network through closed-loop control of virtual commissioning, illustrating how the digital twin-driven OMS adapts to dynamic production requirements. Finally, the proposed method is verified by the performance of comparative experiment. Experimental results show that for randomly arriving products, the proposed method guarantees timely training and scheduling decisions and has the highest total system profit compared to other competing methods (Hybrid Multi-Agent System Negotiation and Ant Colony Optimization (HMA), Onto_MDP, and Deep Q Networks (DQN)). Also, the proposed method shows better scheduling performance in terms of average decision time, average training time and number of finished products when resources are abnormal.}, archive = {J_EAAI}, author = {Hao Tang and Minghao Cheng and Uzair Aslam Bhatti and Bo Xu and Nan Zhou and Rong Guo and Bing Wei}, doi = {10.1016/j.engappai.2025.111754}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111754}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Digital twin-driven reinforcement learning-based operational management for customized manufacturing}, volume = {159}, year = {2025}, }
• (2025). Prediction of frequency spectrum of wall pressure fluctuations beneath turbulent boundary layers based on deep learning. EAAI, 159, 111752. (www)

  @article{ , abstract = {The prediction of wall pressure fluctuation spectrum is crucial in unsteady aerodynamics, aeroelasticity and aeroacoustics of various engineering applications. This paper proposes a surrogate model based on deep learning, which takes the boundary layer parameters of a steady flow field as input and combines a global dimensionless method to train a fully connected neural network through high fidelity unsteady simulation data, achieving fast prediction of wall pressure fluctuation spectrum. The advantages of the new dimensionless method are discussed, and the parameter sensitivity of the model is analyzed. The generalization ability of the model is tested on different operating conditions and airfoils, and compared with traditional semi-empirical models. The results show that the new model has an average prediction error of 2.11–2.53 dB/Hz for four types of airfoils, and its generalization ability is better than semi-empirical models. The prediction accuracy of the model under zero pressure gradient and adverse pressure gradient is significantly better than that of the semi-empirical model, with root mean square error less than 1.75 dB/Hz, and coefficient of determination above 0.92. A single prediction takes only 0.5 s, which is 4 orders of magnitude more efficient than unsteady numerical simulation. This study provides an efficient solution for engineering rapid prediction of wall pressure fluctuation spectra under complex flow conditions. In the future, further improvements of the modeling ability for low-frequency components caused by flow separation and periodic discrete components caused by vortex shedding is needed.}, archive = {J_EAAI}, author = {Yijun Mao and Yongqi Zhang and Chen Xu and Kang Cheng}, doi = {10.1016/j.engappai.2025.111752}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111752}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction of frequency spectrum of wall pressure fluctuations beneath turbulent boundary layers based on deep learning}, volume = {159}, year = {2025}, }
• (2025). Knee osteoarthritis network: A hybrid transformer-based approach for enhanced detection and grading of knee osteoarthritis. EAAI, 159, 111751. (www)

  @article{ , abstract = {Background: Knee osteoarthritis is a prevalent and debilitating condition that significantly affects the elderly population worldwide, leading to pain, joint deformity, and reduced mobility. Early diagnosis and severity grading of knee osteoarthritis are critical for timely intervention but remain challenging due to subjective clinical assessments and the complex progression of the disease. This study aims to develop an automated and reliable artificial intelligence-based deep learning framework for detecting and grading knee osteoarthritis severity using radiographic images. Methods: We propose the Knee Osteoarthritis Network, an innovative hybrid deep learning-based system designed for the automatic detection and grading of knee osteoarthritis. The framework integrates advanced transformer-based models, including Vision Transformer, Swin Transformer, and Faster Vision Transformer, leveraging transfer learning to enhance feature extraction from knee X-ray images. Features extracted from each model are combined using a serial fusion strategy, followed by a hybrid feature selection method employing mutual information and genetic algorithms. The system is evaluated on the publicly available Osteoarthritis Initiative dataset, consisting of 8,260 radiographic images, using stratified five-fold cross-validation. Results: The proposed framework outperforms existing state-of-the-art methods, achieving a classification accuracy of 97.03%, a Cohen’s kappa score of 0.98, a mean absolute error of 0.0312, and a mean squared error of 0.0374. To improve interpretability, we apply explainable artificial intelligence techniques, including Layer Class Activation Mapping and SHapley Additive exPlanations, to visualize class-discriminative regions and identify important features. Conclusions: The experimental results demonstrate that the proposed Knee Osteoarthritis Network offers a reliable and interpretable solution for the automated diagnosis and grading of knee osteoarthritis from radiographic images. By combining transformer-based feature extraction with robust fusion and selection techniques, the framework improves diagnostic accuracy and reduces reliance on manual clinical grading.}, archive = {J_EAAI}, author = {Sarmad Maqsood and Nabeel Maqsood and Shehryar Shahid and Fazal E. Subhan and Muhammad Abdullah Sarwar and Musyyab Yousufi and Ahmad Qurthobi and Ahsan Zafar and Muhammad Attique Khan and Robertas Damaševičius and Rytis Maskeliūnas}, doi = {10.1016/j.engappai.2025.111751}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111751}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Knee osteoarthritis network: A hybrid transformer-based approach for enhanced detection and grading of knee osteoarthritis}, volume = {159}, year = {2025}, }
• (2025). Key node propagation-based overlapping spammer group detection algorithm on e-commerce platforms. EAAI, 159, 111750. (www)

  @article{ , abstract = {With the rapid growth of e-commerce platforms, spammer groups have increasingly used fake reviews to influence consumer decisions, posing significant challenges to platform governance. This issue has become even more pronounced with the widespread use of large language models, which have made fake reviews harder to detect. However, existing spammer group detection algorithms have certain limitations. For example, they often overlook the core–periphery structure within spammer groups, failing to adequately focus on the core reviewers who play a crucial role in group operations. Additionally, these algorithms struggle to detect spammers who are active across multiple groups. To address these challenges, we propose an overlapping spammer group detection algorithm based on key node propagation (KNP-OSG). First, we model the review data as a co-review graph and use the Deep Q-Network algorithm combined with an action filtering mechanism to identify key reviewers, or key spammers, who have a critical impact on spammer group detection. Subsequently, based on the structural relationships among pivotal spammers, an improved label propagation algorithm, copra-g, is proposed to further identify spammer groups. Experimental results show that the KNP-OSG algorithm outperforms existing methods on real-world datasets, demonstrating its effectiveness in detecting overlapping spammer groups.}, archive = {J_EAAI}, author = {Chaoqun Wang and Ning Li and Shuang Chen and Xiaoqing Bu and Shujuan Ji}, doi = {10.1016/j.engappai.2025.111750}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111750}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Key node propagation-based overlapping spammer group detection algorithm on e-commerce platforms}, volume = {159}, year = {2025}, }
• (2025). A vehicle detection method based on cross-scale feature fusion. EAAI, 159, 111749. (www)

  @article{ , abstract = {This paper presents improved algorithms developed to enhance the detection performance of multi-scale vehicles across various lanes. The proposed methods specifically address the challenge that small-scale vehicles in video surveillance systems are susceptible to false positives and missed detections under complex conditions. These limitations ultimately lead to imbalanced detection outcomes across vehicles of different scales. Firstly, the cross-scale feature fusion structure is proposed to enhance the deep fusion capability of multi-scale features. This effectively addresses the issue of deep feature maps struggling to capture small-scale vehicle information caused by excessive downsampling. Secondly, the cross-scale feature fusion module is proposed to enable the model to dynamically capture features from multiple dimensions, facilitating a deeper understanding of both coarse- and fine-grained data and thereby significantly enhancing the performance of multi-scale vehicle detection. Additionally, the downsampling convolution is optimized using Receptive-Field Attention to improve the model's ability to understand the detailed features of multi-scale vehicles. Finally, the Wise-Intersection over Union (Wise-IoU) loss function is utilized to improve the detection performance for low-quality vehicle samples. Experiments on the VisDrone and Vehicle datasets show the number of parameters and model size of the proposed algorithm in this paper have been significantly streamlined. The algorithm can effectively balance the detection performance of multi-scale vehicles, thus obtaining higher overall detection accuracy. In addition, the test results on the DOTAv1 dataset show the proposed method has good generalization ability and cross-scene detection capability.}, archive = {J_EAAI}, author = {Yuyu Meng and Yinbao Ma and Jiuyuan Huo and Hongrui Su}, doi = {10.1016/j.engappai.2025.111749}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111749}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A vehicle detection method based on cross-scale feature fusion}, volume = {159}, year = {2025}, }
• (2025). Pedestrian crossing intention prediction in obstructed scenarios using graph-based feature reconstruction and multi-scale self-attentive encoding. EAAI, 159, 111748. (www)

  @article{ , abstract = {In urban traffic environments, delayed perception of pedestrian motion by drivers increases the risk of pedestrian-vehicle accidents, highlighting the need for accurate prediction of pedestrian crossing intention. However, existing methods struggle with identifying pedestrians under obstructed scenarios and are often affected by redundant environmental features. To address these challenges, this paper proposes a model named Obstruction-CityWalk , which incorporates artificial intelligence (AI) techniques for pedestrian detection and intention prediction. The model consists of two key modules: a pedestrian capture module and an intention prediction module. The pedestrian capture module integrates a graph auto-encoder with an anti-obscuration tracker to reconstruct missing pedestrian features caused by obscuration, ensuring robust detection, and tracking in complex urban traffic scenes. The intention prediction module employs a multi-scale non-uniform self-attentive encoder, enhancing pedestrian crossing feature representation while mitigating background noise and improving prediction accuracy. Experimental results on the selected Joint Attention in Autonomous Driving (JAAD) and Pedestrian Intention Estimation (PIE) datasets show that the proposed model achieves state-of-the-art performance on the PIE dataset, surpassing existing approaches with a 3 % improvement in accuracy, alongside a 6 % improvement in accuracy on the JAAD dataset. Furthermore, the anti-obscuration tracker enhances the consistency of pedestrian identification, improving tracking performance in obstructed scenarios, while the multi-scale non-uniform self-attentive encoder boosts accuracy by 6 %. These findings demonstrate the effectiveness of the proposed approach in pedestrian crossing intention prediction using AI in obstructed urban scenarios.}, archive = {J_EAAI}, author = {Zhuofan Liu and Yigang Xue and Lan Zhu}, doi = {10.1016/j.engappai.2025.111748}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111748}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pedestrian crossing intention prediction in obstructed scenarios using graph-based feature reconstruction and multi-scale self-attentive encoding}, volume = {159}, year = {2025}, }
• (2025). Fault diagnosis of multi-unit nonlinear industrial processes based on a global–local attention convolutional neural network. EAAI, 159, 111747. (www)

  @article{ , abstract = {Recently, fault diagnosis technology has attracted much attention in multi-unit nonlinear industrial processes, which are composed of multiple interconnected subunits, with their complexity arising from nonlinear coupling relationships among these units. Unfortunately, existing fault diagnosis models have not well balanced the influence between inner-unit and cross-unit information. To address the above problem, a global–local attention convolutional neural network (GLACNN) method is proposed to improve the accuracy and interpretation of fault diagnosis results. Firstly, two special one-dimensional convolution kernels are designed to extract global features from the whole system. Secondly, local features from different units are derived in parallel approaches. Then, the corresponding global and local features are stacked to obtain the fusion features. The contribution of each unit to fault diagnosis will be assigned a proper weight through a channel attention module, which represents the local and global significance of different features in each unit. The efficacy of the proposed method is validated through its application in the Tennessee Eastman (TE) process and the penicillin production fermentation process.}, archive = {J_EAAI}, author = {Xueqin Yang and Yun Wang and Lijuan Qian and Gangyue Ye and Weirong Ye and Yuchen He}, doi = {10.1016/j.engappai.2025.111747}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111747}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis of multi-unit nonlinear industrial processes based on a global–local attention convolutional neural network}, volume = {159}, year = {2025}, }
• (2025). An interpretable dynamic bayesian network method for time-varying seismic liquefaction uplift risk assessment of underground rectangular structures. EAAI, 159, 111746. (www)

  @article{ , abstract = {Earthquakes can cause liquefaction-induced uplift in subsurface structures within saturated sandy soils, leading to severe damage. Existing methods fail to fully analyze this dynamic process and consider time-varying characteristics of influencing factors. This study proposes a liquefaction uplift risk assessment method using Dynamic Bayesian Networks (DBN). The DBN method predicts final uplift displacement and liquefaction triggering time while reflecting time-varying characteristics of uplift-influencing factors. A 0.5 s time slice is found to optimize prediction accuracy. Results show DBN improves back judgment accuracy by 1.21 % and 7.71 %, and prediction accuracy by 8.09 % and 5.34 % compared to the two static Bayesian Network (BN) methods. Dynamic sensitivity analysis reveals the water table as the main early-stage influencing factor, with the influence of the Aria's intensity and relative density of sand increasing with depth and liquefaction completion. The method's validity and generalization were verified using centrifuge test data, achieving 0.78 accuracy compared to 0 for traditional formula methods. This study introduces a novel approach for considering time-varying characteristics in engineering design, risk management, and decision-making for liquefaction-prone areas, offering improved insights into the dynamic process of liquefaction-induced uplift.}, archive = {J_EAAI}, author = {Jing Wang and Jilei Hu and Luou Pang}, doi = {10.1016/j.engappai.2025.111746}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111746}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An interpretable dynamic bayesian network method for time-varying seismic liquefaction uplift risk assessment of underground rectangular structures}, volume = {159}, year = {2025}, }
• (2025). A new deep sparse unfolding network with variational bayesian enhancement for high-resolution localization of acoustic sources. EAAI, 159, 111744. (www)

  @article{ , abstract = {The Deep sparse unfolding network has been proven to be interpretable and efficient, which has broad prospects for high-resolution acoustic beamforming. However, the performance of the network model depends on the accurate establishment of high-quality datasets and constraint equations. Therefore, the performance and stability of the network model may decline under the presence of complex source excitation and uncertain interference conditions. To address these issues, we propose a high-resolution acoustic beamforming sparse unfolding network enhanced by variational Bayesian. Firstly, we introduce an adaptive proxy model to generate a high-quality target map unaffected by frequency influence and construct a complete dataset. Then, we establish a deconvolution optimization problem and threshold iteration solution method. On this basis, we construct the Learned Iterative Shrinkage Thresholding Algorithm with Nesterov to enhance solution rates and unroll it into a network model. Subsequently, we propose a variational Bayesian enhancement mechanism to enhance the sparsity of the iterative solution results, improve the diversity of model training data, and characterize imaging position deviation loss. By combining energy deviations, we construct a new loss function to ensure a balance between imaging accuracy and energy concentration. We have developed an interpretable high-resolution imaging network, which we refer to as the Leaning-Nesterov-Variational-Bayesian-Enhancements Network (LNVBENet). Finally, we conduct research and analysis on two available datasets and experimental data to assess the feasibility and superiority of LNVBENet. The results show that LNVBENet not only enhances imaging accuracy and stability but also offers faster imaging speeds and robust anti-interference capabilities.}, archive = {J_EAAI}, author = {Siguo Wen and Kai Zheng and Pu Li and Yuying Fang and Yin Bai and Dewei Yang}, doi = {10.1016/j.engappai.2025.111744}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111744}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new deep sparse unfolding network with variational bayesian enhancement for high-resolution localization of acoustic sources}, volume = {159}, year = {2025}, }
• (2025). Neural network-based analysis of clinical and demographic variables for predicting platelet counts and prothrombin time in chronic kidney disease patients. EAAI, 159, 111741. (www)

  @article{ , abstract = {Objective In chronic kidney disease (CKD), critical health indicators such as Platelet count, Prothrombin Time (PT), and depression play pivotal roles in patient management and outcome prediction. Platelets and PT are essential in assessing the coagulation status, which can be compromised in CKD, while depression significantly impacts the quality of life and treatment adherence. This study employs machine learning (ML) techniques to unravel the complexities of variable interactions in CKD and their collective influence on Platelet counts and PT levels. Specifically, we aim to leverage the capabilities of neural networks to dissect and understand how various patient characteristics—including biological sex, age, and pre-existing medical conditions—affect these crucial blood parameters. By doing so, we seek to provide clinically actionable insights that could support personalized monitoring and more effective risk stratification in CKD management. Data source and setting The data for this study were derived from an open-source dataset, originally utilized in the study by Li et al. and also published on Kaggle, an open data platform. Methods Data were sourced from the Medical Information Mart for Intensive Care (MIMIC-III) Database, comprising a cohort of 1177 patients for the analysis. The dataset was randomly split into a training set and a validation set. We utilized a sequential neural network (NN) model trained using backpropagation to predict the levels of platelet count, PT, and depression indicators. Performance of the model was evaluated on the validation set, focusing on loss reduction over training iterations and accuracy improvement. Additionally, quantitative metrics such as Mean Squared Error (MSE) and Mean Absolute Error (MAE) were used to assess prediction accuracy. Results The NN model was trained to predict platelet count and PT levels, which are critical indicators in patients with chronic kidney disease. The model achieved an accuracy of 65.01 % for platelet count prediction (MSE = 0.0098) and 75.85 % for PT prediction (MSE = 0.0091), with the training loss rapidly decreasing to a plateau—indicating stable convergence. For platelet count, regression analysis identified blood acidity (pH) (β = 0.180, p < 0.001) and age (β = −0.162, p < 0.001) as the most significant predictors. In contrast, Shapley Additive Explanations (SHAP)-based feature importance analysis ranked pH and gender as the most impactful features, followed by hyperlipemia and age, highlighting the non-linear interactions captured by the neural network. Similarly, PT prediction was significantly influenced by hyperlipemia (β = 1.4162, p < 0.05) and hypertensive status (β = −1.5667, p < 0.05) in the regression model, and these two variables were also identified by SHAP as the most influential contributors. These findings underscore the complementary strengths of traditional statistical methods and explainable artificial intelligence (AI) techniques like SHAP in capturing both direct effects and complex, non-linear relationships, ultimately improving model interpretability and clinical applicability in predictive healthcare analytics. Conclusions The Backpropagation NN model utilized in this study represents a robust approach for analyzing and predicting key coagulation indicators in CKD patients. By identifying the most influential clinical variables driving platelet count and PT variation, this framework offers practical value in enhancing patient-specific risk assessment, monitoring strategies, and individualized treatment planning. Despite limitations related to dataset size, the model provides a foundation for clinically meaningful decision-support tools in nephrology.}, archive = {J_EAAI}, author = {Simin Nazari and Amira Abdelrasoul}, doi = {10.1016/j.engappai.2025.111741}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111741}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural network-based analysis of clinical and demographic variables for predicting platelet counts and prothrombin time in chronic kidney disease patients}, volume = {159}, year = {2025}, }
• (2025). Intrinsic-motivation multi-robot social formation navigation with coordinated exploration. EAAI, 159, 111740. (www)

  @article{ , abstract = {This paper investigates the application of reinforcement learning (RL) to multi-robot social formation navigation, a critical capability for enabling seamless human–robot coexistence. While RL offers a promising paradigm, the inherent unpredictability and often uncooperative dynamics of pedestrian behavior pose substantial challenges, particularly concerning the efficiency of coordinated exploration among robots. To address this, we propose a novel coordinated-exploration multi-robot RL algorithm introducing an intrinsic motivation exploration. Its core component is a self-learning intrinsic reward mechanism designed to collectively alleviate policy conservatism. Moreover, this algorithm incorporates a dual-sampling mode within the centralized training and decentralized execution framework to enhance the representation of both the navigation policy and the intrinsic reward, leveraging a two-time-scale update rule to decouple parameter updates. Empirical results on social formation navigation benchmarks demonstrate the proposed algorithm’s superior performance over existing state-of-the-art methods across crucial metrics. Our code and video demos are available at: https://github.com/czxhunzi/CEMRRL .}, archive = {J_EAAI}, author = {Hao Fu and Wei Liu and Shuai Zhou}, doi = {10.1016/j.engappai.2025.111740}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111740}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intrinsic-motivation multi-robot social formation navigation with coordinated exploration}, volume = {159}, year = {2025}, }
• (2025). Evaluation of a newly developed ploughshare: An ensemble deep learning approach for soil surface roughness classification with explainable artificial intelligence. EAAI, 159, 111739. (www)

  @article{ , abstract = {The roughness of the soil surface directly affects seedbed preparation, sowing operations and crop yield. The roughness of the soil surface depends on the equipment used in the tillage process. Evaluating the performance of this equipment is important for increasing agricultural productivity. Recent advances in artificial intelligence (AI) and image processing technologies use convolutional neural network (CNN) models that can automatically learn meaningful patterns from images, enabling the objective analysis of soil surface roughness. Rather than analyzing a single CNN model, ensemble deep learning (EDL) approaches combining predictions from multiple CNN models improve classification accuracy. The aim of this study is to classify soil surface images obtained from primary tillage operations performed with a newly developed mouldboard ploughshare (NDP) and a conventional mouldboard ploughshare (CP) using the EDL method. Images were taken from the field to evaluate the soil surface structure using NDP and CP after primary tillage. These images were then enhanced using the gradient-weighted class activation mapping (Grad-CAM) method to generate heat maps, and the generated images were classified using known CNN and EDL models. The hard voting technique was used for the EDL approach. The Voting 6 model achieved 98.3 % accuracy. In addition, according to the results of the profilometer measurement, it was determined that the soil cutting and fragmentation performance of NDP was higher than that of CP. As a result, this image-based method can significantly contribute to the testing process of primary tillage tools.}, archive = {J_EAAI}, author = {Ali Yavuz Şeflek and Birkan Büyükarıkan and Keziban Yalçın Dokumacı}, doi = {10.1016/j.engappai.2025.111739}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111739}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evaluation of a newly developed ploughshare: An ensemble deep learning approach for soil surface roughness classification with explainable artificial intelligence}, volume = {159}, year = {2025}, }
• (2025). Semantic knowledge and data modeling of environmental justice. EAAI, 159, 111736. (www)

  @article{ , abstract = {Most environmental justice information and data are unstructured, posing substantial challenges for effective risk management, disaster response, and comprehensive environmental justice assessments through software tools. To overcome these barriers, we created the Environmental Justice Ontology, a structured framework that organizes essential environmental justice concepts and their complex relationships, enabling machines to interpret environmental justice knowledge. The ontology applies the Semantic Web Rule Language to model EJScreen measures, including % Unemployment and % People of Color. Leveraging the widely adopted Basic Formal Ontology and Common Core Ontologies, Environmental Justice Ontology organizes environmental justice resources—including reports, charts, and datasets—to enable seamless data exchange across various domains. We applied the ontology to structure environmental justice data gathered from the Proctor Creek Watershed in Atlanta, Georgia, using the U.S. Environmental Protection Agency's EJScreen tool. Applying Neo4J AuraDB and natural language processing techniques in Python, we developed two knowledge graphs to support the visualization, querying, and analysis of this environmental justice data. Together, the ontology and knowledge graphs create a structured framework for uncovering complex interactions and correlations between natural and social systems, fostering a comprehensive understanding of the interconnections between environmental, socioeconomic, demographic, and other influencing factors. This framework empowers users and software to assist policymakers in designing sustainable solutions for overburdened communities facing environmental, health, and climate-related challenges. The ontology encompasses a broad range of classes—including disparities, burdens, discrimination, bias, and the challenges faced by disadvantaged communities—that extends beyond the capabilities of EJScreen, enabling seamless integration of future environmental justice data and knowledge.}, archive = {J_EAAI}, author = {Armita Davarpanah and Hassan A. Babaie and Fatemeh Shafiei and Na'Taki Osborne Jelks}, doi = {10.1016/j.engappai.2025.111736}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111736}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semantic knowledge and data modeling of environmental justice}, volume = {159}, year = {2025}, }
• (2025). Unsupervised domain adaptation framework with global-local adversarial learning and masked image consistency for fish counting in deep-sea aquaculture. EAAI, 159, 111735. (www)

  @article{ , abstract = {Accurate fish counting is crucial for effective management in deep-sea aquaculture. However, the diverse deep-sea aquaculture environments pose significant challenges to the generalizability of fish counting models. Current methods rely heavily on extensive labeled datasets and struggle to adapt to unseen scenarios. To address these limitations, this study proposes an unsupervised domain adaptation framework that leverages global-local adversarial learning and masked image consistency for cross-domain fish counting. A global-local discriminator is designed to extract the domain-invariant features by aligning the density map prediction across domains at the image and patch levels. Additionally, a masked image consistency module is designed to enhance the utilization of spatial context in the target image by enforcing prediction consistency between masked and complete target images. To validate our approach, we established four fish counting datasets using cameras with varying lighting conditions and viewpoints from two actual deep-sea cages. The framework was evaluated on three domain adaptation tasks: (1) natural to artificial lighting, (2) fixed to free viewpoints, and (3) “Shenlan 1” to “Genghai 1” cages. Experimental results demonstrate that the proposed method significantly improves generalization to diverse aquaculture scenarios without requiring additional data annotations. It outperformed state-of-the-art methods, reducing the mean absolute percentage error by 17.34 % for cross-view counting and 6.77 % for cross-cage counting.}, archive = {J_EAAI}, author = {Hanchi Liu and Xin Ma}, doi = {10.1016/j.engappai.2025.111735}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111735}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised domain adaptation framework with global-local adversarial learning and masked image consistency for fish counting in deep-sea aquaculture}, volume = {159}, year = {2025}, }
• (2025). Towards understanding human actions through long-short-term semantic motion encoding. EAAI, 159, 111733. (www)

  @article{ , abstract = {Human action recognition has achieved competitive performance since the advancement of deep learning and large-scale training datasets. However, recent approaches tend to leverage contextual features instead of interpreting human actions based on semantic definitions. This study presents a long-short-term semantic motion encoding method (LSME) based on human pose sequences, aiming to understand generically semantic action representation over an entire video. The three-dimensional convolutional neural network is proposed to model short-term motions, followed by a two-dimensional temporal convolution for long-term semantic encoding. Finally, the Euclidean distance and Additive Angular Margin Loss based methods are explored to encode action signatures of unseen actions in the open-set recognition problem. Therefore, LSME can understand actual semantic action definitions and recognise unseen actions with negligible data and computational cost adjunction instead of retraining a new model. By fusing a spatial pathway, the proposed model has achieved state-of-the-art performance on action recognition according to the experiments carried out on NTU (Nanyang Technological University) and Kinetics datasets, with the accuracy of 93.82 % on NTU 60, 53.04 % top-1 and 79.32 % top-5 accuracies on Skeleton-Mimetics. These results show that the proposed model is capable of encoding semantic action features whilst speeding up understanding events for real-world applications. It has also demonstrated promising applicability and extensibility for solving challenging real-world human action understanding problems.}, archive = {J_EAAI}, author = {Chaolong Zhang and Yuanping Xu and Zhijie Xu and Chao Kong and Benjun Guo and Jin Jin and Xiaowei Song and Weiye Wang}, doi = {10.1016/j.engappai.2025.111733}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111733}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards understanding human actions through long-short-term semantic motion encoding}, volume = {159}, year = {2025}, }
• (2025). Mamba enhanced you only look once network with multiscale spatial attention for remote sensing object detection. EAAI, 159, 111732. (www)

  @article{ , abstract = {Despite significant progress in remote sensing object detection (RSOD), most existing algorithms still fail to effectively integrate global and local features and adequately aggregate precise location information with high-level semantic features. This results in insufficient feature representation for small objects and limits further improvements in detection accuracy. To address these problems, we proposed an improved you only look once network. First, we design a Mamba enhanced spatial pyramid pooling-fast (SPPF-Mamba), which can integrate the local features with the global features. Second, we design a novel multiscale adaptive spatial attention (MASA) module and integrate it into the high-level screening-feature pyramid network (HS-FPN), which is named enhanced spatial feature high-level screening-feature pyramid network (ES-HS-FPN). ES-HS-FPN can efficiently aggregate the abstract semantic information with precise location information of the objects. In addition, the channel and spatial mechanism in ES-HS-FPN has the ability to inhibit complex background information and enhances the prominence of small objects in RSIs. Extensive experiments have been conducted on three widely recognized remote sensing image datasets. The results indicate that our method has superior detection performance compared with many state-of-the-art (SOTA) methods.}, archive = {J_EAAI}, author = {Yinggan Tang and Ziteng Zhao and Ouhan Huang}, doi = {10.1016/j.engappai.2025.111732}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111732}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mamba enhanced you only look once network with multiscale spatial attention for remote sensing object detection}, volume = {159}, year = {2025}, }
• (2025). An enhanced framework for small object detection with middle-order interaction and adaptive cross-scale aggregation. EAAI, 159, 111730. (www)

  @article{ , abstract = {Small object detection holds significant research value in fields such as autonomous driving, intelligent traffic monitoring, and military reconnaissance. However, in practical applications, the performance of commonly used detection algorithms is often suboptimal due to limited pixel information, dense distribution, and susceptibility to background interference for small objects. To enhance the accuracy of small object detection, we propose an improved algorithm based on You Only Look Once version 8 (YOLOv8), termed You Only Look Once version 8-For Small-object Detection (YOLOv8-FSD). First, we introduce Spatial-Channel Aggregation Module (SCA Module) into feature extraction network, leveraging multi-order gated aggregation mechanism to adaptively capture and integrate interactions among different orders of features, thereby enhancing representation of multi-order features. Second, we design Attention-Guided Path Aggregation Feature Pyramid Network (AG-PAFPN) to add pathways for extracting and aggregating information from large feature maps, generating more discriminative features. Additionally, Multi-Scale Feature Extraction Module (MSFE Module) and Three-Branch Feature Fusion Module (TBFF Module) are proposed to extract and fuse multi-scale spatial information across multiple scales and finer granularity. Finally, we optimize the bounding box regression loss function to accelerate convergence and improve detection accuracy. Experimental results on Visual Detection of Drones (VisDrone) dataset show that the proposed YOLOv8s-FSD achieves mean Average Precision at Intersection over Union (IoU) = 0.50 (mAP50) and mean Average Precision at IoU = 0.50-0.95 (mAP50-95) improvements of 8.3% and 5.5% over YOLOv8s. Migration experiments on TinyPerson and Dataset for Object deTection in Aerial images (DOTA) further validate the effectiveness and robustness of the proposed enhancements.}, archive = {J_EAAI}, author = {Zihan Guo and Xingyu Mu and Chao Chang and Weijie Cheng and Xincheng Tian}, doi = {10.1016/j.engappai.2025.111730}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111730}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An enhanced framework for small object detection with middle-order interaction and adaptive cross-scale aggregation}, volume = {159}, year = {2025}, }
• (2025). Multi-scale frequency attention fusion network for infrared and visible image fusion. EAAI, 159, 111728. (www)

  @article{ , abstract = {The goal of visible and infrared image fusion is to generate a composite image that not only preserves fine-grained textures from visible images but also highlights salient targets from infrared images. However, existing methods often struggle to capture detailed features and typically rely solely on spatial domain information, overlooking the complementary advantages offered by frequency domain features. To address these limitations, we propose a feature-level fusion approach based on a multi-scale frequency attention fusion network, which incorporates a spatial-frequency attention fusion module with cross-attention and a multi-scale depthwise separable convolution block equipped with coordinate attention. Moreover, a multi-scale compensation fusion is incorporated within the spatial frequency attention fusion module to reduce cross-modal domain discrepancies. This work applies deep learning-based artificial intelligence (AI) techniques to the field of multi-modal image fusion. Experiments on three public datasets show that our method achieves competitive performance across multiple evaluation metrics compared to state-of-the-art methods. Our method demonstrates superior performance in preserving fine details and enhancing target saliency. The code will be available at https://github.com/ioschunsheng1230 .}, archive = {J_EAAI}, author = {Yong Wang and Xueyuan Zhao and Jianfei Pu and Lulu Zhang and Duoqian Miao}, doi = {10.1016/j.engappai.2025.111728}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111728}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale frequency attention fusion network for infrared and visible image fusion}, volume = {159}, year = {2025}, }
• (2025). Skin lesion classification network based on improved MobileViT. EAAI, 159, 111726. (www)

  @article{ , abstract = {Computer-aided diagnosis can alleviate the pressure on dermatologists while helping patients to recognize skin lesions at an early stage. Recent studies have shown that integrated networks can effectively handle dermatological classification, but they are not conducive to practical applications. Meanwhile, lightweight networks are often limited by network depth, which somewhat affects their performance when facing complex classification tasks. For this purpose, we present a skin lesion classification network. The network acquires local features via a convolutional layer and then models the global features using the Transformer structure. We designed the enhanced feature representation module, which focuses on the spatial features of the receptive field while further incorporating the directional weights, and is able to provide a richer feature representation. In addition, the cosine similarity downsampling module is proposed in order to reduce the information loss caused by the baseline network during downsampling as well as to enhance the ability to capture key features. This module uses parallel downsampling branches and further introduces cosine similarity for weighting to focus on similar feature information. We perform validation experiments on three publicly available dermatological datasets, which include clinical images and dermoscopic images. Experimental results on three datasets show that our network is able to better balance model complexity and skin lesion classification performance compared to some state-of-the-art classification networks and skin lesion classification networks.}, archive = {J_EAAI}, author = {Fuyun He and Rongqing Wu and Xuqing Zeng and Haixing Song and Guanglian Li and Zhen Wei}, doi = {10.1016/j.engappai.2025.111726}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111726}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Skin lesion classification network based on improved MobileViT}, volume = {159}, year = {2025}, }
• (2025). Economic dispatch of virtual power plant using reinforcement learning method with improved state space and hybrid action representation. EAAI, 159, 111725. (www)

  @article{ , abstract = {The uncertainty of renewable energy sources (RESs) has a serious impact on the cost-effectiveness of virtual power plant (VPP). To address this issue, this paper proposes a VPP economic dispatch method based on reinforcement learning (RL) with improved state space and hybrid action representation, aiming to minimize the operating cost of the VPP to the greatest extent while ensuring its stable operation. Firstly, a data-driven method was designed to construct typical random output scenarios of renewable energy, expand the state space, and enable the agent to use more effective data information and better capture environmental uncertainty. Secondly, this study proposes a hybrid action latent representation method that embeds the dependence between discrete and continuous actions using an encoder-decoder structure and is trained via environmental dynamics prediction. This solves the problem that in the traditional hybrid action deep reinforcement learning (DRL) method, scalability, stability and discrete-continuous action dependence cannot be taken into account simultaneously. Finally, the hybrid action policy learning is implemented using the twin delayed deep deterministic policy gradient (TD3) algorithm. Numerical simulations in two VPP systems of different sizes compare model predictive control and existing hybrid action DRL methods. The proposed method achieves the lowest operating cost in all cases, proving that it can provide a more economical operation solution for the VPP system with fluctuations in renewable energy.}, archive = {J_EAAI}, author = {Mingde Liu and Jianquan Zhu and Mingbo Liu}, doi = {10.1016/j.engappai.2025.111725}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111725}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Economic dispatch of virtual power plant using reinforcement learning method with improved state space and hybrid action representation}, volume = {159}, year = {2025}, }
• (2025). Dynamic anomaly detection in blast furnace operations using a transformer-enhanced isolation forest framework. EAAI, 159, 111724. (www)

  @article{ , abstract = {Ensuring the stable operation of blast furnaces is essential for sustaining production efficiency and ensuring product quality. However, detecting operational anomalies remains a significant challenge due to the complexity of industrial processes and the limitations of traditional detection methods, which struggle with high-dimensional data and dynamic patterns. This study proposes a novel Transformer-Enhanced Isolation Forest (TF-IF) method that seamlessly integrates the deep feature extraction capabilities of Transformers with the efficiency of Isolation Forest models. The TF-IF approach uniquely captures multi-scale dynamic representations across multiple time windows using Transformer architectures, enabling precise identification of subtle and complex anomaly patterns. It further employs the Isolation Forest for efficient anomaly scoring and robust detection. Experimental results show that TF-IF consistently outperforms traditional unsupervised models, delivering superior detection accuracy, stable score distributions, and improved efficiency while effectively minimizing false negatives. This innovative framework addresses industrial challenges in anomaly detection by bridging advanced deep learning techniques with efficient anomaly detection methods. The TF-IF method provides a practical, scalable, and highly effective solution for blast furnace condition monitoring and anomaly warning systems, offering substantial potential to enhance safety and operational efficiency in industrial metallurgy.}, archive = {J_EAAI}, author = {Junyi Duan and Jinchao Feng}, doi = {10.1016/j.engappai.2025.111724}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111724}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic anomaly detection in blast furnace operations using a transformer-enhanced isolation forest framework}, volume = {159}, year = {2025}, }
• (2025). Monocular three-dimensional object detection for proximity monitoring in human-machine collision warning systems on construction sites. EAAI, 159, 111722. (www)

  @article{ , abstract = {Monitoring workers’ proximities to avoid struck-by hazards has aroused great concern in construction safety management. Existing methods are either too laborious and costly to apply extensively or lack spatial perception for accurate monitoring. This study proposes a novel framework for proximity monitoring using only an ordinary two-dimensional (2D) camera to realize human-machine collision warning, which integrates a monocular three-dimensional (3D) object detection model and a post-processing classification module to identify four proximity categories: Dangerous , Potentially Dangerous , Concerned , and Safe . A new dataset containing 22,500 virtual and real-world construction images with 3D bounding box annotations has been created and publicly released to facilitate system development and evaluation. Experiments show that the implemented system is rapid-response and camera carrier-independent, achieving promising proximity detection performance on both virtual and real-world data, with mean precision, recall, and F1 scores of approximately 0.8, 0.7, and 0.8, respectively, within a range of 50 meters. This study preliminarily reveals the potential and feasibility of proximity monitoring using only a 2D surveillance camera, providing a new, promising, and affordable way for early warning of human-machine collisions.}, archive = {J_EAAI}, author = {Yuexiong Ding and Qiong Liu and Ankang Ji and Heng Li and Xiaowei Luo}, doi = {10.1016/j.engappai.2025.111722}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111722}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Monocular three-dimensional object detection for proximity monitoring in human-machine collision warning systems on construction sites}, volume = {159}, year = {2025}, }
• (2025). Coverage-guaranteed speech emotion recognition via calibrated uncertainty-adaptive prediction sets. EAAI, 159, 111721. (www)

  @article{ , abstract = {Road rage, often triggered by emotional suppression and sudden outbursts, significantly threatens road safety by causing collisions and aggressive behavior. Speech emotion recognition technologies can mitigate this risk by identifying negative emotions early and issuing timely alerts. However, current speech emotion recognition methods, such as those based on hidden Markov models and Long short-term memory networks, primarily handle one-dimensional signals, frequently experience overfitting, and lack calibration, limiting their safety–critical effectiveness. We propose a novel risk-controlled prediction framework providing statistically rigorous guarantees on prediction accuracy. This approach employs a calibration set to define a binary loss function indicating whether the true label is included in the prediction set. Using a data-driven threshold β , we optimize a joint loss function to maintain an expected test loss bounded by a user-specified risk level α . Evaluations across five baseline models and three benchmark datasets demonstrate our framework consistently achieves a minimum coverage of 1 − α , effectively controlling marginal error rates despite varying calibration-test split ratios (e.g., 0.1). The robustness and generalizability of the framework are further validated through an extension to small-batch online calibration under a local exchangeability assumption. We construct a non-negative test martingale to maintain prediction validity even in dynamic and non-exchangeable environments. Cross-corpus and cross-language tests confirm our method’s ability to uphold reliable statistical guarantees in realistic, evolving data scenarios.}, archive = {J_EAAI}, author = {Zijun Jia and Diyin Tang and Hongyu Long and Jinsong Yu}, doi = {10.1016/j.engappai.2025.111721}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111721}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Coverage-guaranteed speech emotion recognition via calibrated uncertainty-adaptive prediction sets}, volume = {159}, year = {2025}, }
• (2025). Forecasting childhood and adolescent obesity prevalence in türkiye through 2030: A novel application of artificial neural networks for public health planning. EAAI, 159, 111720. (www)

  @article{ , abstract = {This study employs artificial neural networks (ANNs) to predict childhood and adolescent obesity trends in Türkiye from 1990 to 2030, with key input variables such as population, GDP, and health indicators. Comparative analyses determined that the 5-neuron model achieved the highest accuracy with an R 2 value of 0.9737, demonstrating superior prediction performance compared to other models with lower mean squared error and mean absolute error values. ANNs were observed to provide more reliable results compared to traditional curve fitting methods. According to these estimates, obesity prevalence among children and adolescents in Türkiye is projected to increase by at least 1 % and reach 12 % by 2030. Garson, Yoon, and Shapley additive explanations algorithms were employed for the relative importance analysis of the inputs of the ANNs. In the best-performing 5-neuron model, the weights of input variables were largely similar in the Garson and Yoon algorithms; however, the ‘population’ variable was found to have a higher weight than other inputs in both Yoon (13.7 %) and Garson (16.7 %) analyses. The findings provide a significant reference for shaping public health policies and developing early intervention strategies. Future studies should integrate diverse machine learning approaches and analyze comprehensive socioeconomic data.}, archive = {J_EAAI}, author = {Halil Şenol and Halil Çolak}, doi = {10.1016/j.engappai.2025.111720}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111720}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forecasting childhood and adolescent obesity prevalence in türkiye through 2030: A novel application of artificial neural networks for public health planning}, volume = {159}, year = {2025}, }
• (2025). Order reduced gaussian process fused physical information for assessing long-term multiple properties using very little high-variance experimental data -a real-world concrete dam application. EAAI, 159, 111718. (www)

  @article{ , abstract = {Predicting the multi-dimensional temporal material properties presents significantly concerns and challenging, especially using the very limited experimental data having high variance. Therefore, we present a novel data-driven artificial intelligence algorithm, the Order Reduced Gaussian Process (ORGP), designed to accurately and simultaneously predict the long-term multiple temporal mechanical properties of engineering material using only very little high-variance experimental data fused physical information. In which, we leverage Bayesian inference with a well-defined structure that provides clear physical significance to kernel functions and a clear understanding of the effects of hyperparameters. Meanwhile, we fuse material physical information (e.g. the boundedness and time-varying monotonicity of material properties) with the limited and highly-variant experimental data, achieving the prediction conforms to objective physical laws naturally. Specifically, taking the hydraulic concrete materials of a real-world dam as the example of engineering applications, we first comprehensively analyze the inherent high variance of experimental hydraulic concrete data over time and different ingredients. We then integrate physical information with ORGP and construct a mapping from concrete ingredients to dam's long-term multiple engineering properties, accurately predicting eight temporal material properties (e.g., R-Square> 99.83 %), e.g., slump, compressive strength and ultimate tensile strain. More importantly, we compare the results by ORGP with ones by three well-established models, namely multi-layer perceptron, K-nearest neighbors and random forest, as well as comprehensively investigating its convergency, the influence of training data size, demonstrating the ORGP can achieve significantly higher accuracy (e.g., the smallest Root Mean Square Error) with faster convergence speed using few data (e.g., 80 samples of training set), having strong robustness and noise immunity.}, archive = {J_EAAI}, author = {Yunguo Cheng and Mengxi Zhang and Shuai Zhang and Dan Tian and Mingchao Li and Jingyi Zhang and Chensen Ding}, doi = {10.1016/j.engappai.2025.111718}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111718}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Order reduced gaussian process fused physical information for assessing long-term multiple properties using very little high-variance experimental data -a real-world concrete dam application}, volume = {159}, year = {2025}, }
• (2025). Cooperative reconnaissance coverage for heterogeneous unmanned aerial vehicle swarm in unknown communication interference environments. EAAI, 159, 111717. (www)

  @article{ , abstract = {In complex environments with uncertain communication interference, heterogeneous unmanned aerial vehicle (HUAV) swarm often encounter the challenge of communication disruptions while collaboratively executing reconnaissance and coverage missions, hindering efficient information sharing and subsequently leading to substantial issues of redundant coverage. Given that existing research overlooks the impact of communication interference, this paper proposes a Coverage-Oriented Multi-Agent Cooperative Artificial Potential Field (MACAPF) algorithm. Firstly, a communication model for HUAV swarm under communication interference is considered to accurately reflect real-time communication status. Secondly, an autonomous collaborative distributed-concentrated architecture is devised, which dynamically adjusts the distributed-concentrated state of the swarm based on varying communication conditions, providing support for resolving communication disruptions faced by the HUAV swarm. Lastly, addressing the limitations of traditional artificial potential field (APF) algorithm in unknown communication interference environments, individualized definitions of the HUAV potential field are introduced, and the MACAPF algorithm is designed based on the autonomous collaborative distributed-concentrated architecture. This algorithm effectively guides HUAVs experiencing communication disruptions to restore communication, enhancing the communication efficiency and cooperative operation capabilities of the HUAV swarm. Simulation results demonstrate that the proposed MACAPF algorithm exhibits significant advantages over other state of the art (SOTA) algorithms across multiple dimensions under various signal interference intensities.}, archive = {J_EAAI}, author = {Yongjian Fan and Bing Chen and Yunlong Zhao and Feng Hu and Chunyan Liu and Yang Li}, doi = {10.1016/j.engappai.2025.111717}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111717}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cooperative reconnaissance coverage for heterogeneous unmanned aerial vehicle swarm in unknown communication interference environments}, volume = {159}, year = {2025}, }
• (2025). A novel lightweight model integrating convolutional neural network and self-attention mechanism for corn seeds quality image recognition. EAAI, 159, 111716. (www)

  @article{ , abstract = {In the field of agriculture, accurate identification of seeds quality based on artificial intelligence algorithms is revolutionary for enhancing crop yields and qualities. The adoption of convolutional neural networks (CNNs) with scalability and self-attention mechanisms with robustness greatly improves the accuracy of seeds quality recognition. However, CNNs rely mainly on local receptive fields, making it challenging to capture long-range spatial dependencies, while self-attention mechanisms are computationally intensive and consume significant computational resources to achieve effective training. To address these limitations, a lightweight, efficient, and high-accuracy model architecture (denoted as LightMCS) is proposed, which consists of an efficient self-attention mechanism (ESA), partial convolution (PConv), and context broadcasting (CB). The LightMCS gradually extracts and refines features hierarchically. Notably, the PConv and ESA efficiently learn local-global representations, and the CB improves the model's capacity and generalization ability. Additionally, training acceleration techniques are employed to further enhance accuracy and reduce training time. The comparable results show that the proposed LightMCS demonstrates a higher classification accuracy (84.12 %) in the quality recognition task of corn seeds. More importantly, LightMCS has only 3.66 million parameters and a total training time of 243 min. The LightMCS achieves a balance between lightness and efficiency while improving classification accuracy. Thus, the proposed LightMCS provides an effective solution for the accurate identification of corn seeds quality, offering a promising tool in agricultural fields where accurate sorting of defective seeds scenarios is required.}, archive = {J_EAAI}, author = {Shuai Zhang and Shuqi Ma and Xin Luo and Hancheng Chai and Jie Zhu}, doi = {10.1016/j.engappai.2025.111716}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111716}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel lightweight model integrating convolutional neural network and self-attention mechanism for corn seeds quality image recognition}, volume = {159}, year = {2025}, }
• (2025). A new adaptive robust multi-objective optimization algorithm for dispatching of microgrids design. EAAI, 159, 111715. (www)

  @article{ , abstract = {Uncertainty widely exists in real-world applications. When uncertainty occurs, the robustness of solutions obtained by optimization is critical to the operation of the application. In the process of solving robust solutions, the balance between robustness and convergence is the key problem. Attaching undue importance to robustness leads to local optimality, while excessive convergence leads to loss of robust solutions. Motivated by above, a new adaptive robust multi-objective optimization algorithm is proposed. This algorithm proposes both a convergence-driven strategy and a robustness-driven strategy, and it adaptively selects between them by evaluating the population's evolutionary state. In the robust driven strategy, a new robustness metric is proposed to evaluate the robustness of individuals under disturbance. A penalty function based on the proposed metric guides the environmental selection process to enhance robustness. In the convergent driven strategy, the environment selection is mainly driven by the non-dominant relationship. A region robustness estimation strategy is proposed to evaluate the individuals in critical layer. Compared to four state of-the-art multi-objective algorithms on benchmark suits, the proposed algorithm performs better than its peers. In addition, the robust optimization problem of the dispatching of microgrids is constructed and the proposed algorithm is used for optimization of microgrids. The effectiveness of experimental results reveal that our methods are very promising in tackling the dispatching of microgrids problems.}, archive = {J_EAAI}, author = {Xue Feng and Lei Cui and Dexin Ren and Anqi Pan and Juchen Hong}, doi = {10.1016/j.engappai.2025.111715}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111715}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new adaptive robust multi-objective optimization algorithm for dispatching of microgrids design}, volume = {159}, year = {2025}, }
• (2025). Deep neuro-neutrosophic systems for detection of online sexism. EAAI, 159, 111714. (www)

  @article{ , abstract = {This paper proposes a deep neuro-neutrosophic system, a novel framework that combines DNN with the Neutrosophic Inference System (NIS) to improve prediction capacity. With the increasing usage of online social media platforms and anonymity, there is a multifold increase in sexism and hateful comments online. The proposed framework is utilized on the SemEval-2023 Task 10 dataset (with three subtasks) for the explainable detection of online sexism. The proposed approach for subtask A (to detect if a particular comment) is the pre-trained RoBERTa (Robustly Optimized BERT Approach) base model combined with a NIS based on Single-Valued Neutrosophic Set (SVNS) to classify the comments considering ambiguity. The Bidirectional Long Short-Term Memory (BiLSTM) model and suitable NIS based on refined neutrosophic sets were combined for subtasks B and C (to classify into the type of sexism). The proposed system provided higher accuracy than the traditional classification method and other existing models in all subtasks; it obtained 93.1% accuracy in subtask A, 91.55% in subtask B, and 97.98% in subtask C, compared to deep neural network (DNN) models that obtained 91.6% accuracy in subtask A, 78.74% in subtask B, and 78.77% in subtask C. The proposed framework can be used in various real-world scenarios that involve content moderation, including detecting online sexism spread through social media platforms. This research concludes that deep neuro-neutrosophic systems improve on existing deep learning models and are good at detecting online sexism and producing accurate results.}, archive = {J_EAAI}, author = {Ekta Biju and Ilanthenral Kandasamy and Samiksha Sai Paladugu and Ashwati Naduparambil Alakkal Sajit and Ratnala Baby Sowmya}, doi = {10.1016/j.engappai.2025.111714}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111714}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep neuro-neutrosophic systems for detection of online sexism}, volume = {159}, year = {2025}, }
• (2025). ETMA-net: Edge-and-threshold-guided multistage attention network for skin lesion image segmentation. EAAI, 159, 111712. (www)

  @article{ , abstract = {Accurate delineation of skin lesion regions in dermoscopic images is crucial for the early detection of melanoma, yet it presents challenges due to factors like irregular lesion shapes and indistinct boundaries. While conventional convolutional neural networks are commonly employed for segmenting lesion regions, they often struggle to capture global contextual features, leading to poor segmentation results. To address this challenge, this study proposes ETMA-Net (edge-and-threshold-guided multistage attention network), a lightweight parallel convolutional network architecture designed for automated skin lesion image segmentation in dermoscopic images. This framework effectively captures essential pixel dependencies and global contextual information by integrating a grouped multi-scale Hadamard attention module into the edge-threshold parallel structure, with minimal computational overhead. It also utilizes multistage aggregation bridges to merge high and low-level features, enhancing the model's understanding of both coarse-grained and fine-grained features, thereby improving segmentation accuracy. Through the repeated utilization of feature information at multiple levels and the continuous validation of extracted key features by the parallel convolutional network architecture, ETMA-Net demonstrates a robust understanding of skin lesion characteristics while maintaining computational efficiency. The study conducted comparative evaluations of ETMA-Net using the PH2, HAM10000 and ISIC (International Skin Imaging Collaboration) 2016, 2017, and 2018 datasets, yielding mIoU (mean Intersection over Union) test scores of 88.84 %, 84.41 %, 80.02 %, and 80.64 % respectively. With a parameter count of 0.63M and computational complexity of 0.38G, ETMA-Net outperformed existing state-of-the-art segmentation techniques. The model's efficacy was validated through comprehensive testing, including ablation studies, cross-dataset evaluations, and assessments on a polyp cross-medical segmentation dataset, showcasing ETMA-Net as a segmentation framework with superior performance, robust generalization capabilities, and high reliability.}, archive = {J_EAAI}, author = {Shijie Zhao and Fanshuai Meng and Liang Cai and Yuanshuai Chen}, doi = {10.1016/j.engappai.2025.111712}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111712}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ETMA-net: Edge-and-threshold-guided multistage attention network for skin lesion image segmentation}, volume = {159}, year = {2025}, }
• (2025). Frequency-prompt guided spectral–spatial transformer for hyperspectral image classification. EAAI, 159, 111711. (www)

  @article{ , abstract = {Convolutional neural networks (CNNs) and vision transformers (ViTs) have been applied to hyperspectral image (HSI) classification. However, CNNs have difficulty in capturing global context information, while ViTs lack the excellent ability of CNNs to capture local relationships in images. How to integrate CNNs and ViTs to design an effective network remains a challenge. In addition, most of the HSI classification methods are built upon the original image domain, failing to utilize the prior knowledge of the frequency domain. Although some studies explore the frequency-domain information, they cannot effectively achieve spatial-spectral-frequency information interactions and deep feature fusion. To solve the above challenges, we propose a frequency-prompt guided spectral–spatial transformer (FPGSST) for HSI classification. The backbone of FPGSST adopts a two-branch hierarchical network structure, where each branch extracts local and global spatial/spectral feature based on multi-stage convolutions and transformers. To guide the network to learn discriminative spatial and spectral features, we carefully design spatial-wise and spectral-wise frequency prompt modules. These prompts are structural which contain low-frequency priors and learnable high-frequency components. By embedding them into the self-attention mechanism, our model can focus on frequency-domain information in both spatial and spectral domains. We also design a spectral–spatial fusion transformer and a multi-stage fusion transformer to integrate comprehensive features. The resulting FPGSST achieves local–global information interaction and multi-stage spectral–spatial-frequency feature fusion by flexibly harnessing the variants of transformer architecture. Experimental results on several datasets demonstrate the superiority of FPGSST over the compared methods for HSI classification.}, archive = {J_EAAI}, author = {Tiecheng Song and Lei Mao and Longlong Zhang and Anyong Qin and Feng Yang and Chenqiang Gao}, doi = {10.1016/j.engappai.2025.111711}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111711}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Frequency-prompt guided spectral–spatial transformer for hyperspectral image classification}, volume = {159}, year = {2025}, }
• (2025). Graph-based framework for temporal human action recognition and segmentation in industrial context. EAAI, 159, 111710. (www)

  @article{ , abstract = {Industry 5.0 places human operators at the center of industrial processes. In this context, analyzing human movements has become crucial for ensuring operator safety and improving productivity. More specifically, an accurate system for action recognition and segmentation is essential to identify and break down each action an operator performs. These systems enable a range of applications, such as detecting fatigue and ergonomic risks, and providing targeted feedback to optimize workstations and task organization. However, existing action segmentation systems face challenges that limit their deployment in real industrial environments, including the complexity of industrial tasks and the similarity of assembly gestures. To address this issue, this work presents a system explicitly designed for industrial assembly environments that accurately segments and recognizes operator actions. It uses skeletal data and leverages graph-based representations to capture operator posture and model spatio-temporal movement patterns. It employs an encoder–decoder architecture, enhanced with an attention mechanism and an aggregation block, to effectively learn action dynamics. Additionally, it integrates decoder outputs at multiple temporal resolutions to precisely detect the start and end of actions. The proposed approach achieved competitive Mean Over Frame accuracies of 74.59% on the Industrial Human Action Recognition Dataset and 93.11% on the Human Action Multi-Modal Monitoring in Manufacturing dataset. Extensive experiments and ablation studies were conducted to provide deeper insights into its strengths. Visualizations are presented to illustrate the method’s handling of complex assembly video segmentation while also revealing specific cases for further improvement. The code is available in this repository . 1}, archive = {J_EAAI}, author = {Toufik Benmessabih and Rim Slama and Vincent Havard and David Baudry}, doi = {10.1016/j.engappai.2025.111710}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111710}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph-based framework for temporal human action recognition and segmentation in industrial context}, volume = {159}, year = {2025}, }
• (2025). Fuzzy decision-analytics based lithium-ion battery selection for maximizing the efficiency of electric vehicles. EAAI, 159, 111709. (www)

  @article{ , abstract = {The increasing reliance on fossil fuel-powered transportation significantly contributes to global warming and environmental change. Both industrial sectors and governmental bodies face challenges in addressing this issue through the adoption of green energy solutions. Electric vehicles present a more sustainable alternative due to their environmentally neutral characteristics and reduced operational costs. Batteries are integral to the functionality of electric vehicles, making the selection of an optimal battery essential. In this study, we conducted a comprehensive analysis to identify the most suitable lithium-ion batteries for the electric vehicle sector, employing fuzzy multi-criteria decision-making approaches (F-MCDM). We introduced a Complex Interval q-Rung Orthopair Fuzzy Set (CIVq-ROFS), which integrates three fuzzy sets: a complex fuzzy set, an interval-valued fuzzy set, and a q-rung orthopair fuzzy set. This integration yields more favorable outcomes compared to individual fuzzy sets. The proposed set is utilized for calculating attribute weights and ranking alternatives through the Criteria Importance Through Intercriteria Correlation (CRITIC) and Compromise Ranking of Alternatives from Distance to Ideal Solution (CRADIS) method. The algorithm was tested in a case study with five alternatives and six attributes, indicating a preference for the lithium nickel cobalt aluminum oxide battery for electric vehicles. The validity of the proposed model was established by comparing it with eight F-MCDM approaches, along with Spearman rank correlation and four sensitivity analyses to evaluate their effectiveness and robustness.}, archive = {J_EAAI}, author = {Thirumalai Nallasivan Parthasarathy and Samayan Narayanamoorthy and Dragan Pamucar and Daekook Kang}, doi = {10.1016/j.engappai.2025.111709}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111709}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzy decision-analytics based lithium-ion battery selection for maximizing the efficiency of electric vehicles}, volume = {159}, year = {2025}, }
• (2025). Fabric defect detection via explicit de-background. EAAI, 159, 111708. (www)

  @article{ , abstract = {Fabric defect detection under complex backgrounds faces challenges like high interference, false positives, and lack of robustness. To address these issues, a frequency-domain Explicit De-Background method is proposed to separate background from defects and enhance defect focus. The network uses a pixel-level De-background Layer to suppress noise and emphasize defects after extracting multi-scale features with Swin Transformer. This layer includes the Frequency-domain Background Extraction Module (F-BEM) and the Background Suppression Unit (BSU): F-BEM leverages Fourier amplitude to capture global background features, while BSU suppresses them to produce a difference map that accentuates defect regions. The De-Background Attention (DBA) module leverages the difference map as a weighting matrix to enhance spatial focus on defect features while minimizing background interference. To integrate multi-scale information, the Feature Cross-Shrinking Decoder (FCSD) progressively fuses adjacent layers via Cross Aggregation Nodes(CAN), ensuring semantic consistency, reducing redundancy, and mitigating information loss and gradient vanishing for precise defect segmentation. Our method enhances robustness and accuracy in complex backgrounds using explicit background separation and multi-stage feature processing. It surpasses state-of-the-art techniques on fabric defect datasets and shows good generalization in complex and transfer learning scenarios, providing a practical solution for industrial defect detection.}, archive = {J_EAAI}, author = {Yuntao Chen and Hao Liu and Jiuzhen Liang}, doi = {10.1016/j.engappai.2025.111708}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111708}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fabric defect detection via explicit de-background}, volume = {159}, year = {2025}, }
• (2025). Multisource aerodynamic data reconstruction method using an enhanced multifidelity neural network. EAAI, 159, 111707. (www)

  @article{ , abstract = {The acquisition of aircraft aerodynamic pressure distribution typically relies on wind tunnel tests or numerical simulations. However, discrepancies in accuracy and efficiency between multisource data present challenges for aerodynamic data reconstruction. In the present work, we propose a novel multifidelity architecture to enhance the model’s applicability to nonlinear inconsistency problems commonly encountered in transonic aerodynamic problems. By incorporating difference operations into the multifidelity neural network, the model can adaptively find suitable mapping relationships from potential low fidelity data. This method can reduce modeling errors when there is a trend inconsistencies between high and low fidelity data, which is a challenge that traditional multifidelity models struggle to address. To demonstrate the efficiency of our proposed ideas, we conducted multifidelity modeling on classic numerical examples and aerodynamic cases. Predictive results indicate that inconsistencies between multifidelity data can significantly affect traditional models, whereas the proposed multifidelity approach can enhance generalization performance. The reconstruction results of transonic pressure distribution for airfoils and the Office National d’Études et de Recherches Aérospatiales (ONERA) M6 wing indicate that the enhanced multifidelity model can effectively capture shock wave locations, thereby improving modeling accuracy. Analysis of the reconstruction results for pressure distribution indicates that the proposed method can reduce reconstruction error by over 30% compared to deep neural networks and multifidelity neural network. This method is also applicable for the data fusion of experimental and simulation data in various engineering problems.}, archive = {J_EAAI}, author = {Xu Wang and Huailu Li and Haitao Lin and Hui Tang and Weiwei Zhang}, doi = {10.1016/j.engappai.2025.111707}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111707}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multisource aerodynamic data reconstruction method using an enhanced multifidelity neural network}, volume = {159}, year = {2025}, }
• (2025). Super-resolution reconstruction of WorldView-3 multispectral satellite images based on generative adversarial networks. EAAI, 159, 111706. (www)

  @article{ , abstract = {Due to the limitations of optical sensors, satellites often sacrifice spatial resolution to obtain higher spectral resolution. Higher-resolution multispectral images contribute to accurate land feature identification, change monitoring, and environmental analysis. However, satellite remote sensing images encompass diverse terrains and surface features, which causes existing models for image reconstruction to frequently encounter challenges such as loss of high-frequency texture details and artifacts. To tackle this issue, this paper proposes a super-resolution method based on generative adversarial networks to enhance the spatial resolution of WorldView-3 satellite images from 1.2 m to 0.3 m across 8 spectral bands. Firstly, we introduce coordinate attention mechanisms in the generator to enhance the network’s ability to reconstruct spatial high-frequency details. In the discriminator, spectral normalization layers are integrated to enhance the stability of the adversarial network. Furthermore, a joint loss function is devised, comprising Charbonnier loss, spectral loss, and adversarial loss weighting. The Charbonnier loss effectively handles outliers, thereby mitigating artifact generation, while the spectral loss enriches the network’s spectral extraction capability, preventing spectral distortion in multispectral images. The adversarial loss incentivizes the generator to generate realistic images. The proposed method is evaluated using real WorldView-3 multispectral satellite images. Experimental results on two datasets demonstrate that the proposed method not only preserves more accurate high-frequency texture details and effectively reduces artifacts, but also maintains greater spectral fidelity than state-of-the-art approaches. It achieves optimal Learned Perceptual Image Patch Similarity (LPIPS) and Spectral Angle Mapper (SAM) scores, with improvements of 0.005 and 0.014 over the second-best method, respectively.}, archive = {J_EAAI}, author = {Mingqiang Guo and Hanbin Huang and Ran Liu and Chenglong Shao and Ying Huang}, doi = {10.1016/j.engappai.2025.111706}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111706}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Super-resolution reconstruction of WorldView-3 multispectral satellite images based on generative adversarial networks}, volume = {159}, year = {2025}, }
• (2025). An optimized plane detection-based topological metric for LiDAR simultaneous localization and mapping evaluation. EAAI, 159, 111704. (www)

  @article{ , abstract = {Mobile robot localization in complex indoor environments heavily depends on Simultaneous Localization and Mapping (SLAM). Traditional SLAM evaluation methods, which rely on Global Position System (GPS), laser trackers, or pre-calibrated anchors, are not effective for GPS-denied unknown environments. SLAM registration errors often cause noticeable plane noise and misalignment in point cloud maps, manifesting as translational and rotational inaccuracies. To address this, the paper proposes a novel topological SLAM evaluation strategy that estimates the rotational and translational error components of SLAM registration using the Plane Normal Entropy (PNE) and Co-Planar Variance (CPV). This approach enables the decoupling of those two error estimates, improving upon previous topological metrics. Binary and ternary correlation analyses validate the correlation between the proposed PNE and CPV with the traditional relative pose error (RPE). We introduce deep learning-based point cloud semantic segmentation combined with robust plane detection to enhance efficiency and accuracy of plane detect. The proposed evaluation metrics are tested on a well labeled dataset from local real garage in two tasks: short-range frame-to-frame registration error estimation and long-range loop closure detection. Extensive quantitative evaluation results demonstrate that the proposed topological metrics can assess SLAM errors solely based on point cloud map information in GPS-denied environments.}, archive = {J_EAAI}, author = {Zhenchao Ouyang and Yuchen Zhang and Huangcheng Jia and Dongyu Li and Qinglei Hu}, doi = {10.1016/j.engappai.2025.111704}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111704}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An optimized plane detection-based topological metric for LiDAR simultaneous localization and mapping evaluation}, volume = {159}, year = {2025}, }
• (2025). Analysis of the frequency of football players’ movement patterns by means of lorentzian self-organizing maps. EAAI, 159, 111703. (www)

  @article{ , abstract = {The recognition of frequent movement patterns of football players is of particular interest to assess, from a tactical perspective, the performance of both single individuals and the team as a whole. This paper introduces a novel variant of Self-Organizing Maps (SOMs), called Lorentzian Self-Organizing Maps (LSOMs). Whereas current research in sports science addresses a variety of relevant tasks, the dynamic and evolving nature of player movements patterns throughout a match remains underexplored. Starting from players’ velocity and acceleration data, acquired through a sensorized wearable vest during competitive matches, movement units are built, by clustering ranges of similar velocity-acceleration pairs. Subsequently, this units are grouped to form patterns of varying length, and the most frequent ones for five different players’ role are detected and extracted. Its comparison against other state-of-the-art clustering algorithms shown a consistent improvement. The testing has been performed by relying on 3 performance indexes, namely normalized mutual information, clustering accuracy and adjusted rand index. LSOMs resulted in the highest average scores on all the three metrics. Furthermore, in order to assess the tactical relevance of the proposed approach, a questionnaire was administered to 5 professional football trainers, about the goodness of the clustering proposed by LSOMs, against the other clustering techniques. The trainers feedback suggested that LSOMs can be profitably used to analyze players’ frequent movement patterns and that this kind of analysis can be helpful in finding non-trivial connections between the roles and the movements and, consequently, in gaining additional tactical insights.}, archive = {J_EAAI}, author = {Daniele Melloni and Andrea Zingoni}, doi = {10.1016/j.engappai.2025.111703}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111703}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Analysis of the frequency of football players’ movement patterns by means of lorentzian self-organizing maps}, volume = {159}, year = {2025}, }
• (2025). Underwater acoustic signal recognition system with multi-scale hybrid cepstral feature strategy and joint deep network. EAAI, 159, 111702. (www)

  @article{ , abstract = {In this paper, we propose a new underwater acoustic signal recognition system to address the recognition difficulties caused by the susceptibility of signals to complex noise interference in underwater acoustic environments. Specifically, the proposed system includes two stages: feature extraction and recognition. Feature extraction: a multi-scale hybrid cepstral feature strategy is proposed. It uses new singular spectrum decomposition to obtain multi-scale components and then extracts the Mel-frequency cepstral coefficients, inverse Mel-frequency cepstral coefficients, Gammatone frequency cepstral coefficients, and linear prediction cepstral coefficients of each component. After feature enhancement and selection, a novel multi-scale hybrid cepstral feature set is constructed. This feature set realizes the complementarity and enhancement of different cepstral features and effectively solves the problems of single feature expression and data redundancy. Recognition: a new joint deep network model is proposed. It adopts the unique design of one-dimensional convolutional neural network (1DCNN) and bidirectional gated recursive unit (BiGRU), which realizes the mutual complement of spatial information extracted by 1DCNN and dependent information captured by BiGRU and effectively improves the processing ability of the model for complex feature sets. In addition, the Kepler optimization algorithm and self-concern mechanism are introduced into the network, which solves the problem of selecting network parameters and improves the focus ability of the model on key features. By setting up multiple groups of comparison and ablation experiments, the recognition results of underwater acoustic data, including ship-radiated noise signals and marine biological signals, show that the recognition accuracy of the proposed system reaches 96.11 % and 98.67 %, respectively, which is better than all comparison methods. In addition, we further verified that the system still has high robustness under a low signal-to-noise ratio, which provides new ideas for research in the field of underwater acoustic signal recognition.}, archive = {J_EAAI}, author = {Hong Yang and Jinmei Li and Guohui Li and Chao Wang}, doi = {10.1016/j.engappai.2025.111702}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111702}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Underwater acoustic signal recognition system with multi-scale hybrid cepstral feature strategy and joint deep network}, volume = {159}, year = {2025}, }
• (2025). Traffic signal control based on deep reinforcement learning using state fusion and trend reward. EAAI, 159, 111701. (www)

  @article{ , abstract = {Deep reinforcement learning (DRL) methods are increasingly applied in traffic signal control. However, existing methods suffer from the high volatility of traffic flow and the challenges posed by sparse rewards in DRLs. To address these issues, this study proposes a traffic signal control model based on dueling double deep Q-learning. Additionally, a state fusion extraction method is designed for real-world, multisensor, intersection environments, allowing for a concise and accurate description of traffic dynamics through the incorporation of multidimensional spatial features as state inputs. To tackle the problem of reward sparsity in reinforcement learning, this study optimizes vehicle waiting time as the objective and designs a dynamic reward function based on trend of vehicle waiting time changes, weakening the influence of historical rewards to enable instantaneous strategy adjustments. Simulation results demonstrate that the proposed model notably outperforms other schemes with various states and reward design options, reducing cumulative waiting time by 19.89% and vehicle queues by 17.96% while increasing average speed by 5.17% at least.}, archive = {J_EAAI}, author = {Xiaoxue Tan and Yonghua Zhou and Xiangmeng Jiao}, doi = {10.1016/j.engappai.2025.111701}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111701}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Traffic signal control based on deep reinforcement learning using state fusion and trend reward}, volume = {159}, year = {2025}, }
• (2025). Artificial intelligence-driven methodological insights into wireless charging and sustainable energy management schemes: A survey. EAAI, 159, 111700. (www)

  @article{ , abstract = {Wireless rechargeable sensor networks (WRSNs) consist of sensors equipped with fast-rechargeable batteries and wireless power receivers, where energy storage plays a crucial role in bridging the intermittent nature of wireless power transfer (WPT) with the continuous energy demands of the sensors. The mobile charger (MC) enables network functionality through wireless power transmission, effectively addressing the charging requirements of these networks. This survey seeks to illuminate various aspects of wireless charging schemes regarding management strategies and technical methodologies, particularly in the context of WPT and energy consumption management from a methodological standpoint. By presenting a comprehensive, methodology-driven analysis of wireless charging schemes for WRSNs, this survey makes a significant contribution to the field. Furthermore, this survey aims to provide artificial intelligence (AI)-driven methodological insights into various facets of wireless charging schemes within the framework of sustainable energy management. It examines innovative management strategies and technical methodologies that utilize artificial intelligence to enhance the efficiency of WPT and energy consumption. Each wireless charging scheme is scrutinized through three distinct lenses: technical methodology, charge methodology, and evaluation methodology, considering the specific design parameters relevant to each approach. The technical methodology explores the capabilities and limitations of each scheme, detail the analytical parameters involved, and define the operational objectives. The charge methodology focuses on charging characteristics, including management patterns, operational modes, and categorization of charging processes. The evaluation methodology provides a holistic overview of how these schemes are assessed, encompassing evaluation metrics, simulation configurations, and the simulation environment. To better understand the landscape of these methodologies, this survey categorizes the reviewed schemes into five distinct classes: heuristic-based, metaheuristic-based, fuzzy-based, hybrid approaches that integrate both metaheuristic and fuzzy logic techniques, and reinforcement learning-based techniques. This categorization is grounded in the aforementioned three innovative aspects of technical, charge, and evaluation methodologies. Importantly, this survey aims to offer an exhaustive overview of the capabilities, limitations, and emerging research directions concerning these schemes from a methodology-driven perspective, particularly in the realm of WRSNs. Additionally, it seeks to pave the way for future research by identifying new avenues for exploration, filling the gap left by the lack of comprehensive surveys in this crucial field.}, archive = {J_EAAI}, author = {Fakhrosadat Fanian and Marjan Kuchaki Rafsanjani}, doi = {10.1016/j.engappai.2025.111700}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111700}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence-driven methodological insights into wireless charging and sustainable energy management schemes: A survey}, volume = {159}, year = {2025}, }
• (2025). Physics-informed neural networks for linear elastic fracture mechanics: Application of assessing stress intensity factor and transverse stress. EAAI, 159, 111699. (www)

  @article{ , abstract = {Structural health monitoring (SHM) is essential for preserving infrastructure integrity, ensuring public safety, and extending the lifespan of structures. Real-time monitoring and assessment of structural health allow for the early detection of potential damage or deterioration, preventing catastrophic failures. However, current SHM methods often depend on sensor data and traditional physical models, which are limited when faced with complex environmental changes and incomplete data. Particularly in quantifying fracture behavior, existing methods struggle to accurately estimate key fracture parameters and fail to capture the microscopic changes and complex mechanical behaviors during crack propagation. A generalized framework based on Physics-informed neural networks (PINN) is proposed to identify the key fracture parameters, such as stress intensity factor (SIF) and transverse stress ( T -stress), at the crack tip. By integrating the fracture-related formulation and Airy stress function, the key fracture parameters around the crack tip can be successfully identified from sparse displacement field data. Through the verification of several typical fracture cases, the proposed framework is demonstrated to accurately predict the key fracture parameters and physical fields around the crack tip. It is anticipated that the proposed framework could be incorporated into comprehensive SHM systems, improving predictive capabilities for crack growth, fatigue analysis, and damage detection, thereby enabling more efficient maintenance strategies and enhancing safety for critical infrastructure.}, archive = {J_EAAI}, author = {Xiang Guo and Zhigong Song}, doi = {10.1016/j.engappai.2025.111699}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111699}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed neural networks for linear elastic fracture mechanics: Application of assessing stress intensity factor and transverse stress}, volume = {159}, year = {2025}, }
• (2025). A model predictive trajectory tracking control strategy for heavy-duty unmanned tracked vehicle using deep koopman operator. EAAI, 159, 111698. (www)

  @article{ , abstract = {Among the numerous technologies for the heavy-duty unmanned tracked vehicle (HDUTV), trajectory tracking is the key function to support the maneuverability. Unlike Ackermann steering vehicles, HDUTVs are easily affected by disturbances during the steering process, leading to different steering characteristics. The variable steering characteristics pose challenges for precise tracking control. Motivated by this challenge, a high accuracy model predictive trajectory tracking method is proposed to improve the tracking performance of HDUTVs. First, a deep Koopman operator-based tracked vehicle model is established. The proposed learning-based model provides an accurate description of the complex nonlinear dynamics of HDUTVs while maintaining the model linearity. Utilizing the model, the real-time performance of the trajectory tracking process is guaranteed. Second, a trajectory tracking control strategy is established considering the steering characteristic of the HDUTV to improve the tracking performance. Third, the deep Koopman operator-based model is integrated into the model predictive control framework to enhance predictive accuracy while ensuring the real-time performance of the trajectory tracking controller. Finally, the proposed method is validated through simulations and experiments with a full-sized HDUTV. Experiment results indicate that the proposed model enhances predictive ability for vehicle states, with a 59.51 % improvement in the accuracy of the sideslip angle. And the proposed trajectory tracking strategy improves the tracking accuracy by 57.93 %.}, archive = {J_EAAI}, author = {Yinchu Zuo and Chao Yang and Shengfei Li and Weida Wang and Changle Xiang and Tianqi Qie}, doi = {10.1016/j.engappai.2025.111698}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111698}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A model predictive trajectory tracking control strategy for heavy-duty unmanned tracked vehicle using deep koopman operator}, volume = {159}, year = {2025}, }
• (2025). Data driven-based thesis defense scheduling system: A theoretical framework and empirical development. EAAI, 159, 111697. (www)

  @article{ , abstract = {The thesis defense, essential for students to obtain their degree during the undergraduate program, is a formal event for presenting students’ theses to committees. This paper studies the thesis defense scheduling problem, considering equality and efficiency after the committees are determined. A data-driven method calculates the matching similarity degree between students and committees. A nonlinear mathematical model is proposed to formulate the studied problem. A linearized model is introduced to make the model solvable by mixed-integer solvers. A mathematical formulation-based genetic algorithm is proposed to solve large-scale problems. This paper developed a decision-support system for solving the studied problem by integrating the proposed solution method. Various experiments were conducted, and the experimental results showed that the proposed method could effectively solve the studied problem for large-scale instances. Finally, a web-based system is presented to show the capability of applying the proposed method.}, archive = {J_EAAI}, author = {Yanjie Zhou and Zhicheng Zhang and Ningning Song and Jincan Zhang}, doi = {10.1016/j.engappai.2025.111697}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111697}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data driven-based thesis defense scheduling system: A theoretical framework and empirical development}, volume = {159}, year = {2025}, }
• (2025). Evolutionary dynamic multiobjective optimization using a bayesian vector autoregression prediction model. EAAI, 159, 111696. (www)

  @article{ , abstract = {Dynamic multiobjective optimization problems (DMOPs), inherently involving the simultaneous optimization of conflicting objectives under time-varying environments, exhibit ubiquitous presence in real-world applications spanning diverse engineering disciplines. A prevalent limitation in existing dynamic multiobjective optimization evolutionary algorithms (DMOEAs) lies in inadequate utilization of historical information and neglect of interdependencies among decision variables, which frequently induces suboptimal initial population predictions deviating from true Pareto optimal sets. To mitigate these limitations, we propose MOEA/D-BVAR, a novel DMOEA framework incorporating a Bayesian vector autoregressive (BVAR) model that conceptualizes solution dynamics through holistic vector forecasting rather than isolated variable-specific prediction. The algorithm initially clusters variables via mutual information correlation analysis, subsequently constructing BVAR models for each cluster to project their evolutionary trajectories. Independently varying variables are rapidly predicted through differential forecasting models. A multivariate interaction optimization mechanism enhances search efficiency. Comprehensive empirical evaluations on 14 benchmark suites compare MOEA/D-BVAR against six state-of-the-art DMOEAs developed over the past five years. Statistical analysis of experimental outcomes demonstrates the proposed algorithm’s superior competitiveness in handling complex DMOPs.}, archive = {J_EAAI}, author = {Kai Gao and Wenxiang Jiang and Lihong Xu}, doi = {10.1016/j.engappai.2025.111696}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111696}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evolutionary dynamic multiobjective optimization using a bayesian vector autoregression prediction model}, volume = {159}, year = {2025}, }
• (2025). Post-earthquake residual track irregularities prediction using wavelet-packet-transformation-powered domain adaptive neural network. EAAI, 159, 111695. (www)

  @article{ , abstract = {Prompt identification of residual track irregularities resulting from seismic events is critical for timely post-earthquake rescue operations and ongoing activities. To tackle the computational burden of purely physics-based surrogate models and alleviate the overfitting issue faced by purely data-driven models, we propose a novel knowledge-guided data-driven model called the Wavelet-Packet-Transformation-powered Domain Adaptive Neural Network (WPT-DANN). The adoption of the Prior-Domain-Knowledge-adaptive ensemble learning strategy enables model training without relying on time-history response data, thus lowering the barrier to incorporating Prior Domain Knowledge (PDK). The Response-Spectra-Method (RSM)-based PDK, capable of rapidly generating RSM-inferred irregularities, not only fulfill essential functions within the ensemble strategy to alleviate overfitting issue but also contribute an extra limited penalty in the loss function to prevent unreasonable predictions. The Wavelet-Packet-Transform (WPT), with its proficiency in lossless time-frequency decomposition and reconstruction, executes tensor shape adjustments in the model, optimizing training by increasing dimensionality while reducing time complexity. Analyzing WPT-DANN models with various PDKs, trained on reliable samples validated through shaking-table tests, demonstrates that integrating PDKs reduces sample size to attain comparable accuracy levels. Comparative assessments against classic neural network-based models highlight WPT-DANN's superiority, with a minimum 79.1% and 4.6 % advantage in training efficiency and accuracy of residual irregularities prediction, respectively. Notably, the application of WPT to classic neural network-based models presents a promising avenue for performance enhancement, particularly in reducing training time. This study contributes to the development of related research by optimizing network architecture and integrating prior information into seismic structural deformation prediction tasks.}, archive = {J_EAAI}, author = {Yuntai Zhang and Wang-Ji Yan and Li-Zhong Jiang and Ka-Veng Yuen and Wangbao Zhou}, doi = {10.1016/j.engappai.2025.111695}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111695}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Post-earthquake residual track irregularities prediction using wavelet-packet-transformation-powered domain adaptive neural network}, volume = {159}, year = {2025}, }
• (2025). Density cluster-based feature selection: An information theory approach. EAAI, 159, 111694. (www)

  @article{ , abstract = {Feature selection plays a crucial role in data mining and machine learning. However, evident challenges exist: (1) current methods cannot autonomously identify the optimal feature set, requiring manual parameter adjustment based on the learning algorithm; (2) heuristic methods, which are widely used, often struggle to ensure the maximization of the objective function. To address these challenges, this paper proposes a density cluster-based feature selection (DCFS) method leveraging information theory, which involves the application of artificial intelligence (AI) in the clustering process. First, a novel initial feature selection method that maximizes feature-relevance and feature-difference is introduced to automate the selection of an initial feature subset. Second, a new density-centric automatic clustering (DAC) algorithm, an AI-based clustering approach, is proposed. This algorithm synthesizes non-parametric density estimation, decision graph-based density center selection strategies, and adaptive domain search clustering methods to enhance the precision and robustness of clustering outcomes. Third, a feature space selection method based on maximizing feature relevance is established to construct a comprehensive feature subspace. This feature space selection method converts the maximization of the objective function into an automated density clustering process, facilitating the automatic selection of the most optimal features. Extensive experiments conducted across 14 datasets have demonstrated the superior performance of the proposed DCFS in terms of effectiveness and robustness. To the best of our knowledge, this paper is the first work attempting automatic feature selection through clustering, thus pushing the frontier of feature selection algorithm development.}, archive = {J_EAAI}, author = {Jingya Dong and Shuai Tao and Chunhe Song and Peiming Ning and Tao Zhang}, doi = {10.1016/j.engappai.2025.111694}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111694}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Density cluster-based feature selection: An information theory approach}, volume = {159}, year = {2025}, }
• (2025). Knowledge alignment based on category-attributes for blood cell classification. EAAI, 159, 111692. (www)

  @article{ , abstract = {Blood cell image classification is crucial for diagnosing blood diseases. In this field, deep learning models have been extensively applied due to their high accuracy and efficiency. While the interpretability within models inherently possess expanded deployment potential, numerous interpretable approaches remain obscure and difficult for human comprehension. More detrimentally, interpretable models requiring domain-knowledge grounding face critical deployment barriers, as their reliance on costly expert-annotated datasets fundamentally restricts scalability across application domains. To address this issue, we introduce a novel knowledge alignment interpretability mechanism and the implementation, Knowledge-Aligned Attribute Network (KAAN), which leverages category-attributes alignment and cytomorphology. The concrete implementation, KAAN approach allows models of arbitrary architectures to learn cytomorphological knowledge through category-attribute labels, identify the true single-value attributes of images from multi-value category-attribute labels, and leverage these attributes for effective classification. Users can leverage reality-aligned attribute outputs to assess the credibility of a model’s classification results, while freely manipulating these attributes to fulfill practical engineering requirements such as false sample correction or manual hematological cell morphology training program. Meanwhile, knowledge alignment implementation could preserve the model’s original performance. More critically, the category-attribute labels are directly derived from existing knowledge texts, significantly reducing annotation costs during cross-domain deployment of KAAN approach. Comprehensive quantitative and qualitative experiments demonstrate that KAAN outperforms current state-of-the-art counterparts in both attribute recognition and classification accuracy, while exhibiting superior knowledge alignment characteristics. This novel interpretability mechanism not only demonstrates strong practical value but also fills the gap in interpretable models within the field of blood cell classification.}, archive = {J_EAAI}, author = {Qianhang Guo and Xingyu Zhao and Yaodong Liang and Xiexin Liao and Jie Li and Jianping Guo and Qingmao Zhang and Kunyuan Xu and Qiongxiong Ma}, doi = {10.1016/j.engappai.2025.111692}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111692}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Knowledge alignment based on category-attributes for blood cell classification}, volume = {159}, year = {2025}, }
• (2025). Automotive fuse & relay box plug-in modules assembly correctness detection system based on machine vision. EAAI, 159, 111691. (www)

  @article{ , abstract = {The automotive fuse and relay box is vital for electrical safety and reliability, demanding stringent quality control before leaving the factory. However, existing methods face limitations such as light interference, inability to detect non-fuse plug-in modules, lack of worker-friendly interfaces, insufficient data recording features, and a lack of comparative diagnostic capabilities for detection results. To address these issues, an artificial intelligence (AI)-powered automotive fuse and relay box assembly correctness detection system based on machine vision is proposed. This system incorporates a closed image acquisition setup, advanced machine vision techniques, and My Structured Query Language (MySQL) database operations for efficient data management. A comprehensive detection rule-setting subsystem, developed with Python Qt 5 (PyQt5) graphical user interface (GUI), integrates classification detection, similarity detection, color detection, and text recognition, allowing users to easily create detection rules. Additionally, a PyQt5-based template selection subsystem further streamlines template identification for various scenarios. The detection system combines these four methods with an object detection method for real-time, accurate assembly verification. The core You Only Look Once version 11 extra-large (YOLOx) model provides fast and precise localization, while supplementary modules—Residual Neural Network with 18 layers for classification detection, Siamese network-based similarity detection, binary character recognition, and color detection—work synergistically to enhance detection robustness and accuracy. The system achieves an average detection time of 0.141 s per module for correct assemblies and 1.398 s for faulty assemblies. Demonstrating 99.9 % accuracy, high adaptability, and efficient detection, the system is highly suitable for large-scale, real-world production environments.}, archive = {J_EAAI}, author = {ZhengWei Gong and Jun Song and Ping Zhang}, doi = {10.1016/j.engappai.2025.111691}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111691}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automotive fuse & relay box plug-in modules assembly correctness detection system based on machine vision}, volume = {159}, year = {2025}, }
• (2025). A novel data-driven sample augmentation method using interpretable space and adversarial mechanism for dioxin risk warning. EAAI, 159, 111690. (www)

  @article{ , abstract = {Dioxin risk warning plays a crucial role in ensuring the long-term sustainability of municipal solid waste incineration (MSWI) plants. However, dioxin, which is difficult to detect in real time, poses significant challenges in developing data-driven models. One widely employed solution to overcoming the limitations of data-driven modeling is the virtual sample generation (VSG) method. Nevertheless, incomplete generative procedures and unreliable randomness produce difficulties for researchers utilizing VSG. Addressing these issues, this article proposes a novel data augmentation method based on interpretable space and adversarial mechanism. Initially, two-dimensional spatial latent samples are derived using a variational autoencoder to aid visualization. Subsequently, two-dimensional virtual latent samples are generated via multi-angle rotations in diverse circular spaces to enhance interpretability. Next, these two-dimensional virtual latent samples are input into the generative adversarial network to produce candidate virtual samples in their original dimensions. Finally, co-training is employed to meticulously select high-quality candidate virtual samples. Experimental verification of the proposed VSG method utilizes real dioxin datasets of the MSWI process, demonstrating its effectiveness.}, archive = {J_EAAI}, author = {Canlin Cui and Jian Tang and Junfei Qiao and Heng Xia}, doi = {10.1016/j.engappai.2025.111690}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111690}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel data-driven sample augmentation method using interpretable space and adversarial mechanism for dioxin risk warning}, volume = {159}, year = {2025}, }
• (2025). A fuzzy three-criteria optimization-based currency trading system with adaptive criteria shapes and money management. EAAI, 159, 111689. (www)

  @article{ , abstract = {The paper is focused on the development of algorithmic trading modeling — the modern branch of financial engineering. The concepts of negative overfitting risk and irregular shape of fuzzy criteria in currency trading modeling have been introduced. The first concept is based on observations, according to which deep drawbacks of the profit curve at the optimization stage can lead to negative overfitting (loss of profit during the testing period), as well as high local peaks on this profit curve. Thus, we can say that any significant irregularity of the profit curve can lead to losses due to the emergence of the effect of overfitting. To implement this concept, two relevant indicators of technical analysis and, based on them, risk criteria were developed. The second concept is based on the informal but logically justified assumption that irregular optimized forms of used membership functions may better reflect current market conditions. To implement these concepts, two fuzzy three-criteria optimization-based currency trading models comprised the profitability, reliability and risk local criteria with adaptive criteria shapes and money management were developed. These models were confirmed by the results obtained by modeling intraday trading over the past five years and four months on a 1-hour timeframe and the six most traded currency pairs. A comparison of these results with the results obtained by other authors confirmed their significant superiority and thus confirmed the correctness of the concepts introduced.}, archive = {J_EAAI}, author = {Krzysztof Kaczmarek and Pavel Sevastjanov and Ludmila Dymova and Adam Kulawik and Leszek Rutkowski}, doi = {10.1016/j.engappai.2025.111689}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111689}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A fuzzy three-criteria optimization-based currency trading system with adaptive criteria shapes and money management}, volume = {159}, year = {2025}, }
• (2025). Detection of error in static and dynamic visual stimulation via electroencephalogram and eye-tracking systems. EAAI, 159, 111688. (www)

  @article{ , abstract = {Human responses such as electroencephalogram (EEG), eye-tracking, and heart rate have been studied for error detection during visual stimulation, often in controlled settings with single-target fixation. This study delves into constructing machine learning (ML) models for binary error classification across diverse visual stimulation conditions, including static, dynamic, and single or multiple targets, using EEG and eye-tracking data. When constructing these models, using gaze fixation data for epoch extraction can enhance the ability to extract salient, stimulus-induced responses from EEG and eye-tracking data. These features can be strongly associated with changes in visual stimulation. Among 30 ML models tested, the best-performing ML models built on a personalized approach consistently achieved over 90 % accuracy across conditions. For feature importance, we integrate a repetition approach with the Boruta SHapley Additive exPlanations (BorutaSHAP) algorithm to enhance the legitimacy of key feature selection. Feature analysis revealed distinct patterns, e.g., eye-tracking features like log energy entropy being particularly prominent under dynamic conditions, EEG features from the delta, and theta bands being significant across all conditions. Interestingly, an increase in the number of visual targets led to a reduction in the importance of EEG features, especially during dynamic stimulations. These insights have the potential to enhance the ML models through tailored feature selection. While this study acknowledges certain limitations concerning real-time applicability, generalizability, etc, the novelty of our models p[presents opportunities for various applications in human-computer/robot interaction (HCI/HRI), monitoring systems, rehabilitation systems, assistive technologies for individuals with limited mobility and driving, etc.}, archive = {J_EAAI}, author = {Hyowon Lee and Ning Jiang and Siby Samuel}, doi = {10.1016/j.engappai.2025.111688}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111688}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detection of error in static and dynamic visual stimulation via electroencephalogram and eye-tracking systems}, volume = {159}, year = {2025}, }
• (2025). Multitask gated interactive network for automatic international classification of diseases coding with dual denoising mechanism. EAAI, 159, 111687. (www)

  @article{ , abstract = {Recent advancements in deep learning have significantly impacted the field of medicine and healthcare, offering promising solutions for automatic International Classification of Diseases (ICD) coding. ICD coding plays a vital role in mapping clinical records to the corresponding ICD codes, providing doctors with faster and more accurate diagnostic support. However, traditional manual coding is inefficient and error-prone, while deep learning models face challenges such as noisy and lengthy clinical text, complex connections among labels, and imbalanced label distributions. Therefore, we propose a novel Multitask Gated Interactive Network (MGIN) with a dual denoising mechanism. Specifically, the gated recalibration module is designed to combine cascaded convolutions and gated units to filter irrelevant text and extract the most ICD-relevant fragments from lengthy clinical notes. The multitask interactive sharing module aims to jointly train ICD and Clinical Classification Software (CCS) codes for sharing parameter information and incorporates a co-occurrence frequency matrix to capture the complex associations between medical codes. Additionally, a dual denoising module simultaneously combats class imbalance by emphasizing rare samples through adaptive loss weighting while further suppressing noisy samples by dynamically discard high-loss training instances. Finally, comprehensive experiments on real-world datasets have demonstrated that our proposed model MGIN outperforms multiple competitive baselines.}, archive = {J_EAAI}, author = {Xiaobo Li and Yijia Zhang and Xiaodi Hou and Shilong Wang and Wen Qu}, doi = {10.1016/j.engappai.2025.111687}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111687}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multitask gated interactive network for automatic international classification of diseases coding with dual denoising mechanism}, volume = {159}, year = {2025}, }
• (2025). Research on enhancing model performance by merging with korean language models. EAAI, 159, 111686. (www)

  @article{ , abstract = {This study proposes a novel and straightforward approach to enhancing the capabilities of top-ranking Large Language Models (LLMs) from the Open LLM Leaderboard, leveraging the Drop and Rescale (DARE) technique. DARE facilitates efficient model merging by minimizing delta parameter redundancy from fine-tuned models. We integrate a top-performing multilingual LLM with a specialized Korean language model using DARE. The merged model is evaluated on six benchmark tasks and a multi-turn question set (MT-Bench), focusing on reasoning capabilities. Results show that incorporating the Korean language model achieves a significant performance improvement of 1.69% on average across the six benchmark tasks, and notably demonstrates over 20% higher performance on Grade School Math 8K (GSM8K), which requires complex reasoning skills. This suggests that the inherent complexity and rich linguistic features of the Korean language contribute to enhancing LLM reasoning abilities. Moreover, the model exhibits superior performance on MT-Bench, demonstrating its effectiveness in real-world reasoning tasks. This study highlights DARE’s potential as an effective method for integrating specialized language models, demonstrating the ability to unlock existing language models for advanced tasks. 1}, archive = {J_EAAI}, author = {Taewan Cho and Rina Kim and Andrew Jaeyong Choi}, doi = {10.1016/j.engappai.2025.111686}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111686}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on enhancing model performance by merging with korean language models}, volume = {159}, year = {2025}, }
• (2025). Multi-attribute decision-making with extended VlseKriterijuska optimizacija i komoromisno resenje method based on generalized fermatean probabilistic hesitant fuzzy distance measure and cross-entropy. EAAI, 159, 111684. (www)

  @article{ , abstract = {Amidst the escalating complexity and ambiguity inherent in real-world decision-making within uncertain environments, the Fermatean probabilistic hesitant fuzzy set (FPHFS) emerges as a significant extension of the Fermatean fuzzy set (FFS), gaining momentum in the realm of multi-attribute decision-making (MADM). This paper proposes a novel MADM approach that integrates the distance measure and cross-entropy of FPHFSs with the VlseKriterijuska Optimizacija I Komoromisno Resenje (VIKOR) method. This approach is designed to concurrently account for the objective weighting of attributes, the stochastic nature of evaluation data, and the psychological preferences of decision-maker (DM). Initially, leveraging the definitions of distance measures and entropy for FPHFSs, various forms of distance and cross-entropy measures are proposed. Subsequently, by integrating the proposed generalized Fermatean probabilistic hesitant fuzzy cross-entropy with the maximum deviation method, two models for determining the objective weights of attributes are developed. To intuitively capture and process DMs' preference information, a linear preference function within the decision-making process is constructed using the introduced generalized Fermatean probabilistic hesitant fuzzy distance. Ultimately, the preference function-based Fermatean probabilistic hesitant fuzzy VIKOR (FPHF-VIKOR) approach is established. By applying it to a numerical example of supply chain finance risk assessment, the decision-making procedure of the method is thoroughly demonstrated. The excellent performance of the distance and cross-entropy measures within this model is examined through parameter sensitivity analysis. Additionally, a real-world application concerning financing enterprise selection highlights the practicality and scalability of the proposed method. Comparative analysis with existing MADM methods and VIKOR variants further confirms the scientific feasibility and effectiveness of our approach.}, archive = {J_EAAI}, author = {Chuanyang Ruan and Lin Yan and Shouzhen Zeng}, doi = {10.1016/j.engappai.2025.111684}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111684}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-attribute decision-making with extended VlseKriterijuska optimizacija i komoromisno resenje method based on generalized fermatean probabilistic hesitant fuzzy distance measure and cross-entropy}, volume = {159}, year = {2025}, }
• (2025). Deep learning in forensic analysis: Optical coherence tomography image classification in methamphetamine detection. EAAI, 159, 111682. (www)

  @article{ , abstract = {Detecting drug addiction in forensic science traditionally relies on expensive and time-consuming laboratory tests. This study proposes a rapid, non-invasive approach that uses optical coherence tomography images combined with deep learning techniques to identify methamphetamine users. A novel convolutional neural network was developed, incorporating depthwise and pointwise convolutions, patchify-based downsampling, and inception blocks to improve feature extraction and classification accuracy. To further enhance model performance, we introduced a grid-based deep feature engineering model that extracts and selects discriminative features using iterative neighborhood component analysis. The proposed model achieved 91.02 % accuracy, surpassing the 88.57 % accuracy of Mobile Network version 2 on the same dataset. By integrating the grid-based feature engineering model, classification accuracy was further improved to 93.27 %, demonstrating a significant enhancement over traditional deep learning approaches. The dataset consisted of 2172 optical coherence tomography images collected from 54 methamphetamine users and 60 control subjects, ensuring a diverse and representative sample. This research marks the first application of optical coherence tomography imaging in drug addiction detection, bridging biomedical imaging and forensic science. By employing gradient-weighted class activation mapping visualization, we identified key retinal features that distinguish methamphetamine users from non-users, thereby making the model more interpretable and clinically relevant. Given its high accuracy, lightweight architecture, and non-invasive nature, the proposed method offers a promising forensic tool for rapid, artificial intelligence-driven drug addiction screening with potential real-world applicability in forensic investigations and healthcare.}, archive = {J_EAAI}, author = {Nilifer Gurbuzer and Alev Lazoglu Ozkaya and Elif Topdagi Yaylali and Elif Ozcan Tozoglu and Mehmet Baygin and Burak Tasci and Sengul Dogan and Turker Tuncer}, doi = {10.1016/j.engappai.2025.111682}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111682}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning in forensic analysis: Optical coherence tomography image classification in methamphetamine detection}, volume = {159}, year = {2025}, }
• (2025). Transformer-based auto-encoder with combined multi-head-attention for industrial soft-sensor modeling. EAAI, 159, 111681. (www)

  @article{ , abstract = {Soft-sensor modeling is common in industrial production, but high data dimensionality, a lack of labeled features, and inadequate methods complicate extracting nonlinear feature representations. This paper proposes a Transformer-based auto-encoder with a combined multi-head-attention approach (TAE-CMHA) for soft-sensor modeling, which offers advantages for nonlinear feature representation. It introduces a combined multi-head-attention mechanism (CMHA) that improves feature-extraction accuracy and robustness. The Transformer's global feature extraction capabilities are leveraged in the auto-encoder for better nonlinear feature extraction. Additionally, label information optimizes the auto-encoder's reconstruction loss function which improves feature acquisition for predicting target outputs. Compared to supervised methods, the unsupervised auto-encoder uses abundant unlabeled industrial data to improve generalizability. Experiments were conducted on the industrial steam flow and debutanizer column datasets. The results show that the mean squared error (MSE) of the proposed method reaches a minimum of 0.00297 and the coefficient of determination (R 2 ) is 0.881 in debutanizer column datasets, which shows the advantages of the model.}, archive = {J_EAAI}, author = {Yanhong Li and Shiwei Gao and Wenfeng Zhao}, doi = {10.1016/j.engappai.2025.111681}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111681}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer-based auto-encoder with combined multi-head-attention for industrial soft-sensor modeling}, volume = {159}, year = {2025}, }
• (2025). Can large language models understand depth in monocular images without prior vision knowledge?. EAAI, 159, 111680. (www)

  @article{ , abstract = {Monocular Depth Estimation (MDE) is vital for robotics and autonomous driving, yet it usually relies on task-specific vision backbones and large labeled datasets. This paper explores that the large language models can effectively understand depth with minimal supervision, using efficient resource utilization and a consistent neural network architecture. We propose Large Language Model-Monocular Depth Estimation (LLM-MDE), a unified framework that treats depth estimation as a language-reasoning problem. A frozen Vision Transformer first encodes an image; cross modal reprogramming then aligns visual tokens with a small set of learned text prototypes, and an adaptive depth prompt generator turns simple image statistics into natural-language cues that guide a pretrained large language model. Only lightweight Low-Rank Adaptation (LoRA) adaptors are tuned, keeping the computational cost low. Comprehensive experiments on real-world datasets confirm the effectiveness and superiority of LLM-MDE, which excels in few/zero-shot tasks while minimizing resource use. These show that large language models, when guided by minimal visual reprogramming and adaptive prompts, can reason about geometric structure without prior vision-specific training, opening a practical path to low-resource MDE on edge devices. Code are publicly available at: https://github.com/651784766/LLM-MDE .}, archive = {J_EAAI}, author = {Zhongyi Xia and Tianzhao Wu and Boqi Wu}, doi = {10.1016/j.engappai.2025.111680}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111680}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Can large language models understand depth in monocular images without prior vision knowledge?}, volume = {159}, year = {2025}, }
• (2025). Integrating physics-based simulations, machine learning, and bayesian inference for accurate detection and metrology of elongated nanoscale analytes using high-frequency capacitance spectroscopy. EAAI, 159, 111679. (www)

  @article{ , abstract = {Elongated analytes are simple general-purpose model systems for nucleic acid strands, bacteriophages, nanoplastic fibers, nanotubes, nanorods, etc., and are characterized by numerous unknowns (e.g., material composition, length, orientation, etc.) that are difficult to measure accurately in real-time. This paper aims to advance the state-of-the-art of nanoscale sensing and metrology for these elongated, high aspect-ratio analytes, utilizing advanced data analysis methods specially developed for high-frequency capacitance spectroscopy measurements at nanoelectrode arrays. A model-based approach is proposed, integrating: (1) advanced supervised learning algorithms trained on an extensively augmented dataset derived from accurate physics-based numerical simulations; (2) a Markov Chain Monte Carlo (MCMC) Bayesian estimation framework for the parameters extraction task. The proposed algorithm achieves substantial speed enhancements while maintaining high accuracy, even at the resolution limits of the sensor. The test case is developed for multispectral capacitance images of 200-1000 nanometers long nanorods captured with an advanced 256 × 256 pixels nanocapacitor array. The proposed approach minimizes the need for time-consuming, physics-based simulations in sensor behavior prediction and Bayesian inference iterations. It is applicable to other elongated nanoscale analytes whose state is defined by many parameters. As a result, a robust and scalable solution for efficient and precise metrology of elongated analytes is established for high parallelism and high throughput nanocapacitor array sensor applications.}, archive = {J_EAAI}, author = {Ehsan Khodadadian and Daniele Goldoni and Jacopo Nicolini and Amirreza Khodadadian and Clemens Heitzinger and Luca Selmi}, doi = {10.1016/j.engappai.2025.111679}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111679}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating physics-based simulations, machine learning, and bayesian inference for accurate detection and metrology of elongated nanoscale analytes using high-frequency capacitance spectroscopy}, volume = {159}, year = {2025}, }
• (2025). Comprehensive analysis and machine learning-based solutions for drift behavior in ambient atomic force microscope conditions. EAAI, 159, 111678. (www)

  @article{ , abstract = {This study outlines the effectiveness of combining numerical methods, Computer Vision (CV) and Machine Learning (ML) approaches to analyze and predict drift behavior in high-resolution Atomic Force Microscope (AFM) scanning procedures. Using Long Short-Term Memory (LSTM) models for time series analysis and the Light Gradient Boosting Machine (LightGBM) algorithm for predictive modeling, significant progress was achieved in understanding the dynamic and variable nature of drift and mitigating its impact on scanning. The models demonstrated a robust predictive capability, achieving approximately 94% accuracy in drift predictions. The study emphasizes the nonstationary characteristics of drift and demonstrates how the selection of features directly related to the target variable enhances the efficiency of the model and enables adaptive real-time correction. These findings confirm the predictive strength of the models and highlight the potential for integrating ML predictions with real-time feedback mechanisms to improve the resolution and stability of AFM imaging in both scientific and industrial applications.}, archive = {J_EAAI}, author = {D. Gemici Deveci and T. Karakoyun Barandır and Ö. Ünverdi and C. Celebi and L.Ö. Temur and D.Ç. Atilla}, doi = {10.1016/j.engappai.2025.111678}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111678}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comprehensive analysis and machine learning-based solutions for drift behavior in ambient atomic force microscope conditions}, volume = {159}, year = {2025}, }
• (2025). Detail-focused and polarization-guided multi-modality fusion for underwater image clarity enhancing. EAAI, 159, 111677. (www)

  @article{ , abstract = {Underwater optics imaging typically suffer from the impurities scattering and light absorption in dynamic and complex underwater environment, which significantly effect the clarity and visibility of images. Existing cutting-edge Underwater Image Enhancement (UIE) methods mostly focus on color correction and contrast enhancement neglect the textual and detail information of objects, leading to imbalance exposure and edge features missing. To overcome these problems, we propose a novel detail-focused and polarization guided multi-modality fusion network (DFPG-Net), for enhancing underwater images. Unlike the previous methods, we first construct a Detail-Focused Convolution (DFC) block for extracting features from underwater multimodal images, which integrates difference convolutions to capture prior and edge information. Meanwhile, polarization information is introduced with a Multi-scale Polarization Guided (MPG) fusion module, which intends to maintain and enhance the texture and details information from the degree of polarization information and angle of polarization information obtained from the polarized image. Additionally, a parallel progressive attention network is designed to explore and combine the valuable and discriminative information in feature learning stage. Extensive experiments on the constructed underwater dataset validate the effectiveness and superior performance of the proposed DFPG-Net, which against state-of-the-art methods in both machine evaluation metrics and visual perception.}, archive = {J_EAAI}, author = {Mingze Yao and Huibing Wang and Yudong Li and Wenzhe Liu and Xianping Fu}, doi = {10.1016/j.engappai.2025.111677}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111677}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detail-focused and polarization-guided multi-modality fusion for underwater image clarity enhancing}, volume = {159}, year = {2025}, }
• (2025). A rescheduling strategy based on deep reinforcement learning for new order insertion. EAAI, 159, 111676. (www)

  @article{ , abstract = {Manufacturing shops have common dynamic factors that often disrupt pre-established schedules. In production, a buffer time is typically set to account for such dynamic disturbances. However, which increases scheduling complexity and expands the Flexible Job Scheduling Problem (FJSP) into the Flexible Job Shop Rescheduling Problem (FJSRP). Consequently, an efficient method is urgently needed. In this context, the insertion of a new order alters the scheduling state. This study aims to address the objectives of minimizing the earliest completion time of orders and maximizing the utilization of machine resources. For the FJSRP, it is essential to establish both the independent insertion time and pre-processing buffer time while also devising a rescheduling method that balances the number of jobs across orders. To tackle these challenges, the Proximal Policy Optimization for Hybrid Feature Extraction Network (PPO-HFEN) strategy was developed. Innovations in this strategy include an attention mechanism that integrates a Multi-layer Hybrid Feature Extraction Network (HFEN) for feature extraction and job prioritization. Furthermore, a Mixed Reward Function (MRF) is introduced to resolve convergence issues during the training process. Experimental evaluations demonstrated that PPO-HFEN outperformed four state-of-the-art algorithms, yielding superior solutions in 21 out of 40 instances, thereby underscoring the effectiveness of the HFEN and MRF components. Moreover, the validity of the proposed rescheduling strategy is further validated on a real production foundry instance. The primary objective is to analyze the impact of varying order sizes, insertion times, and pre-processing times on the test results of diverse sampling methods.}, archive = {J_EAAI}, author = {Hao Sun and Zhiwei Hao and Ziyu Hu and Jin Lv and Cong Wang}, doi = {10.1016/j.engappai.2025.111676}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111676}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A rescheduling strategy based on deep reinforcement learning for new order insertion}, volume = {159}, year = {2025}, }
• (2025). Multi-objective virtual calibration method for parameters in two control strategies of a complex hybrid electric vehicle. EAAI, 159, 111669. (www)

  @article{ , abstract = {This paper presents a virtual calibration method for parameters in two coupled control strategy of a complex hybrid electric vehicle. Firstly, the sampling inputs are determined through the optimal Latin hypercube design, and the response are obtained by employing a validated physics-based model, forming the training dataset. Based on this dataset, the deep neural network model is employed to establish a surrogate model connecting control parameters and response, achieving a mean square error of 7.47 × 10 −7 . Using the deep neural network model, sensitivity analysis of calibration parameters is carried out through the Sobol indices method. The results of this analysis reveal calibration paths of key parameters and combinations, effectively reducing the search region. Finally, the double distance sorting genetic particle swarm optimization algorithm is employed to solve the multi-objective optimization problem. The optimization results indicate a 2.09 % reduction in the mean square error of speed, a 24.28 % reduction in equivalent fuel consumption, and a 14.71 % reduction in battery capacity loss compared to the initial values. These outcomes clearly confirm the effectiveness of the proposed calibration method in enhancing calibration efficiency and provide valuable theoretical guidance for actual calibration.}, archive = {J_EAAI}, author = {Jiangbin Yu and Xin Zhang and Haohua Yan and Xinlin Li}, doi = {10.1016/j.engappai.2025.111669}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111669}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective virtual calibration method for parameters in two control strategies of a complex hybrid electric vehicle}, volume = {159}, year = {2025}, }
• (2025). Spatio-temporal graph interaction networks for teacher behavior description in classroom scene. EAAI, 159, 111668. (www)

  @article{ , abstract = {Teacher behavior description is crucial for improving teaching effectiveness and understanding classroom dynamics. Traditional methods rely on meticulous manual observation and recording, which is extremely complex and time-consuming when dealing with many teaching videos. Recently proposed video captioning technologies can automatically analyze and describe object behaviors, reducing manual intervention and providing powerful support for improving teaching effectiveness. Existing video captioning methods, however, fall short in describing teachers’ behaviors because they overlook the temporal progression of teachers’ actions, the background changes between frames, and the frequent interactions between teachers and students in real classroom environments. To address these issues, we propose a video captioning method leveraging a spatio-temporal graph interaction network (STGIN) to describe teachers’ behaviors in classroom scenes. Comprising the teacher–student spatial interaction module (TSI), teacher temporal context module (TTC), and description generator, STGIN captures spatial interaction relationships between teacher and students, extracts temporal dynamics features of the teachers’ behaviors, and precisely generates teacher behavior descriptions. To validate the proposed scheme, we collected a teacher behavioral description (TBD) dataset consisting of 2000 videos. Extensive experiments on our TBD dataset and two generic scene datasets confirm the effectiveness and robustness of the proposed STGIN in real classroom environments.}, archive = {J_EAAI}, author = {Yu Xiong and Chengyang He and Lulu Chen and Ting Cai}, doi = {10.1016/j.engappai.2025.111668}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111668}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatio-temporal graph interaction networks for teacher behavior description in classroom scene}, volume = {159}, year = {2025}, }
• (2025). Artificial intelligence-driven protocol for secure and standardized maneuver control in electrical substations. EAAI, 159, 111667. (www)

  @article{ , abstract = {Notwithstanding recent advances in substation automation, no existing protocol integrates human–machine interaction, intelligent interlocking, operational autonomy, and artificial intelligence analysis in sequential maneuvering contexts. This study proposes an automated interface to optimize and control switching operations in electrical substations by integrating operational protocols, automated documentation generation, and artificial intelligence techniques with interactive graphical visualization. The developed solution enables sequential command execution, classification of operational events, and automatic generation of auditable reports, enhancing accuracy and traceability in operations. A total of 108 real files, corresponding to 54 events with documented failures, were analyzed and used to train and validate a recurrent convolutional neural network model. The system achieved an accuracy of 82.92% in error detection, along with reductions of 42.7% in the average operational response time and 38.5% in failure frequency. In addition to standardizing procedures, the interface demonstrated adaptability to different substation topologies and configurations, establishing itself as a scalable, secure, and efficient alternative for assisted operation environments. The results suggest that the proposed solution contributes to reducing inconsistencies, increasing decision-making autonomy, and strengthening operational safety in the power sector.}, archive = {J_EAAI}, author = {Gustavo Havila F. Campos and Viviane M. Gomes Pacheco and Marcio Rodrigues C. Reis and Clóves Gonçalves Rodrigues and Saulo Rodrigues Silva and Antonio Paulo Coimbra and Wesley Pacheco Calixto}, doi = {10.1016/j.engappai.2025.111667}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111667}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence-driven protocol for secure and standardized maneuver control in electrical substations}, volume = {159}, year = {2025}, }
• (2025). Multimodal prompt state space models for unified adverse weather removal. EAAI, 159, 111666. (www)

  @article{ , abstract = {Image restoration under adverse weather conditions can improve the robustness of outdoor machine vision systems. Most existing methods are designed to deal with just one type of adverse weather degradation using uni-modal image information. Moreover, popular restoration backbones such as convolutional neural networks and transformer structures have inherent limitations. In this paper, we propose a multimodal locally enhanced state space model for unified adverse weather removal. It leverages multimodal information to provide comprehensive prompts, enabling the model to achieve unified adverse weather removal from diverse degradation types and levels effectively. The locally enhanced state space model simultaneously achieves local feature representation and long-range dependency modeling with linear complexity, which promotes the performance and efficiency of the restoration network. Moreover, we introduce a novel multiscale gated feed-forward network to realize the effective propagation of multiscale features. Specifically, our method first uses parallel multimodal encoders to represent text and image features adequately. Then, it utilizes multiple degradation semantic adaptors to conduct adaptive learning from the encoded features to generate descriptions of weather-specific knowledge. Finally, the multimodal semantic information is fed into the decoder of the restoration network to guide adverse weather removal. Experiments show that our method achieves state-of-the-art results across four tasks and 33 benchmarks. Our method improves the average peak signal-to-noise ratio (PSNR) by 2.45 decibel (dB) and 2.09 decibel (dB), and structural similarity index measure (SSIM) by 3% and 1.6% over the sub-optimal all-in-one and one-by-one methods on four tasks, respectively. The running time is only 0.38 s.}, archive = {J_EAAI}, author = {Pengyue Li and Wentao Li and Jiandong Tian and Yandong Tang}, doi = {10.1016/j.engappai.2025.111666}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111666}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodal prompt state space models for unified adverse weather removal}, volume = {159}, year = {2025}, }
• (2025). A multi-agent-based cooperative optimization control method for the motor energy consumption and tracking accuracy of an unmanned electric tractor. EAAI, 159, 111665. (www)

  @article{ , abstract = {The unmanned driving of electric tractors and its related research has become a new direction in tractor development. Poor coordination between the drive system and energy system, high motor energy consumption, and low path tracking accuracy are the main challenges restricting the development of unmanned driving in electric tractors. Current research primarily focuses on rotary tillage and plowing operations, while studies on unmanned cooperative control under transportation conditions are severely lacking. This paper proposes an unmanned cooperative optimization control method for single-motor electric tractors under transportation conditions, taking into account energy consumption and tracking accuracy. The research includes longitudinal velocity control for the tractor under transportation conditions, based on an improved Deep Deterministic Policy Gradient (DDPG) algorithm, aimed at enhancing motor efficiency, and lateral velocity control using a Deep Q-Network (DQN) algorithm to improve path tracking accuracy during unmanned operation. Finally, bench tests were conducted on a physical prototype, and a complete validation system was established. The results show that this method reduces the total machine energy consumption of the electric tractor under transportation conditions by 38.8 % on hard surfaces and by 25.3 % on off-road terrain, while also improving path tracking accuracy under both surface types, with lateral deviation during straight-line travel maintained below 3.5 cm (cm). This study provides a theoretical foundation and technical support for realizing intelligent control in advanced unmanned agricultural cooperative systems.}, archive = {J_EAAI}, author = {Wenxiang Xu and Yejun Zhu and Maohua Xiao and Mengnan Liu and Yanpeng Yang and Xinyuan Yue and Tai Chen}, doi = {10.1016/j.engappai.2025.111665}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111665}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-agent-based cooperative optimization control method for the motor energy consumption and tracking accuracy of an unmanned electric tractor}, volume = {159}, year = {2025}, }
• (2025). PAVEN: A perceptual algorithm for versatile video encoding using neural networks. EAAI, 159, 111664. (www)

  @article{ , abstract = {This work introduces the Perceptual Algorithm for Versatile video Encoding using Neural Networks (PAVEN), a subjective video coding algorithm designed to reduce the bit rate in videos encoded with the Versatile Video Coding (VVC) standard without compromising subjective video quality. The algorithm uses a deep learning model specifically trained by the authors to account for the specific characteristics of video signals. The trained model outperforms others in the literature by more accurately identifying areas of the frames where viewers are most likely to focus their attention. The output of the deep learning model is further processed to merge all disjoint areas and adapt the result to the Coding Tree Unit (CTU) size in VVC, allowing for greater compression in less important areas. The results show an average reduction in bit rate of 7% while maintaining the same subjective video quality, validated through viewer interviews using the Mean Opinion Score (MOS) metric.}, archive = {J_EAAI}, author = {Pablo Fernández-Lagos and Belén Ríos and Hari Kalva and Gabriel Cebrián-Márquez and Guillermo Vigueras and Antonio Jesus Diaz-Honrubia}, doi = {10.1016/j.engappai.2025.111664}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111664}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {PAVEN: A perceptual algorithm for versatile video encoding using neural networks}, volume = {159}, year = {2025}, }
• (2025). Sketch & fetch: Draw a logo sketch to fetch the suspect from surveillance footage. EAAI, 159, 111663. (www)

  @article{ , abstract = {Logo Sketch-assisted Suspect Retrieval (LSSR) enables rapid identification of individuals based on eyewitness-drawn logo sketches, especially when facial details are obscured or unavailable in surveillance footage. The proposed Sketch-Fetch Generative Adversial Network (SF-GAN) translates sketches into realistic logo images, while the enhanced E-YOLOv7 (Elite-You Only Look Once version 7) detects logos on clothing in real-time. Local self-similarity descriptors with Euclidean matching are used to retrieve the query person. SF-GAN is trained on 100 sketch-image logo classes and shows adaptability to new designs. It achieves a low FID (Fréchet Inception Distance) score of 0.2, indicating high-quality generation. The system is tested on benchmark datasets under challenging conditions, including blur, occlusion, and low resolution. Achieving 95.6 % accuracy, the LSSR framework outperforms state-of-the-art approaches in logo-based suspect retrieval.}, archive = {J_EAAI}, author = {Yogameena Balasubramanian and Nagavani Chandrasekaran}, doi = {10.1016/j.engappai.2025.111663}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111663}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sketch & fetch: Draw a logo sketch to fetch the suspect from surveillance footage}, volume = {159}, year = {2025}, }
• (2025). A survey of federated learning-based gradient compression for internet of vehicles. EAAI, 159, 111662. (www)

  @article{ , abstract = {The Federated Learning (FL) paradigm in the Internet of Vehicles (IoV) leverages vehicular networks and intelligent technologies to integrate data from traffic devices, aiming to develop a smart transportation system with high throughput, low latency, privacy protection, and collaborative multi-party training. However, frequent exchanges of model parameters between numerous vehicle nodes and roadside units (RSUs) in FL lead to uplink channel overload, which poses a significant challenge to system development. To this end, existing works introduce Gradient Compression (GC) technologies, incorporating methods such as count sketch, dynamic adjustment, and sparse matrices. These methods help to reduce communication overhead, enhance transmission efficiency, and maintain FL model training accuracy, making GC a crucial solution for overcoming communication barriers in IoV-based FL systems. In this paper, we first investigate GC technologies, systematically categorizing them by Quantization, Sparsification, and Driving Strategy-orientated compression, followed by using metrics such as efficiency, load, and latency to evaluate them. Secondly, we compare the pros and cons of different compression technologies in addressing communication problems and analyze their key characteristics. Finally, considering the deep integration of IoV and FL, we explore future research directions of the FL framework for IoV, analyze potential challenges, and propose corresponding solutions in conjunction with current mainstream deep learning models.}, archive = {J_EAAI}, author = {Ling Xing and Zhaocheng Luo and Jianping Gao and Kaikai Deng and Honghai Wu and Huahong Ma}, doi = {10.1016/j.engappai.2025.111662}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111662}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A survey of federated learning-based gradient compression for internet of vehicles}, volume = {159}, year = {2025}, }
• (2025). A dual-objective contrastive learning approach with dynamic self-adaption for zero-shot fault diagnosis. EAAI, 159, 111660. (www)

  @article{ , abstract = {Fault type classification and fault severity identification are two critical and complementary tasks in fault diagnosis of industrial machines, providing essential information for the maintenance and safety of the machines. However, variable operating conditions in industrial settings make it hard to collect comprehensive fault data covering all possible types and severities, thereby limiting diagnostic efficiency. To overcome these challenges, a novel multi-task network approach is proposed to detect fault type and severity simultaneously even with zero novel samples. Discriminative features are extracted through a contrastive network with task-specific projection heads, enabling the capture of distinct representations for fault type and severity. Two zero-shot mapping spaces are constructed to diagnose fault types and severity by aligning feature representations with the semantic information of fault types and severity. A dynamic self-adaptation optimization mechanism is introduced considering the dependency of fault severity on fault types. It enhances the identification of fault severity. The proposed method was evaluated on two bearing datasets. It achieved up to 89.4 % accuracy for fault type and 83.42 % for fault severity under zero-shot settings, outperforming baselines and demonstrating strong real-world applicability.}, archive = {J_EAAI}, author = {Yifan Wu and Min Xia}, doi = {10.1016/j.engappai.2025.111660}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111660}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual-objective contrastive learning approach with dynamic self-adaption for zero-shot fault diagnosis}, volume = {159}, year = {2025}, }
• (2025). Machine learning methodologies in video capsule endoscopy-based bleeding analysis: A systematic review of progress and prospects (2008–2024). EAAI, 159, 111659. (www)

  @article{ , abstract = {The rising global mortality and morbidity of gastrointestinal (GI) bleeding, coupled with the limitations of traditional endoscopic methods, highlight the urgent need for innovative solutions in this field. Video Capsule Endoscopy (VCE) has emerged as a significant innovation, enabling a non-invasive visualization of the GI tract, crucial for identifying bleeding sources which may be inaccessible by traditional endoscopic methods. However, the efficacy of VCE is limited by challenges such as labor-intensive analysis and human error, where Machine Learning (ML) can offer significant improvements. This systematic review explores the role of ML in automating the analysis of GI bleeding in VCE, aiming to enhance diagnostic accuracy, reduce manual effort, and improve patient outcomes. A total of 114 studies published between 2008 and 2024 were systematically reviewed and categorized into four groups: segmentation, classification, detection, and combined methodologies. Each group was analyzed to assess the effectiveness and quality of the studies, using established assessment tools such as Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) and Prediction model Risk Of Bias Assessment Tool (PROBAST). An overview of open-source datasets, evaluation metrics, ML and its interpretability, and image processing methodologies used in the field was also conducted. Additionally, the review discusses the limitations and challenges encountered and highlights future prospects for ML methodologies in analyzing VCE frames for GI bleeding.}, archive = {J_EAAI}, author = {Tanisha Singh and Shreshtha Jha and Nidhi Bhatt and Palak Handa and Nidhi Goel and S. Indu}, doi = {10.1016/j.engappai.2025.111659}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111659}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning methodologies in video capsule endoscopy-based bleeding analysis: A systematic review of progress and prospects (2008–2024)}, volume = {159}, year = {2025}, }
• (2025). Enhanced light detection and ranging simultaneous localization and mapping based on three-dimensional moving object tracking in dynamic scenes. EAAI, 159, 111658. (www)

  @article{ , abstract = {Moving objects can cause incorrect feature matches, significantly affecting light detection and ranging (LiDAR) odometry accuracy and mapping quality in simultaneous localization and mapping (SLAM). This paper proposes a learning-based SLAM framework that reduces the impact of moving objects on SLAM by estimating their motion through segmentation and tracking. We design a deep dynamic LiDAR odometry (DyLO) network by introducing a hierarchical attention mechanism and a dynamic mask network in point clouds to improve odometry performance in dynamic scenes. Additionally, we design a dynamic object segmentation and tracking module to extract motion information from dynamic scenes and combine it with DyLO to form a joint factor graph backend optimization for SLAM, thereby developing a complete SLAM system. Our SLAM system is validated on the public dataset with 64-line LiDAR and a self-established campus dataset with 16-line LiDAR. We compare the proposed DyLO with several outstanding iterative closest point (ICP) methods, the deep learning-based method of LiDAR odometry network (LO-Net), and the feature-based odometry approach. Compared with the outstanding LO-Net, the proposed algorithm reduces the translation error by 4.5% and the rotation error by 3.3% on average. Furthermore, our ablation experiments demonstrate that incorporating motion estimation of moving objects and backend optimization can remarkably improve the odometry accuracy.}, archive = {J_EAAI}, author = {Tao Gao and Hu Ran and Hongyu Chi and Dunwen Wei}, doi = {10.1016/j.engappai.2025.111658}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111658}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced light detection and ranging simultaneous localization and mapping based on three-dimensional moving object tracking in dynamic scenes}, volume = {159}, year = {2025}, }
• (2025). Decoding industrial-scale battery manufacturing process through integration of causal graphs into explainable artificial intelligence. EAAI, 159, 111657. (www)

  @article{ , abstract = {Modeling and analyzing industrial-scale lithium-ion battery (LIB) manufacturing process present significant challenges due to the numerous variables and their complex interrelationships. While previous studies have utilized machine learning and explainable artificial intelligence to discern complex patterns and identify crucial variables from data, these methods often overlook the causal connections among input variables. This oversight can potentially lead to inaccurate interpretations and limited insights, particularly regarding how influences accumulate throughout the process. To address these limitations, this study introduces a comprehensive modeling and explanatory framework that incorporates causal information among input variables. By employing the Shapley flow algorithm, which propagates attributions along the edges of the causal graph, our framework successfully identified key process variables that could have not been isolated under conventional approaches. Furthermore, a detailed analysis of impact pathways for individual process parameters was also obtained by focusing on relevant edges. Through preliminary validation with a simulated system and subsequent application to real-world data from leading commercial LIB manufacturing enterprise, we confirmed the method’s efficacy in accurately pinpointing significant variables. Our analysis also introduced novel insights into the impact pathways of process parameters, previously unexplored by previous approaches. This new understanding offered engineers deeper insights and actionable strategies, boosting the potential for enhanced process analysis and decision-making capabilities.}, archive = {J_EAAI}, author = {Haechang Kim and Ji Young Yun and Eunjoo Jung and Bora Lee and Hyeongseok Kim and Jong Min Lee}, doi = {10.1016/j.engappai.2025.111657}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111657}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Decoding industrial-scale battery manufacturing process through integration of causal graphs into explainable artificial intelligence}, volume = {159}, year = {2025}, }
• (2025). Consistent learning for incomplete multi-view clustering. EAAI, 159, 111656. (www)

  @article{ , abstract = {With the rapid development of artificial intelligence in multi-view data analysis, clustering incomplete multi-view data has become a hot direction for studying data deficiency in real scenarios. Numerous innovative methods have been proposed to tackle these problems effectively. However, most studies ignore the fact that samples with different numbers of available instances should have different weights. In addition, as the structure of multi-view data becomes increasingly complex, we should consider introducing a hyper-Laplacian matrix rather than the traditional Laplacian matrix to mine higher-order semantic information. This paper proposes an effective multi-view clustering approach to address the issues above. We incorporate weight vectors reflecting the number of available instances into the learning process of the consensus representation matrix. In addition, we consider using hyper-Laplacian matrix coupling to represent the structural information of the matrix and the sample. This paper conducts many experiments on eight different datasets and selects nine advanced incomplete multi-view clustering methods for comparison. A large number of experiments demonstrate the effectiveness of our method on incomplete multi-view clustering.}, archive = {J_EAAI}, author = {Haixia Shi and Youlong Yang}, doi = {10.1016/j.engappai.2025.111656}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111656}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Consistent learning for incomplete multi-view clustering}, volume = {159}, year = {2025}, }
• (2025). Enhancing pine wilt disease detection with synthetic data and external attention-based transformers. EAAI, 159, 111655. (www)

  @article{ , abstract = {The catastrophic effects of Pine Wilt Disease (PWD) and the lack of a defined cure make it a danger to the world’s forests. The early detection of PWD becomes essential when establishing efficient approaches to mitigation. However, collecting data, labelling, and proposing a system for PWD detection is a challenge. This work presents an innovative PWD system and uses synthetic data sets to overcome these challenges. In particular, it emphasises two contributions: (1) the implementation of a multi-head external attention mechanism focused on improving computational effectiveness and model performance, and (2) the integration of synthetic data to address the problem of data scarcity in PWD detection. To solve the high computational cost of the self-attention function from the original transformer, we introduce external attention, which enables two linear transformations to take fewer amounts of parameters. Whereas the original self-attention is designed to capture relations inside a single sample, the external attention seeks inter-sample connections through memory modules. It is also computationally efficient compared to the wider context information, and it may consume as much computational power as the 1 by 1 convolution. Our external attention mechanism is multi-head attention, which enables the model to pay close attention to multiple features in the data and find different types of relationships. The proposed methodology achieves high performance across multiple metrics for both individual and ensemble datasets. For individual datasets, the model achieves accuracy (90%–98%), Global Detection Rate (GDR) (90%–98%), Matthews correlation coefficient (MCC) (0.80–0.96) and Kappa (0.80–0.96). For ensemble datasets, the model achieves accuracy (94%–96%), GDR (94%–96%), MCC (0.88 - 0.92) and Kappa (0.88 - 0.92). These results confirm that the presented technique is valid and effective in the early-stage detection of PWD in trees, and can be implemented in forest management.}, archive = {J_EAAI}, author = {Sareer Ul Amin and Yonghoon Jung and Muhammad Fayaz and Bumsoo Kim and Sanghyun Seo}, doi = {10.1016/j.engappai.2025.111655}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111655}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing pine wilt disease detection with synthetic data and external attention-based transformers}, volume = {159}, year = {2025}, }
• (2025). Aspect-based sentiment analysis with semantic and syntactic enhanced multi-layer fusion model. EAAI, 159, 111654. (www)

  @article{ , abstract = {Aspect-based sentiment analysis (ABSA) aims to identify the sentiment polarity of specific aspect words or phrases in a sentence. Although recent studies have used attention mechanisms or syntactic relations of dependency trees to establish links between aspect terms and sentences, these approaches are imperfect in effectively fusing syntactic and semantic contextual information. Therefore, in this paper, we propose a novel multi-layer fusion model (MLFM) based on artificial intelligence (AI) techniques to efficiently fuse semantic and syntactic information for sentiment analysis. In the model, we first propose a new bi-graph convolutional network module for aspect term-centered aspect nodal attention (Aspect-NA) to enhance Semantic and Syntactic learning. Within Aspect-NA, we introduce dependency embedding and propose a dual embedding update mechanism that pays more attention to the influence of dependency types and semantics. In addition, we propose an adaptive hierarchical cross-attention (AHCA) for fusing the semantic information of aspect term with their associated syntactic features. AHCA not only effectively fuses features between syntax and semantics of the context, but also carries out the key features. We conducted experiments on six benchmark datasets, and the results show that our proposed model outperforms most baseline methods. The code and datasets involved in this paper are provided on https://github.com/jims-bug/MLFM.git .}, archive = {J_EAAI}, author = {Song Jin and Qing He and Yuji Wang and Nisuo Du and Wenjing Lei}, doi = {10.1016/j.engappai.2025.111654}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111654}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Aspect-based sentiment analysis with semantic and syntactic enhanced multi-layer fusion model}, volume = {159}, year = {2025}, }
• (2025). Physics-informed cross layer temporal frequency transformer network for remaining useful life prediction of rolling bearings. EAAI, 159, 111653. (www)

  @article{ , abstract = {In response to the challenge of accurately capturing long-term degradation dependencies in the remaining useful life (RUL) prediction task for rolling bearings under extreme operating conditions, which leads to low accuracy of prediction results, a physics-informed cross-layer temporal frequency Transformer method for rolling bearing RUL prediction, named WeiCLTF, has been proposed. Firstly, a cross-layer information transfer mechanism is established between adjacent layers within the Transformer, which introduces cross-layer regularization to facilitate alternate information sharing between adjacent Transformer layers, reducing the discrepancy between the key matrix of the multi-head attention module and the query matrix of the preceding layer. The current layer's feed-forward module weights are then utilized as the weights for the subsequent lower layer, ensuring that the higher layers of the Transformer can accurately refine the information from the lower layers, thus enhancing the network's ability to capture long-term dependencies. Secondly, The self-attention layers in the first two layers of the Transformer are replaced with Fourier modules. This alternation between the encoder and decoder enables the model to capture time-frequency information from the alternating Fourier transform and inverse Fourier transform, thereby enhancing the accuracy of predictions. Finally, use the Weibull failure model to describe the failure probability of the bearing over its lifecycle, and integrate it into the loss function of the neural network training to enhance the physical consistency of the training. The proposed model was applied and verified using both public datasets and datasets of wind turbine gearboxes from real wind farms to confirm its superiority.}, archive = {J_EAAI}, author = {Runyan Zhao and Baoping Tang and Xiaolong Zhang and Lei Deng}, doi = {10.1016/j.engappai.2025.111653}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111653}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed cross layer temporal frequency transformer network for remaining useful life prediction of rolling bearings}, volume = {159}, year = {2025}, }
• (2025). A multi-point combined prediction model for deformation monitoring of concrete dams. EAAI, 159, 111652. (www)

  @article{ , abstract = {Accurate deformation prediction is essential for dam risk assessment and real-time early-warning. However, most existing approaches concentrate primarily on single-point displacement analysis, neglecting the spatiotemporal difference and intrinsic connections of regional deformation, which limits comprehensive understanding and accurate evaluation of the dam's structural response. Therefore, a novel multi-point combined prediction model (MCPM) for dam displacement prediction is proposed in this study. Cloud digital features generated by the cloud model are introduced to characterize the holistic displacement pattern of each measurement point. Then agglomerative hierarchical clustering is employed to group the measurement points exhibiting similar displacement patterns. Finally, a multi-input multi-output prediction framework, integrating convolutional neural network and bi-directional long short-term memory, is developed to simultaneously predict displacement for measurement points within the partition. Case-study analysis demonstrates that the MCPM achieves prediction accuracy comparable to traditional single-point models while significantly enhancing robustness and stability. Consequently, the proposed model provides valuable insights and practical guidance for dam operational management.}, archive = {J_EAAI}, author = {Jiaquan Yang and Huaizhi Su and Hongchen Liu and Kefu Yao and Yining Qi and Yongbo Wang}, doi = {10.1016/j.engappai.2025.111652}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111652}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-point combined prediction model for deformation monitoring of concrete dams}, volume = {159}, year = {2025}, }
• (2025). Seismic damage states prediction of in-service bridges using feature-enhanced swin transformer without reliance on damage indicators. EAAI, 159, 111651. (www)

  @article{ , abstract = {To achieve real-time and efficient evaluation of seismic damage to structures, this study proposes an improved deep learning-based model, the deep feature-enhanced Swin Transformer (CC-SwinT). This model overcomes the influence of service life on the time-varying damage indicators of bridges. By eliminating the reliance on seismic performance and damage indicators, it predicts the seismic damage state of in-service bridges based solely on the response of structure. The CC-SwinT model integrates continuous wavelet transform (CWT) technology and the context anchored attention (CAA) mechanism to enhance the extraction of structure response features of bridge piers. This integration enables the model to effectively mine time-frequency characteristics and capture non-local long-term dependencies in structure responses. To comprehensively train the CC-SwinT model, a structure response database for in-service bridges was constructed based on a data-driven objectives, analyzing the impacts of service conditions on the seismic performance of bridges. Subsequently, transfer learning methods were applied, and the performance of the CC-SwinT framework was evaluated using various metrics to highlight its exceptional feature extraction and prediction capabilities. Furthermore, the Gradient-weighted Class Activation Mapping (Grad-CAM) interpretability technique was used to explore the decision-making process and feature focus of CC-SwinT. The findings of this study provide a valuable reference for seismic damage prediction of in-service structures and rapid post-earthquake rescue response.}, archive = {J_EAAI}, author = {Yalin Li and Zhen Sun and Sujith Mangalathu and Yaqi Li and Hao Yang and Weidong He}, doi = {10.1016/j.engappai.2025.111651}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111651}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Seismic damage states prediction of in-service bridges using feature-enhanced swin transformer without reliance on damage indicators}, volume = {159}, year = {2025}, }
• (2025). Deep reinforcement learning approach for hybrid renewable energy systems optimization. EAAI, 159, 111650. (www)

  @article{ , abstract = {The sizing of hybrid renewable energy systems (HRES) is a major challenge faced in contemporary energy research. The optimal configuration based on the specific consumption requirements is essential for strategic energy planning. Effective sizing must balance the investment costs, reliability, environmental impacts, and greenhouse gas emissions while satisfying the expected energy requirements. This study proposes a novel multi-criteria sizing approach based on deep reinforcement learning (DRL). The DRL agent is guided by a reward function that integrates three essential performance metrics: energy cost (LCOE), renewable energy fraction (REF), and the loss of power supply probability (LPSP). A penalty function is also included to consider the reliance on external sources, such as diesel generators and the public grid, promoting greater autonomy and renewable usage. The DRL-based approach was implemented and tested on three distinct demand profiles, using hourly data for one year. A comparative analysis was conducted against three established methods: particle swarm optimization (PSO), multi-objective PSO (MOPSO), and non-dominated sorted genetic algorithm (NSGA-II). The results indicate that DRL significantly outperforms all the benchmark methods in terms of economic efficiency. DRL achieves a significant reduction in the energy costs, ranging from 21.33 % to 30.09 % when compared with PSO, 27.89 %–30.27 % when compared with MOPSO, and 27.63 %–28.47 % when compared with NSGA-II. These findings demonstrate that DRL presents a robust and adaptive framework for the sizing and operational control of HRES. DRL presents more autonomous, cost-effective, and scalable renewable energy solutions by minimizing the energy costs while maintaining the system reliability.}, archive = {J_EAAI}, author = {Inoussa Legrene and Tony Wong and Louis-A. Dessaint}, doi = {10.1016/j.engappai.2025.111650}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111650}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep reinforcement learning approach for hybrid renewable energy systems optimization}, volume = {159}, year = {2025}, }
• (2025). Artificial intelligence-enabled defect detection method and engineering application of ceramic mug. EAAI, 159, 111648. (www)

  @article{ , abstract = {In the manufacturing process of ceramic mugs, the detection of micro-surface defects faces technical challenges of high difficulty and low efficiency, and efficient and high-quality production lines are crucial to maintaining market competitiveness. The goal of this study is to improve the accuracy and efficiency of detection by developing a defect detection algorithm and equipment based on deep learning, thereby improving product quality and reducing production costs. The research uses the visual algorithm You Only Look Once version 8 (YOLOv8) in the field of artificial intelligence as the baseline model. Firstly, a slice pre-training layer is designed to reduce the memory loss of large images to the graphics card. Secondly, the model structure is reconstructed to adapt to small target detection. In addition, a mixed local channel cross-stage feature fusion module is proposed to enhance the recognition ability of small targets. Finally, a detection head with shared parameters is designed to further reduce the number of parameters. In terms of engineering application, a set of test equipment was developed and the corresponding software was written. Experiments show that the accuracy of the algorithm is 21.3 % higher than that of YOLOv8, and the parameter amount is reduced by 67 %. Compared with manual detection, the equipment efficiency is increased by 47.06 %, and the detection success rate is 99.6 %. Therefore, the research in this paper provides an efficient and reliable solution for industrial automation detection.}, archive = {J_EAAI}, author = {Wenjie Mao and Hu Wu and Shilong Xie and Linyuxuan Li and Xianhai Yang}, doi = {10.1016/j.engappai.2025.111648}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111648}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence-enabled defect detection method and engineering application of ceramic mug}, volume = {159}, year = {2025}, }
• (2025). Traffic parameter prediction and congestion assessment based on spatiotemporal characteristics of highway network. EAAI, 159, 111646. (www)

  @article{ , abstract = {As the demand for expressway travel increases and congestion becomes more and more prominent, real-time prediction of traffic parameters of each road in the expressway network and assessment of congestion on the road are important guides in reducing traffic congestion and improving the operational efficiency of the road network. This paper first proposes a novel deep learning road network traffic flow prediction algorithm (T-TGCN, Temporal Tensor-Graph Convolutional Network). This study first measures the correlation between roads in the highway network from three perspectives: the topological relationships between road nodes, connectivity distance, and data similarity, resulting in a traffic graph tensor. Next, based on the periodicity of traffic parameters, historical data is segmented. We then combine the sequence-to-sequence model (Seq2Seq) with the tensor graph convolutional network to construct the T-TGCN algorithm, which captures the temporal and spatial correlations of traffic parameters. Finally, using the traffic parameters predicted by the T-TGCN algorithm, along with time periods, weather, and road location information as evaluation metrics, and addressing the issue of imbalanced congestion level categories, we propose a CBL-LightGBM-based (Class Balance Loss-Light Gradient Boosting Machine) algorithm for highway network road congestion assessment. Extensive numerical experiments validate the effectiveness and applicability of the T-TGCN algorithm and the CBL-LightGBM algorithm.}, archive = {J_EAAI}, author = {Kaiyuqi Guan and Zhihang Zhou and Jianghui Wen}, doi = {10.1016/j.engappai.2025.111646}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111646}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Traffic parameter prediction and congestion assessment based on spatiotemporal characteristics of highway network}, volume = {159}, year = {2025}, }
• (2025). Wind turbine blade surface defect detection model based on improved you only look once version 10 small and integrated compression. EAAI, 159, 111645. (www)

  @article{ , abstract = {This paper introduces a new model built on the You Only Look Once version 10 small (YOLOv10s) baseline to address challenges in wind turbine blade surface defect detection, including low accuracy due to complex backgrounds, small targets, and dense defects, as well as issues of model over-parameterization and high memory consumption. Several improvements are incorporated to enhance detection accuracy: (1) the original Spatial Pyramid Pooling Fast (SPPF) module is replaced with a lightweight Contextual Augmentation Module (CAM-DW) to improve feature fusion, (2) Efficient Multi-Scale Attention (EMA) substitutes Partial Self-Attention (PSA) for better feature extraction, and (3) the Wise-Intersection over Union version 1 (WIoU-V1) loss function optimizes detection performance for high-density defect samples. To tackle the problem of excessive parameters and memory usage, an integrated compression method is proposed, which combines isomorphic pruning to reduce parameters and memory usage with channel-wise knowledge distillation to recover accuracy lost during pruning, thus striking a balance between model complexity and performance. Experimental results show that the proposed model reduces parameters by 69.7 % and memory usage by 68.1 % compared to the baseline. Its mean Average Precision mAP50 (prediction confidence threshold: 0.5) and mAP50-95 (prediction confidence thresholds: 0.5–0.95) improved by 3.3 % and 3.8 %, respectively, while detection speed increased by 46.7 Frames Per Second (FPS). These results demonstrate that the proposed model outperforms mainstream models, significantly enhancing the accuracy and efficiency of wind turbine blade surface defect detection, and providing crucial support for intelligent wind power equipment operation and maintenance.}, archive = {J_EAAI}, author = {Hang Liu and Sheng Liu and Zhijian Liu and Ben Niu and Jing Xie and Chi Luo and Zhiyu Shi}, doi = {10.1016/j.engappai.2025.111645}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111645}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Wind turbine blade surface defect detection model based on improved you only look once version 10 small and integrated compression}, volume = {159}, year = {2025}, }
• (2025). A novel framework for crack segmentation using image augmentation and a CannyNet. EAAI, 159, 111644. (www)

  @article{ , abstract = {Computer-vision based crack detection is highly dependent on the quality of the segmentation process and remains a challenging task due to its complexity. In this paper, a framework for image segmentation that incorporates image augmentation method and a CannyNet is proposed to improve segmentation results. The style transfer is employed for image augmentation. A novel deep neural network for crack image segmentation, named CannyNet, is proposed to enhance the recognition capability for tiny cracks. Moreover, to improve the precision of CannyNet predictions, Bayesian optimization approach is employed to optimize network hyperparameters. The proposed framework for crack segmentation was verified using four open-source dataset and a new constructed dataset by conducting experimental test. A comparison of segmentation models indicates that style transfer method enhances the model's generalization, and the CannyNet demonstrates superior performance. The Bayesian optimization strategy is capable of optimizing the architecture of the CannyNet, thereby improving crack segmentation results.}, archive = {J_EAAI}, author = {Gang Liu and Xuming Li and Jin Di and Rui Sun and Fengjiang Qin and Yi Su and Dewei Liu}, doi = {10.1016/j.engappai.2025.111644}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111644}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel framework for crack segmentation using image augmentation and a CannyNet}, volume = {159}, year = {2025}, }
• (2025). Semi-supervised generative adversarial networks for imbalanced skin lesion diagnosis with an unbiased generator and informative images. EAAI, 159, 111643. (www)

  @article{ , abstract = {Skin cancer remains a significant global health challenge, necessitating effective early detection methods. Traditional supervised learning approaches for skin lesion classification often require extensive labeled datasets, which are costly and time-consuming to obtain. This study addresses the limitations of semi-supervised learning in skin cancer diagnosis, particularly issues related to classification bias toward majority classes and the impact of incorrect pseudo-labels from low-confidence unlabeled images. We propose an unbiased bad generator within the weighted bad semi-supervised generative adversarial network (WB-SGAN), which integrates self-social consistency regularization and a new weighted inversed cross-entropy loss function for informative images. Additionally, a weighted cross-entropy loss function is used to reduce the bias of the classifier's predictions on labeled and unlabeled images. This framework enhances the generation of informative fake samples, thereby reducing bias in pseudo-labels and improving classification performance. Our experiments demonstrate that WB-SGAN outperforms existing state-of-the-art (SOTA) methods, achieving balanced accuracies of 80.35 % and 74.19 % on the ISIC-2018 and PAD-UFES datasets, respectively, even with just 5 % labeled data. This approach highlights the visual and volumetric aspects of training images and their labeling, offering a solution for skin lesion classification in semi-supervised domains with limited and imbalanced datasets.}, archive = {J_EAAI}, author = {Mohammad Saber Iraji}, doi = {10.1016/j.engappai.2025.111643}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111643}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised generative adversarial networks for imbalanced skin lesion diagnosis with an unbiased generator and informative images}, volume = {159}, year = {2025}, }
• (2025). Enhanced flower image classification: A multimodal approach leveraging persian texts and deep learning. EAAI, 159, 111642. (www)

  @article{ , abstract = {Image flower classification is a challenging task that has recently been enhanced by incorporating textual modalities, though Persian texts have been neglected. We propose a novel multimodal approach for flower image classification that leverages both visual and textual information. Our method integrates Persian text with image features to improve classification accuracy. Specifically, we utilize pre-trained EfficientNetB2 (a specific variant of EfficientNet), a convolutional neural network architecture known for its high performance and efficiency in image feature extraction, to extract deep features from images and FastText, a word embedding model effective in capturing semantic meaning and handling out-of-vocabulary words in text, to extract deep features from text. These models were chosen for their state-of-the-art performance and suitability for our multimodal task. The extracted textual features are then processed and fused with the image features using a convolutional neural network. Experimental results on the Oxford 102 Category Flower dataset demonstrate an accuracy of 97.17 %, representing a significant improvement over existing unimodal and other multimodal state-of-the-art techniques. Similar results were observed on the Oxford 17 Flower dataset. This high accuracy underscores the effectiveness of our multimodal fusion approach and highlights the potential of incorporating Persian textual data for improved flower classification.}, archive = {J_EAAI}, author = {Zohre Karimi and Mohammad Karamad and Sajjad Rahmany}, doi = {10.1016/j.engappai.2025.111642}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111642}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced flower image classification: A multimodal approach leveraging persian texts and deep learning}, volume = {159}, year = {2025}, }
• (2025). Decorrelating structure via adapters makes ensemble learning practical for semi-supervised learning. EAAI, 159, 111641. (www)

  @article{ , abstract = {In Semi-Supervised Learning (SSL), ensemble learning boosts training by enhancing pseudo-label accuracy through multi-perspective prediction integration. This improvement hinges on ensemble framework diversity. However, balancing ensemble cost and diversity is crucial. High costs can impede practicality or efficiency, while low diversity limits performance gains. Our experiments comparing ensemble methods reveal that diversity is influenced by prediction similarity among predictors. Thus, we propose ”Decorrelating Structure via Adapters” (DSA), using distinct adapters to map input features of each prediction head into independent spaces, fostering diversity. As a lightweight structure, DSA effectively decorrelates without substantially increasing parameters or relying on extra loss functions. We analyze DSA’s mechanism, explaining how adapters reduce prediction similarity. Moreover, we showcase DSA’s significant improvements in semi-supervised classification and pose estimation tasks. The code is publicly accessible at: https://github.com/Qi2019KB/DSA .}, archive = {J_EAAI}, author = {Jiaqi Wu and Junbiao Pang and Qingming Huang}, doi = {10.1016/j.engappai.2025.111641}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111641}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Decorrelating structure via adapters makes ensemble learning practical for semi-supervised learning}, volume = {159}, year = {2025}, }
• (2025). Ensemble learning for short circuit fault location estimation in distribution networks. EAAI, 159, 111640. (www)

  @article{ , abstract = {One of the most challenging tasks in power system operation is finding the exact location of a short circuit fault especially in distribution networks with branched structure. In this article a novel neural network ensemble model-based methodology is presented for fault location determination, which combines fault type classification, fault section identification and fault location estimation, and uses only 3 phase V-I measurements from a single sending end monitoring point as inputs. Two ensemble modelling paradigms are considered, namely, Neural Network (Multilayer Perceptron) Ensembles (NNE) and Random Forests (RF). Several different ensemble learning based structures are created using the proposed models and evaluated for fault location estimation on the IEEE-34 feeder benchmark. The average and maximum prediction errors under different fault conditions are used as performance metrics. The results for both the RF and stacked ensemble methods demonstrate that combining predictions enhances overall performance. State-of-the-art performance is achieved with a confidence-weighted stacked NNE-RF ensemble model.}, archive = {J_EAAI}, author = {Hatice Okumus and Fatih M. Nuroglu and Seán McLoone}, doi = {10.1016/j.engappai.2025.111640}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111640}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ensemble learning for short circuit fault location estimation in distribution networks}, volume = {159}, year = {2025}, }
• (2025). Prognosticate pulmonary pathosis for COVID negative and post-acute COVID patients using chest computed tomography images. EAAI, 159, 111639. (www)

  @article{ , abstract = {A significant number of studies have omitted data regarding fatalities (6 months–2 year) after the recovery from Corona Virus Diseases (COVID). Post-COVID, or long-term COVID, refers to the enduring consequences that individuals who have recovered from COVID-19 commonly suffer. People with chronic lung disorders are more likely to die as the infection progresses rapidly. Chest Computed Tomography (CT) images were used to identify the lung abnormalities and determine the patient's exact condition. The proposed method groups COVID-19-negative patients by chronic lung diseases, post-COVID lung disorders, and severity using an improved mask regional-convolutional neural network (R-CNN) to analyze chest CT scan images. Generate synthetic image with a cycle-consistency generative adversarial network (Cycle-GAN) to avoid overfitting. Enhanced Mask R-CNN using ResNet-101 by incorporating FPN model classifies COVID-negative patients' abnormal lung conditions as chronic or post-COVID disorder. This model achieves an accuracy of 95.71 %, F1 score of 94.21 %, a mean average precision (mAP) of 92.34 %, and a geometric mean (G-mean) of 94.89 %. Further post-COVID disorder can be classified into mild (structural abnormalities) or severe (fibrosis). This model had an accuracy of 90.38 % without using cycle-GAN and 94.35 % of accuracy by applying cycle GAN to generate synthetic images which balances the dataset for the severity classification of post-COVID disorder. It achieves the p-value as 0.0004 where p < 0.01 shows that the augmented dataset showed significantly higher performance. This method helps radiologists diagnose chronic lung disease or post-COVID disorder, which enables them to provide appropriate and effective treatments.}, archive = {J_EAAI}, author = {D. Suganya and R. Kalpana}, doi = {10.1016/j.engappai.2025.111639}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111639}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prognosticate pulmonary pathosis for COVID negative and post-acute COVID patients using chest computed tomography images}, volume = {159}, year = {2025}, }
• (2025). Semantic shapley-based counterfactual explanations for end-to-end autonomous driving. EAAI, 159, 111638. (www)

  @article{ , abstract = {Deep learning-based computer vision models are increasingly being used in safety-critical domains, such as autonomous driving, which has led to concerns about the trustworthiness and explainability of these models. Recently, synthetic visual counterfactual explanations have been proposed to uncover the decision-making processes of these models. Building upon this foundation, This study introduces a novel two-stage semantic Shapley-based counterfactual explanation (SSCE) scheme to generate explanations for significant semantics in complex scenarios. In the first stage, the semantic Shapley of the input image is generated to identify the most critical semantic regions. In the second stage, a counterfactual explanation of critical semantics is produced using current synthetic counterfactual explanation methods. Extensive experiments on an end-to-end autonomous driving decision model validate the ability of SSCE to provide realistic, sparse, and proximate counterfactual explanations in complex scenarios.}, archive = {J_EAAI}, author = {Hengyang Sun and Meng Li and Zhihao Cui and Yanjun Huang and Hong Chen}, doi = {10.1016/j.engappai.2025.111638}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111638}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semantic shapley-based counterfactual explanations for end-to-end autonomous driving}, volume = {159}, year = {2025}, }
• (2025). An artificial intelligence maturity assessment framework based on international standards. EAAI, 159, 111637. (www)

  @article{ , abstract = {In an era dominated by technological integration, Artificial Intelligence (AI) is pivotal across various sectors, driving significant advancements and demanding robust quality measures for its implementations. This paper introduces a novel AI maturity assessment framework designed in alignment with International Standards provided by ISO (International Organization for Standardization) and IEC (International Electrotechnical Commission), specifically with the ISO/IEC 33000 family of standards for process assessment. Our framework aims to provide AI system developers with a structured tool for continuous improvement of their development processes, thereby enhancing the reliability and efficacy of AI applications. To demonstrate the applicability of our proposed framework, we have validated it through a case study in the automotive sector. Specifically, the framework was employed to assess and enhance an AI project to develop a mechanism to determine vehicle behavior from sensor data within the constraints of onboard devices. Our findings identify key improvement points contributing to the iterative enhancement of AI system quality in engineering applications.}, archive = {J_EAAI}, author = {Rubén Márquez and Moisés Rodriguez and Javier Verdugo and Francisco P. Romero and Mario Piattini}, doi = {10.1016/j.engappai.2025.111637}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111637}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An artificial intelligence maturity assessment framework based on international standards}, volume = {159}, year = {2025}, }
• (2025). Overcoming data scarcity: A transfer learning framework with fine-tuned neural networks and multi-sensor satellite image fusion for soil moisture estimation. EAAI, 159, 111636. (www)

  @article{ , abstract = {Training deep learning (DL) models requires extensive data, particularly for soil moisture prediction, where large volumes of in situ measurements are needed to prevent overfitting. To address this challenge, we propose a customised transfer learning framework that adapts a pre-trained DL model to a new study site with a different climate. Specifically, we fine-tune a fully connected feed-forward neural network, originally trained on a large dataset from a humid subtropical region (source domain), using limited data from a semi-arid region (target domain). The proposed framework leverages nine input features extracted and generated from Sentinel-1/2 and Shuttle Radar Topographic Mission (SRTM) images through a linear data fusion technique. We systematically evaluate the performance of the proposed framework against ten benchmark algorithms. We observed that the proposed framework outperforms all benchmark algorithms, achieving a correlation coefficient (R) of 0.81, a root mean square error (RMSE) of 0.05 m 3 / m 3 , and a bias of 0.02 m 3 / m 3 on the target domain. Particularly, this is achieved using 55% less in situ data compared to the source domain. To ensure reliability and robustness, we conduct comprehensive analyses, including error histogram, residual, uncertainty, spatial distribution, ablation, statistical, and complex time complexity analyses. Throughout each evaluation, the proposed framework consistently exhibits a reliable and robust performance. The findings of this study hold promise in facilitating accurate surface soil moisture estimation, particularly in data-scarce regions.}, archive = {J_EAAI}, author = {Abhilash Singh and M. Niranjannaik and Kumar Gaurav}, doi = {10.1016/j.engappai.2025.111636}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111636}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Overcoming data scarcity: A transfer learning framework with fine-tuned neural networks and multi-sensor satellite image fusion for soil moisture estimation}, volume = {159}, year = {2025}, }
• (2025). Efficient dehazing network based on mix structure for single image with uneven haze distribution. EAAI, 159, 111635. (www)

  @article{ , abstract = {Although the defogging algorithm based on convolutional neural networks has made significant progress on synthetic uniform foggy datasets, it still exhibits subpar performance on real non-uniform foggy images. In recent years, the transformer network has been applied in the field of image dehazing and has achieved good results in removing haze from non-uniform hazy images. However, two main issues remain: The neglect of the multi-scale characteristics of the image; And the lack of effective strategies to better combine the convolutional structure with the transformer. In this paper, we propose an efficient image dehazing network framework based on a novel hybrid structure. Specifically, the mixed structure block consists of a convolutional component utilizing a parallel attention mechanism and a Transformer architecture. This design effectively captures extensive areas of blur while simultaneously restoring texture details. Additionally, it takes into account the uneven distribution of haze, thereby addressing the challenges associated with removing uneven fog in single images more effectively.Meanwhile, we propose a fusion structure that comprises a skip branch and a main branch, enabling dynamic adjustment of the receptive field size and selection of the appropriate convolution kernel. The experimental results demonstrate that the dehazing algorithm proposed by us outperforms existing methods in terms of dehazing performance.}, archive = {J_EAAI}, author = {Kangle Yuan and Jianguo Wei and Wenhuan Lu}, doi = {10.1016/j.engappai.2025.111635}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111635}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient dehazing network based on mix structure for single image with uneven haze distribution}, volume = {159}, year = {2025}, }
• (2025). Development of intelligent equipment for weed identification and variable spraying in lettuce fields based on instance segmentation framework. EAAI, 159, 111634. (www)

  @article{ , abstract = {Weeds in the field compete with crops for nutrients, water and sunlight, hindering the early growth of crops. If not controlled in time, weeds may adversely affect crop growth and yield. Although chemical weed control is low cost, efficient and widely applicable, excessive use of chemical agents may lead to herbicide residues and environmental pollution. In this study, an instance segmentation-based intelligent equipment was developed for weed recognition and targeted variable-rate spraying in lettuce fields. The You-Only-Look-Once version 8 segmentation (YOLOv8-seg) model was optimized through three key enhancements. Initially, Depthwise Separable Convolution (DSConv) was adopted to replace standard convolutional layers, effectively reducing model complexity, and improving computational efficiency. After that, a novel Faster Implementation of Cross Stage Partial Bottleneck with 2 Convolutions-Star shaped Convolutional (C2f_Star) module was proposed, which integrated the StarBlock from the Star-shaped Convolutional Neural Network (StarNet) into the existing structure, thereby enhancing the feature extraction capabilities of the model. Finally, the Simple Attention Module (SimAM), a parameter-free attention mechanism, was introduced to improve the model's attention to relevant features without increasing the number of parameters. These improvements led to the development of the YOLOv8n-seg model, which achieved a mean Average Precision (mAP) of 90.15 % at 0.5 Intersection over Union (IoU), with 2,281,702 parameters and an inference speed of 15.7 ms per frame. Compared with the original model, the average precision and inference speed increased by 2.65 % and 4.3 %, respectively, while the number of parameters was reduced by 30 %. By combining this model with post-processing algorithms, a precision variable spraying algorithm and equipment were developed. Laboratory experiments at three different weed density levels demonstrated that the system achieved an average recognition accuracy of 95.2 % and a target spraying success rate of 97.2 % for weeds in lettuce fields. Herbicide dosage was reduced by 88.42 %, 65.25 %, and 37.30 % at the three density levels, respectively. This research provides essential theoretical and technical support for the development of precision spraying and weeding robots.}, archive = {J_EAAI}, author = {Long-Tao Niu and Wen-Hao Su and He-Yi Zhang and Qi Wang and Bo-Wen Dong and Yankun Peng}, doi = {10.1016/j.engappai.2025.111634}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111634}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Development of intelligent equipment for weed identification and variable spraying in lettuce fields based on instance segmentation framework}, volume = {159}, year = {2025}, }
• (2025). A novel compositional zero-shot learning approach based on hierarchical multi-scale feature fusion. EAAI, 159, 111633. (www)

  @article{ , abstract = {Compositional Zero-Shot Learning (CZSL) aims to recognize novel combinations of attributes and objects with pre-existing concepts. Most of existing CZSL methods encounter substantial challenges in managing the complex interplay between attributes and objects, particularly when compositions vary in subtle visual details or scale. Inspired by the point-wise convolution, in this work, we propose a novel Hierarchical Multi-Scale Feature Fusion approach for the compositional zero-shot learning task. The proposed CZSL approach incorporates a patch-aware feature selection mechanism to select informative patches from images, enhancing the model’s ability to capture fine-grained details. Subsequently, we design a hierarchical multi-scale feature fusion strategy that combines visual features from multiple scales, allowing the model to integrate local and global information effectively. The proposed feature fusion strategy works by enhancing the model’s ability to disentangle attributes and objects, thereby facilitating improved recognition of novel compositions. The proposed feature fusion strategy works by enhancing the model’s ability to disentangle attributes and objects for improved recognition of novel compositions. Through extensive experiments on standard CZSL benchmark datasets, the proposed approach demonstrates significant improvement over other state-of-the-art methods in both open-world and closed-world CZSL scenarios. This study not only improves accuracy and robustness in the compositional zero-shot learning task but also provides solutions for complex visual tasks in image understanding and robotics, promoting Artificial Intelligence development in visual data and semantic understanding.}, archive = {J_EAAI}, author = {Wenlong Du and Xianglin Bao and Wei Zhao and Xiaofeng Xu and Xingyu Lu and Ruiheng Zhang}, doi = {10.1016/j.engappai.2025.111633}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111633}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel compositional zero-shot learning approach based on hierarchical multi-scale feature fusion}, volume = {159}, year = {2025}, }
• (2025). Adaptive deformable convolutional neural network framework for depression-related behavioral analysis in mice. EAAI, 159, 111632. (www)

  @article{ , abstract = {The use of approximately 1 billion laboratory animals annually in research highlights the urgent need for advanced methods to analyze behavioral dynamics, particularly in mice. Capturing subtle and prolonged behavioral changes, such as those observed in long-term depression studies, poses a significant challenge. To address this, we propose an Adaptive Deformable Convolutional Neural Network Framework for depression-related behavioral analysis in mice. By integrating DeepLabCut (DLC) with deformable convolutional networks (DCN) and convolutional block attention module (CBAM), the framework captures subtle and prolonged behavioral changes with high precision. Adaptive image deformation encodes joint movements into image representations, enabling robust analysis of spatial and temporal patterns. In depression modeling experiment, the framework achieved over 80% classification accuracy, demonstrating its scalability and efficiency. This non-invasive, automated solution represents a transformative advancement in behavioral analysis, offering a reliable tool for long-term studies in animal models.}, archive = {J_EAAI}, author = {Jian Li and Ziyi Li and Peng Shan and Xiaoyong Lyu and Yu Tian and Chen Du and Ying Wang and Yuliang Zhao}, doi = {10.1016/j.engappai.2025.111632}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111632}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive deformable convolutional neural network framework for depression-related behavioral analysis in mice}, volume = {159}, year = {2025}, }
• (2025). Multiobjective evolutionary algorithm based wrapper approach for hyperspectral band selection. EAAI, 159, 111631. (www)

  @article{ , abstract = {A hyperspectral sensor captures information from a wide range of spectral wavelengths, but the information collected is typically highly correlated. It is often challenging to obtain the pertinent bands without degrading the information content. The present work suggests a wrapper approach consisting of a decomposition-based multiobjective evolutionary algorithm. A simultaneous search is suggested for identifying significant bands and hyperparameter value of the classifier as efficacy of proposed approach is influenced by underlying classifier performance. A power distribution-based mechanism is suggested to choose and generate candidate solutions. The hyperspectral band selection problem is formulated as tri-objective optimization problem with information entropy, the percentage in band reduction, and classification accuracy as the objective functions. Entropy is employed as an objective function as a band subset with a higher entropy value can perform better in classification than another band subset with the same size. The assessment of the proposed approach on five widely referenced hyperspectral datasets demonstrates its effectiveness for band reduction while obtaining significant classification accuracy. Furthermore, the suggested approach outperforms other evolutionary multiobjective optimization strategies in obtaining fewer bands with better spectral information.}, archive = {J_EAAI}, author = {Kamal Deep and Manoj Thakur}, doi = {10.1016/j.engappai.2025.111631}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111631}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multiobjective evolutionary algorithm based wrapper approach for hyperspectral band selection}, volume = {159}, year = {2025}, }
• (2025). Utilizing neural networks to illustrate the dynamics of viscous fluid flow over curved surface with homogeneous and heterogeneous reactions. EAAI, 159, 111629. (www)

  @article{ , abstract = {This study examines the influence of homogeneous-heterogeneous reactions and viscous dissipation on the magnetohydrodynamic (MHD) boundary layer flow over a curved stretching sheet, incorporating the effects of partial slip and a non-uniform heat source. Understanding these interactions is crucial for optimizing heat and mass transfer in industrial applications where precise thermal and solutal control are required. The governing partial differential equations are transformed into a system of coupled ordinary differential equations using similarity transformations and solved numerically via the Runge-Kutta method with a shooting technique. A comparative analysis with existing studies further validates the accuracy of the present findings, providing strength into flow control mechanisms and heat transfer enhancement strategies relevant to industrial thermal systems. Moreover, results indicate that increasing the magnetic field parameter increases the shear rate by 65.62 %, whereas thermal dissipation reduces the heat transfer rate by 13.63 %. Additionally, an Artificial Neural Network (ANN) model is employed to predict drag force, heat transfer, and mass transfer rates, achieving a validation accuracy exceeding 99 % with a mean squared error (MSE) of approximately 10 −11 and a regression coefficient ( R ) close to 1 for each case. Moreover, the inputs for predicting drag force are provided to the ANN by varying the values of curvature parameter ( K ( 1 − 5 ) ) , magnetic parameter ( M ( 1 − 3 ) ) and slip parameter ( λ 1 ( 0.1 − 0.3 ) ) , the heat transfer rate controlled by appropriately adjusting the parameters K ( 1 − 5 ) , M ( 1 − 3 ) , heat source parameters ( A ∗ ( 0.5 − 1.5 ) , B ∗ ( 0.1 − 0.4 ) ) and Eckert number ( E c ( 0.1 − 0.3 ) ) and the solutal rate is determined by adjusting the parameters K ( 1 − 5 ) , Schimdt number ( S c ( 3 − 6 ) ) , homogeneous and heterogeneous reaction parameters ( k 1 & k 2 ( 0.1 − 0.25 ) ) . The ANN predictions show strong agreement with numerical results, confirming the reliability of the model for real-world thermal engineering applications.}, archive = {J_EAAI}, author = {Abhishek Sharma and Ram Prakash Sharma}, doi = {10.1016/j.engappai.2025.111629}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111629}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Utilizing neural networks to illustrate the dynamics of viscous fluid flow over curved surface with homogeneous and heterogeneous reactions}, volume = {159}, year = {2025}, }
• (2025). ForecaState: Framework for industrial internet of things state forecasting using recurrent neural networks with hyperparameters optimization. EAAI, 159, 111627. (www)

  @article{ , abstract = {In the Industrial Internet of Things (IIoT), data analytics plays an important role in the management and planning of production processes. In particular, forecasting future states helps to ensure the operation of the equipment, optimize its performance and detect anomalies and failures in a timely manner. We propose a framework called ForecaState for IIoT state forecasting. ForecaState utilizes variations of recurrent neural networks (RNNs), including long short-term memory (LSTM) and gated recurrent units (GRUs), to capture complex patterns and dependencies in industrial data. These advanced RNN architectures have proven to be highly effective in capturing complex temporal dependencies and patterns in data. We also integrate popular optimization techniques such as grid search, random search, Bayesian optimization, and HyperBand algorithms into our framework to fine-tune the hyperparameters of RNN models. This ensures that models can extract the most relevant information from the data and provide accurate predictions. We evaluate our framework for short-term and long-term forecasting problems using the SWaT (Safe Water Treatment) and ETT (Electricity Transformer Temperature) datasets. These datasets reflect industrial scenarios and serve as a benchmark for comparing different forecasting approaches. The experimental results show that the proposed framework is not inferior or even superior to relevant approaches in terms of the quality of industrial data forecasting. We achieved an improvement in the quality of forecasting by more than 78% compared to the analogs to predict the parameters of the following production processes: water purification and electricity supply.}, archive = {J_EAAI}, author = {Diana Levshun and Dmitry Levshun and Igor Kotenko}, doi = {10.1016/j.engappai.2025.111627}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111627}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ForecaState: Framework for industrial internet of things state forecasting using recurrent neural networks with hyperparameters optimization}, volume = {159}, year = {2025}, }
• (2025). Harmonizing sustainability and affordability in desalination: A disc spherical fuzzy weighted aggregated sum product assessment approach. EAAI, 159, 111626. (www)

  @article{ , abstract = {Addressing the escalating global water demands requires efficient decision-making to ensure fair access and environmental preservation amid increasing demand and climate uncertainties. Desalination emerges as a pivotal solution to combat water scarcity, especially in regions grappling with persistent water shortages. In this context, effective decision-making becomes imperative for selecting sustainable desalination practices, considering diverse technological, environmental, and economic factors. Therefore, this research aims to develop an improved decision-making framework for sustainable desalination through Dombi operations joined with the disc spherical fuzzy (D-SF) arena. The D-SF environment receives an extended weighted aggregated sum product assessment (WASPAS) methodology for decision-making purposes together with the maximizing deviation method (MDM) method for weighing attributes. The new decision-making system improves reliability through weighted averaging and weighted geometric aggregation in D-SF Dombi operations to handle uncertain data more effectively. The proposed framework undergoes case-study evaluation to demonstrate its capacity for developing sustainable desalination solutions between managing technological aspects with environmental considerations and cost control. A comparative analysis showcases the reliability of the proposed D-SF WASPAS-MDM methodology. The proposed model maintains stability according to its sensitivity testing process which proves its ability to deliver dependable decisions during uncertain situations. The D-SF-driven framework maximizes the optimization of the desalination system by its ability to synthesize technological and economic with environmental considerations. Such a cooperative approach ensures that technological advancements meet both environmental criteria and economic demands, resulting in improved efficiency as well as ecologically friendly desalination operations.}, archive = {J_EAAI}, author = {Shahzaib Ashraf and Muhammad Shazib Hameed and Wania Iqbal and Vladimir Simic and Serhat Aydin and Dragan Pamucar and Nebojsa Bacanin}, doi = {10.1016/j.engappai.2025.111626}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111626}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Harmonizing sustainability and affordability in desalination: A disc spherical fuzzy weighted aggregated sum product assessment approach}, volume = {159}, year = {2025}, }
• (2025). Knowledge graph reasoning: Mainstream methods, applications and prospects. EAAI, 159, 111625. (www)

  @article{ , abstract = {Knowledge graphs have emerged as a leading paradigm for knowledge representation due to their robust capacity to mine, organize, and manage massive datasets. Their versatility has led to extensive research and application in various advanced fields. Knowledge graph reasoning plays a critical role in this context by reasoning new facts from existing ones, thereby completing and refining the knowledge base. In this paper, we provide a comprehensive review of knowledge graph reasoning by researching its definition, foundational concepts, and methodological approaches. We systematically categorize reasoning methods into five groups: reasoning based on ontology, reasoning based on rules, neural rule reasoning based on distributed representations, neural rule reasoning based on deep learning, and hybrid reasoning. Furthermore, we analyze the applications of knowledge graph reasoning, discuss their current limitations, and outline promising directions for future research to enhance both performance and scalability in this rapidly evolving field.}, archive = {J_EAAI}, author = {Han Liu and Shaojie Yang and Guokai Shi and Zongcheng Miao}, doi = {10.1016/j.engappai.2025.111625}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111625}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Knowledge graph reasoning: Mainstream methods, applications and prospects}, volume = {159}, year = {2025}, }
• (2025). Transfer learning-based approach for evaluating residual stiffness in carbon fiber reinforced composites and adhesives. EAAI, 159, 111624. (www)

  @article{ , abstract = {This work proposes a transfer learning-based encoder-decoder framework to predict the relationship between loading conditions and residual stiffness in carbon fiber reinforced composites and adhesives. The encoder, built from a Convolutional Neural Network (CNN) and Bidirectional Long Short-term Memory (Bi-LSTM), extracts time-series loading signals into latent variables and captures their dependencies. The decoder employs a Multilayer Perceptron (MLP) to map these latent features to residual stiffness. Transfer learning strategy is used to account for individual variability and further improve accuracy. The model's effectiveness and robustness are validated through random and constant loading fatigue experiments from two different material systems. Under random fatigue data, the model demonstrates strong learning capabilities. Under random fatigue data, the model demonstrates strong learning capabilities. Compared to classical models like Support Vector Machine (SVM) and Random Forest, or simpler deep learning architectures like individual CNN and Bi-LSTM networks, the proposed architecture shows enhanced prediction accuracy and regression results, achieving a Root Mean Square Error (RMSE) of 0.154 and a Coefficient of Determination (R 2 ) of 0.931. In constant amplitude fatigue datasets, the model accurately identifies different materials and exhibits satisfactory robustness when reasonable training dataset size is used.}, archive = {J_EAAI}, author = {Licai Cao and Tianxiao Zhang and Jin Cui and Anastasios P. Vassilopoulos}, doi = {10.1016/j.engappai.2025.111624}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111624}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transfer learning-based approach for evaluating residual stiffness in carbon fiber reinforced composites and adhesives}, volume = {159}, year = {2025}, }
• (2025). Artificial emotion-assisted data and modeling dual-drive methods for voltage control in renewable energy enriched grids. EAAI, 159, 111623. (www)

  @article{ , abstract = {The conventionally automated voltage control framework lacks a consistent time scale at all levels and is prone to issues such as inconsistent scheduling goals and inverse regulation. To address the issue of non-uniform time-scale voltage control at each level, a harmonized first-level voltage control (FVC) framework with a consistent time scale is proposed. This study proposes a data and modeling dual-drive methods named as AEs-TNNs-DDPG-PID (AETDP) which is a unified method combining deep deterministic policy gradient (DDPG), Transformer neural networks (TNNs), artificial emotions (AEs), and proportional-integral-derivative (PID). The AETDP is applied in a harmonized FVC framework to achieve smart grids (SGs) voltage control. The TNNs have superior long-term dependent modeling and parallel processing capabilities in the voltage prediction tasks of smart grids attributed to the unique self-attention layer. The DDPG realizes the parallel operation of affective and cognitive processing by incorporating AEs, which improves the responsiveness and processing efficiency of the system. Emotional and cognitive modules complement each other and work together to enable agents to function in complex environments with flexibility and humane treatment. AETDP is simulated in IEEE 39-bus, 300-bus, and 2736-bus power systems to assess the feasibility and reliability of AETDP. The integral of error squared for AETDP is 36.06 % less than PID and 1.35 % less than DDPG-PID.}, archive = {J_EAAI}, author = {Linfei Yin and Hongzhen Wu}, doi = {10.1016/j.engappai.2025.111623}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111623}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial emotion-assisted data and modeling dual-drive methods for voltage control in renewable energy enriched grids}, volume = {159}, year = {2025}, }
• (2025). Building explainable artificial intelligence for reinforcement learning based debt collection recommender system using large language models. EAAI, 159, 111622. (www)

  @article{ , abstract = {Managing debt collections can be a challenging task. Managing the collection process and customer details requires human effort and time. We built a Reinforcement Learning (RL) based recommendation system to optimize this collection process. The model has to generate suggestions for boosting the collection rate, thereby saving time for manual tasks. The suitable actions were developed according to customers' risk levels and debt collector performance. The reinforcement Learning algorithm works based on the principle of exploration and exploitation. Initially, the model generates the recommendations based on the fixed policies for observed cases. When the model encounters a new scenario during the explorations, it has to create the recommendations by itself. It is necessary to understand why the model generated the particular recommendation. To build such an explainable recommendation system, we used Google Vertex Artificial Intelligence (AI) to generate text-based explanations for decisions made by the model. Moreover, we compared the different large language models through different prompting methods. We used prompt engineering concepts to get the desired text explanations. Bilingual Evaluation Understudy (BLEU) and Recall Oriented Understudy For Gisting Evaluation (ROUGE) were the accuracy metrics used in this work to analyze the performance of the text generation. Our contribution not only advances interpretability in the debt collection process but can also be implemented in various real-world settings such as healthcare, e-commerce recommendations, etc. The proposed method achieved a higher collection rate in three risk categories, such as very low risk, low risk, and medium risk.}, archive = {J_EAAI}, author = {Keerthana Sivamayilvelan and Elakkiya Rajasekar and Subramaniyaswamy Vairavasundaram and Santhi Balachandran and Vishnu Suresh}, doi = {10.1016/j.engappai.2025.111622}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111622}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Building explainable artificial intelligence for reinforcement learning based debt collection recommender system using large language models}, volume = {159}, year = {2025}, }
• (2025). Adversarial environment design for crowd navigation based on deep reinforcement learning. EAAI, 159, 111621. (www)

  @article{ , abstract = {The widespread use of mobile robots has increased the shared space between humans and robots, necessitating advanced solutions for crowd navigation. Recent studies have proposed approaches based on deep reinforcement learning to safely and efficiently achieve this goal. However, these approaches face challenges such as difficulty in presenting diverse pedestrian patterns and limited generalization performance. This study proposes a framework called Simultaneous training Process with Adversarial Crowd Environment (SPACE), which is an implemented artificial intelligence that generates crowd navigation environments. This framework competitively trains a crowd navigation agent and an adversarial crowd environment. In the adversarial crowd environment, the adversarial agent places pedestrians to induce collisions with the crowd navigation agent. By applying artificial intelligence within the episode-generation, this framework addresses vulnerabilities of previous approaches and allows the training of robust crowd navigation agents with high generalization performance. Experimental results demonstrate up to a 24.62% increase in navigation success rate and a 41.6% improvement in minimum distance from pedestrians compared to agents trained in non-adversarial environments, ensuring safer crowd navigation. Furthermore, SPACE exhibits more stable navigation performance in evaluation environment settings that are significantly more complex than the training scenarios. These findings highlight the promise of SPACE for training crowd navigation agents capable of operating effectively under diverse environmental conditions.}, archive = {J_EAAI}, author = {Jeongeun Kim and Hyo-Seok Hwang and Junhee Seok}, doi = {10.1016/j.engappai.2025.111621}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111621}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adversarial environment design for crowd navigation based on deep reinforcement learning}, volume = {159}, year = {2025}, }
• (2025). History matching of a three dimensional channelized reservoir using a beta-convolutional variational autoencoder and ensemble smoother with multiple data assimilation. EAAI, 159, 111620. (www)

  @article{ , abstract = {Three-dimensional channelized reservoir characterization is challenging because complex channel patterns, vertical interaction of reservoir properties, and high computational costs make it hard to carry out inverse calculations. This paper proposes a novel deep learning-based history-matching scheme for three-dimensional channelized reservoir characterization. The study's scheme was divided into three stages. First, a beta-convolutional variational autoencoder network was trained using rock facies of reservoir models. Second, latent vectors of the beta-convolutional variational autoencoder network were updated using an ensemble smoother with multiple data assimilation. Third, permeability values associated with rock facies were acquired by decoding the updated latent vectors. Two multi-layered channelized reservoir model datasets are presented as case studies, one of inclined reservoir models and the other of reservoir models with vertically varying channels. The simulated dynamic behavior of the proposed scheme was shown to be in good agreement with the observed behavior in terms of oil and water production rates based on accurate matching results in channel connectivity and permeability distributions for both cases.}, archive = {J_EAAI}, author = {Youngbin Ahn and Jonggeun Choe and Baehyun Min}, doi = {10.1016/j.engappai.2025.111620}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111620}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {History matching of a three dimensional channelized reservoir using a beta-convolutional variational autoencoder and ensemble smoother with multiple data assimilation}, volume = {159}, year = {2025}, }
• (2025). An innovative contrastive learning approach to improve image recognition robustness and interpretability via simulated environmental perturbations. EAAI, 159, 111619. (www)

  @article{ , abstract = {In the field of pattern recognition, the noise inherent in real-world images poses a significant challenge to traditional image processing methodologies. While existing approaches have made progress in addressing this issue, they often struggle with limited model generalization, data distribution shifts, and domain adaptability discrepancies between simulated environments and real-world contexts, compromising efficiency and robustness. In this paper, we propose a novel contrastive learning strategy for Enhancing Robustness and Interpretability in Image Recognition through Environmental Perturbations (ERIEP) of clear-featured image data. ERIEP meticulously identifies a set of core visual features, termed “invariant features”, which can offer optimal explanations for image predictions. Concurrently, it emphasizes learning noise-resistant strategies to amplify the model’s interpretability. Through ERIEP’s contrastive learning approach, we address complex images, enabling the model to progressively refine its understanding of both the invariant features and noise mitigation technique. Our extensive experiments on CIFAR-10, CIFAR-100, and ImageNet-1K demonstrate that ERIEP significantly outperforms several state-of-the-art image-processing baselines, showing robust performance under various noise intensities and environmental perturbations.}, archive = {J_EAAI}, author = {Leijun Cheng and Xihe Qiu and Xiaoyu Tan and Haoyu Wang and Yujie Xiong}, doi = {10.1016/j.engappai.2025.111619}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111619}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An innovative contrastive learning approach to improve image recognition robustness and interpretability via simulated environmental perturbations}, volume = {159}, year = {2025}, }
• (2025). LUNet: Lightweight U-net with adaptive scaling for devanagari handwritten word segmentation. EAAI, 159, 111618. (www)

  @article{ , abstract = {Word segmentation from handwritten documents is a fundamental task in document analysis, facilitating a range of applications such as digitization, writer identification, topic modeling, and information retrieval. However, handwritten scripts often present unique challenges, including inconsistent spacing, overlapping elements, and skewed text, which make segmentation a non-trivial task. Among these, the Devanagari script stands out due to its distinct characteristics, such as horizontal lines above characters and varying inter-word spacing, which add further complexity to the segmentation process. To address these challenges, this study focuses on developing a robust segmentation method for handwritten Devanagari text. The proposed approach employs a U-Net-based deep learning model to achieve precise word segmentation, enhanced with pre-processing and post-processing steps. Experimental results demonstrate an accuracy of 95.52%, highlighting the model’s effectiveness in tackling the intricacies of Devanagari scripts. This work not only contributes to advancing the segmentation of handwritten Devanagari text but also serves as a foundation for broader applications in document analysis and language processing.}, archive = {J_EAAI}, author = {Rajandeep Kaur and Rajneesh Rani and Roop Pahuja}, doi = {10.1016/j.engappai.2025.111618}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111618}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {LUNet: Lightweight U-net with adaptive scaling for devanagari handwritten word segmentation}, volume = {159}, year = {2025}, }
• (2025). Smart artificial pancreas: Sensor-based glucose and insulin control by deep stochastic policy learning. EAAI, 159, 111616. (www)

  @article{ , abstract = {Objectives: This study presents a novel closed-loop Artificial Pancreas System (cl-PAS) intended to adjust blood glucose (BG) levels and insulin dosage in diabetic patients in real-time. Methods and Novelty: To achieve this, a nonlinear active disturbance rejection controller (NADRC) has been created to control BG levels by managing insulin concentrations while simultaneously eliminating outside uncertainties and unidentified input dynamics. A deep stochastic policy optimization (DSPO) has been incorporated into the main system in the next step to offer real-time updates of the coefficients of the suggested controller and enhance the system’s resilience and adaptability. A second deep neural network serves as the value function in the implemented strategy, estimating the expected return for a given state–action pair. The policy function in the strategy translates states to actions. DSPO is able to train effective control policies in high-dimensional, continuous state and action spaces, especially physiological systems, by updating the value networks and the policy simultaneously. Findings: Many tests are performed to determine whether the developed approach is functional in managing the insulin infusion rate to regulate BG under various conditions. These tests include: (1) Normal condition, (2) robustness analysis (applying to new cases), (3) condition of multiple meal consumption, and (4) Conditions involving uncertainties and disruptions. Comparative evaluations and statistical analyses with alternative and prior methodologies demonstrate the average improvement of 64.90% for patient 1, 59.49% for patient 2, and 64.17% for patient 3 over other approaches.}, archive = {J_EAAI}, author = {Shuguang Li and Shuzhou Han and Mai The Vu}, doi = {10.1016/j.engappai.2025.111616}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111616}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Smart artificial pancreas: Sensor-based glucose and insulin control by deep stochastic policy learning}, volume = {159}, year = {2025}, }
• (2025). Joint domain-adaptive transformer model for bearing remaining useful life prediction across different domains. EAAI, 159, 111615. (www)

  @article{ , abstract = {Rolling bearings are critical components in mechanical systems, and accurately predicting their remaining useful life (RUL) is essential for optimizing maintenance and ensuring safety. While deep learning presents new opportunities for RUL prediction, discrepancies in feature distributions between experimental and real-world data hinder model transfer, and collecting long-cycle industrial data is costly. To address these challenges, we propose the joint domain-adaptive Transformer (JDATransformer), a model tailored to improve the accuracy of cross-domain bearing RUL prediction by introducing global feature distribution alignment within the Transformer framework. Unlike conventional methods that apply domain adaptation at isolated layers, JDATransformer leverages the attention mechanism to capture long-range temporal dependencies and applies joint maximum mean discrepancy to achieve global alignment. This design effectively enhances both domain adaptation and temporal modeling, addressing the two key challenges of distribution shift and degradation pattern learning in bearing RUL prediction. We validated JDATransformer through intra-machine and cross-machine transfer experiments, conducting comprehensive comparisons with both conventional and state-of-the-art methods. On average across all intra-machine and cross-machine transfer experiments, JDATransformer achieved root mean square error reductions of 0.2128 and 0.0626, and mean absolute error reductions of 0.2066 and 0.0524, compared with the baseline method and the single-layer domain adaptation method, respectively. The results demonstrate JDATransformer's consistent superiority across various evaluation metrics. JDATransformer offers a robust solution for accurate and efficient RUL prediction in complex industrial environments, leveraging global alignment and advanced temporal information processing.}, archive = {J_EAAI}, author = {Xin Zhang and Zefang Li and Jiaxu Wang}, doi = {10.1016/j.engappai.2025.111615}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111615}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Joint domain-adaptive transformer model for bearing remaining useful life prediction across different domains}, volume = {159}, year = {2025}, }
• (2025). Interpretable adaptive multiwavelet kernel-driven two-dimensional convolutional neural network for mechanical fault diagnosis. EAAI, 159, 111614. (www)

  @article{ , abstract = {Despite extensive research on convolutional neural networks (CNNs) for intelligent fault diagnosis, several challenges remain, including the limited effectiveness of one-dimensional CNNs, susceptibility to noise, and lack of interpretability. To address these issues, an interpretable adaptive multiwavelet kernel-driven two-dimensional convolutional neural network model called MWKN has been designed for mechanical fault diagnosis in this paper. Specifically, a newly designed adaptive natural convolutional layer based on two-dimensional multiwavelet transform is embedded as a feature extraction module in the shallow layer of a two-dimensional CNN model. Crucially, this novel multiwavelet convolutional layer is jointly optimized with the entire network, enabling the adaptive optimization of its intrinsic multiwavelet convolutional kernel. Additionally, this model incorporates a specifically designed embedded two-dimensional neighboring coefficient shrinkage module to address the issue of CNN susceptibility to strong noise. This study investigates the interpretability of the MWKN model through simulated fault experiments, addressing interpretability deficits observed in CNN. The results demonstrate that the embedded multiwavelet kernel and the inclusive WMKN model as a whole possess synchronized learning and matching rules, substantiating that the learning process of the multiwavelet kernel is neither isolated nor random but follows an inherent adaptive learning mechanism based on the principle of error minimization. Finally, the excellent fault identification capability and robust noise resistance of MWKN are validated through experimental cases of variable speed bearing faults and pump circulation bearing faults under strong noise background. In addition, the industrial applicability and interpretability of the method were further validated in an industrial scenario case.}, archive = {J_EAAI}, author = {Tianheng Hai and Jing Yuan and Huiming Jiang and Qian Zhao}, doi = {10.1016/j.engappai.2025.111614}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111614}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpretable adaptive multiwavelet kernel-driven two-dimensional convolutional neural network for mechanical fault diagnosis}, volume = {159}, year = {2025}, }
• (2025). Deep muscle electromyogram construction using a physics-integrated deep learning approach. EAAI, 159, 111613. (www)

  @article{ , abstract = {Electromyography (EMG)–based computational musculoskeletal modeling is a non-invasive method for studying musculotendon function, human movement, and neuromuscular control, providing estimates of internal variables like muscle forces and joint torques. However, EMG signals from deeper muscles are often challenging to measure by placing the surface EMG electrodes and unfeasible to measure directly using invasive methods. The restriction to the access of EMG data from deeper muscles poses a considerable obstacle to the broad adoption of EMG-driven modeling techniques. A strategic alternative is to use an estimation algorithm to approximate the missing EMG signals from deeper muscle. A similar strategy is used in physics-informed deep learning, where the features of physical systems are learned without labeled data. In this work, we propose a hybrid deep learning algorithm, namely the neural musculoskeletal model (NMM), that integrates physics-informed and data-driven deep learning to approximate the EMG signals from the deeper muscles. While data-driven modeling is used to predict the missing EMG signals, physics-based modeling engraves the subject-specific information into the predictions. Experimental verifications on thirteen test subjects are carried out to investigate the performance of the proposed hybrid framework. The proposed NMM is validated against the joint torque computed from ‘OpenSim’ software. The predicted deep EMG signals are also compared against the state-of-the-art approaches, where the proposed NMM completely outperforms the existing frameworks by a significant margin.}, archive = {J_EAAI}, author = {Rajnish Kumar and Tapas Tripura and Souvik Chakraborty and Sitikantha Roy}, doi = {10.1016/j.engappai.2025.111613}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111613}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep muscle electromyogram construction using a physics-integrated deep learning approach}, volume = {159}, year = {2025}, }
• (2025). A hybrid three-way recommendation considering users variability. EAAI, 159, 111610. (www)

  @article{ , abstract = {Hybrid recommender systems leverage diverse information sources and techniques to enhance performance. Nevertheless, integrating users’ multifaceted preferences remains challenging due to the uneven data. Meanwhile, information insufficiency introduces uncertainty in recommendations while existing strategies (i.e., recommend or not-recommend) lack the flexibility to address it. Additionally, these works mainly overlook that ratings not only reflect preferences but imply users’ attitudes toward the strategies, leading to the same recommendation rule despite users distinctly. To solve these issues, a Hybrid Three-Way Recommender (HTWR) system is proposed to formulate personalized three-way rules. Specifically, users’ historical and predictive preferences are captured via tags and ratings while integrated based on the user’s data distribution. Then, the theory of three-way decision is introduced to address such uncertainty by offering the option of defer-recommend. Finally, the users variability is formally given and incorporated into the loss function to obtain personalized rules. Experiments on three public datasets validate the superiority and flexibility of the proposed HTWR.}, archive = {J_EAAI}, author = {Yu Xie and Jilin Yang and Youlei Meng and Xianyong Zhang}, doi = {10.1016/j.engappai.2025.111610}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111610}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid three-way recommendation considering users variability}, volume = {159}, year = {2025}, }
• (2025). OralTransNet: A novel hybrid model integrating transformer attention and CNN features for accurate diagnosis of mouth and oral diseases. EAAI, 159, 111609. (www)

  @article{ , abstract = {The rising prevalence of mouth and oral diseases (MOD), including gum disease and oral cancer, presents a significant global health challenge. Early detection is crucial for effective intervention. However, existing models often rely on complex preprocessing, computationally expensive operations, and specialized resources, leading to inefficiency and limited practicality. This paper presents a novel lightweight hybrid model that combines the local feature extraction strengths of CNNs with the global contextual power of Transformer attention mechanisms, contributing to the advancement of artificial intelligence (AI) in medical image analysis. The proposed architecture integrates the local feature extraction efficiency of convolutional neural networks (CNNs) with the global context modeling strength of Transformers. This combination enables the model to effectively capture both fine-grained details and broader spatial patterns, while maintaining low computational complexity. By leveraging CNNs' weight-sharing properties for efficient feature extraction and Transformers' ability to model global patterns, the proposed model performs well across datasets of varying sizes and complexities. Its lightweight design emphasizes efficiency, with fewer parameters, reduced floating-point operations (FLOPs), and shorter inference times, making it ideal for real-time AI applications, particularly in resource-constrained environments. The proposed model is also well-suited for deployment on mobile devices and in regions with limited medical infrastructure, providing a scalable solution for early diagnosis in diverse healthcare settings. In the context of medical engineering, the proposed model is applied to the automated detection of mouth and oral diseases (MOD) using both clinical and histopathological images. This approach aims to enhance diagnostic capabilities in resource-constrained clinical environments. The model is rigorously evaluated on three datasets: the MOD dataset (5143 images, 7 classes), the Oral Cancer dataset (241 images, 2 classes), and the Histopathological Oral Cancer dataset (5192 images, 2 classes). The proposed model achieves accuracies of 99.03 %, 97.83 %, and 94.23 %, respectively—surpassing several state-of-the-art (SOTA) models. Its strong performance, lightweight design, and enhanced interpretability position it as a practical and scalable solution for early and reliable oral disease detection in diverse clinical settings.}, archive = {J_EAAI}, author = {Sohaib Asif and Vicky Yang Wang and Dong Xu}, doi = {10.1016/j.engappai.2025.111609}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111609}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {OralTransNet: A novel hybrid model integrating transformer attention and CNN features for accurate diagnosis of mouth and oral diseases}, volume = {159}, year = {2025}, }
• (2025). An attention-guided multi-scale feature cascade network for underwater fish counting. EAAI, 159, 111608. (www)

  @article{ , abstract = {Visual counting is essential for advancing fisheries intelligence, but fish scale variation in open underwater environments has made underwater fish counting a constant challenge. Therefore, we propose an Attention-guided Multi-scale Feature Cascade Network, named AMFCNet, which resolves scale variation and improves the accuracy of fish counting in complex underwater environments. AMFCNet utilizes a multi-scale attention gate for multi-scale feature fusion, and integrates a multi-scale convolution module to capture complex spatial relationships. It also employs a multi-head supervision fusion strategy to mask irrelevant regions, ensuring targeted learning for each scale and generating high-quality multi-scale density maps. Experimental results demonstrate that the proposed method achieves state-of-the-art performance on the proposed dataset with the lowest computational cost, significantly outperforming 11 mainstream counting methods. It also achieves excellent results on other publicly available underwater datasets, with Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Normalized Absolute Error (NAE) values of 1.26, 1.71, and 0.08, respectively. This method shows significant potential for practical applications in aquaculture, such as in marine ranching and pond farming, to assess fish growth conditions and adjust feeding strategies accordingly.}, archive = {J_EAAI}, author = {Hanyu Zhang and Mengping Dong and Fei Li and Zhenbo Li and Ping Hu}, doi = {10.1016/j.engappai.2025.111608}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111608}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An attention-guided multi-scale feature cascade network for underwater fish counting}, volume = {159}, year = {2025}, }
• (2025). A single-demonstration guided manipulation learning with dexterous hand. EAAI, 159, 111606. (www)

  @article{ , abstract = {Demonstration assisted reinforcement learning has been proven to be an extremely effective method for solving complex multi-fingered dexterous hand manipulation tasks. It usually requires costly and time-consuming expert demonstrations for each task, affecting the learning efficiency of the dexterous manipulation policy. To overcome this drawback, this paper devotes to use only one human demonstration per task to obtain a generalizable dexterous manipulation policy. And a novel ‘Basics-before-Extension’ policy learning strategy (BBE) is proposed for this purpose. It consists of two learning stages. In the ‘basics learning stage’, the dexterous hand extracts hand-object contact points as key clues from demonstration, facilitating a quick learning of the expert basic skill. While in ‘extension learning stage’, the designed joint policy training scheme enables the expert knowledge to be transferred and adapted to new environments, outputting generalizable policy. We present the distinctive overall framework from the low-cost demonstration data collection to the policy learning process. Meanwhile, BBE strategy has been experimentally validated on typical grasp and manipulation tasks, including relocating objects, opening door, hammering nails and functional grasp. The results indicate that the proposed BBE strategy can empower the multi-fingered dexterous hand with the intelligence of learning typical grasp and manipulation efficiently and accurately from a single demonstration.}, archive = {J_EAAI}, author = {Jianwen Li and Yinglan Lv and Xiangbo Lin and Jinglue Hang and Xuanheng Li and Yi Sun}, doi = {10.1016/j.engappai.2025.111606}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111606}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A single-demonstration guided manipulation learning with dexterous hand}, volume = {159}, year = {2025}, }
• (2025). Prior-driven refinement network for small organ-at-risk segmentation in head and neck cancer. EAAI, 159, 111605. (www)

  @article{ , abstract = {Accurate segmentation of small organs-at-risk (OARs) in computed tomography (CT) images is crucial for radiotherapy treatment planning in head and neck cancer. However, the low soft tissue contrast, small spatial structures, and the limited training data pose significant challenges for automated segmentation methods. This paper proposes prior-driven refinement network (PRNet), a novel deep learning-based approach that leverages the foundation model’s general-purpose representations and domain-specific knowledge to tackle these challenges. PRNet builds upon the initial coarse segmentation and refines small organs by utilizing the coarse segmentation as prior knowledge. PRNet inherits its architecture from the Segment Anything Model (SAM) but incorporates a novel prior encoder and mask refinement transformer, enabling the fusion of domain-specific knowledge with SAM’s robust representations.The architecture of PRNet is inherited from the Segment Anything Model (SAM), with the addition of the prior encoder and the mask refinement transformer, allowing for the fusion of domain-specific knowledge with SAM’s robust representations. Experiments on three public datasets demonstrate PRNet’s superior performance, with average Dice scores of 75.14% ± 12.81%, 76.56% ± 12.90%, and 82.83 ± 13.49% respectively. These results represent improvements of 3.61%, 3.64%, and 5.14% over current state-of-the-art methods. Moreover, experiments on four diverse datasets demonstrate PRNet’s generalizability across different anatomical regions and imaging modalities, including liver tumors, myocardial pathologies, and thoracic organs. Our proposed method shows potential for improving clinical radiotherapy planning workflows and contributing to more precise treatment delivery in head and neck cancer patients.}, archive = {J_EAAI}, author = {Taibao Wang and Yifan Gao and Bingyu Liang and Qin Wang}, doi = {10.1016/j.engappai.2025.111605}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111605}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prior-driven refinement network for small organ-at-risk segmentation in head and neck cancer}, volume = {159}, year = {2025}, }
• (2025). Contrast gas detection: Improving infrared gas semantic segmentation with static background. EAAI, 159, 111604. (www)

  @article{ , abstract = {Accurate and efficient identification and segmentation of gases play an important role in industrial processes and public safety. Although Optical Gas Imaging (OGI) has proven effective in acquiring images and videos of gas leaks, accurately determining the leakage points and their diffusion areas in real-world conditions remains a significant challenge. To address this, we propose a novel gas segmentation annotation methodology. This approach applies differencing between gas leakage scenes and corresponding static background to delineate precise leakage regions, which are subsequently annotated manually. The resulting dataset, consisting of 926 static background-gas image pairs extracted from 38 comprehensive video sequences, provides a robust foundation for advancing gas segmentation research. Based on this dataset, we present a novel semantic segmentation model specifically designed for the unique characteristics of gas leakage scenarios. Our model integrates a gas contrast attention mechanism to capitalize on static background information, resulting in improved segmentation precision. Comparative evaluations demonstrate that the proposed model achieves state-of-the-art performance on our dataset, outperforming widely adopted semantic segmentation models. All code and the dataset will be made publicly available at https://github.com/ProAlize/CGNet .}, archive = {J_EAAI}, author = {Jue Wang and Jianzhi Fan and Tianshuo Yuan and Dong Luo and Guohua Jiao and Wei Chen}, doi = {10.1016/j.engappai.2025.111604}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111604}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Contrast gas detection: Improving infrared gas semantic segmentation with static background}, volume = {159}, year = {2025}, }
• (2025). Store-and-forward with graph attention: Enhanced multi-agent reinforcement learning for emergency-responsive traffic signal control. EAAI, 159, 111602. (www)

  @article{ , abstract = {This paper investigates the problem of priority access signal control for emergency vehicles (EMVs) in urban road networks. Based on the characteristics of traffic road network, a cooperative multi-agent reinforcement learning model leveraging the store-and-forward paradigm is proposed, as a way to improve emergency vehicle priority and alleviate urban traffic pressure. Initially, we establish cooperative relationships among agents at the network modeling level, modeling the traffic flow between pairs of adjacent agents as a co-optimization objective, which achieves parameter decoupling among algorithms. Subsequently, through regional partitioning, we design a spatiotemporal information extraction algorithm, termed store-and-forward graph attention mechanism for multi-agent proximal policy optimization (SF-GMPPO), to control information exchange among agents, thereby learning optimal traffic signal control policies. Finally, the proposed algorithm model is tested in both synthetic and real-world network environments. Experimental results demonstrate that the proposed algorithm surpasses similar algorithms in terms of training convergence and control performance. Specifically, in the 5x5 synthetic road network, the average speed of EMVs increased by 2.463 meters per second (m/s) compared to social vehicles. In the Lanzhou Chengguan road network with 128 intersections, the proposed algorithm achieved high traffic control efficiency, resulting in the speed of social vehicles approaching that of EMVs, with EMVs operating at an average speed of 13.061 m/s and social vehicles at 13.032 m/s under efficient traffic flow management. These results highlight the effectiveness of the proposed approach in enhancing EMV priority while maintaining overall traffic efficiency.}, archive = {J_EAAI}, author = {Kangkang Yang and Zhiwen Wang and Xinyou Meng and Long Li and Yaoke Shi and Yuling Yu and Haoxu Wang and Ziheng Yao}, doi = {10.1016/j.engappai.2025.111602}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111602}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Store-and-forward with graph attention: Enhanced multi-agent reinforcement learning for emergency-responsive traffic signal control}, volume = {159}, year = {2025}, }
• (2025). A deep learning based iterative denoising algorithm for multiple frequency lines recovery. EAAI, 159, 111601. (www)

  @article{ , abstract = {Passive detection technology constitutes a crucial research direction in underwater acoustic target detection. It has been the subject of ongoing investigations to address the pressing need for stealth capabilities. The most formidable hurdle that all types of detectors must overcome is the extraction of line spectral components relevant to the target, given the convoluted underwater environment teeming with significant noise pollution. In this paper, a pioneering deep learning-based algorithm, known as the Additive Diffusion Probabilistic Denoising Model (ADPDM), is proposed to rectify the performance inadequacies of neural network-based approaches when operating under low signal-to-noise ratios (SNRs). To begin with, the ADPDM was ingeniously crafted. It was designed to astutely modify the representation of underwater signals by transforming the generative inference process of the diffusion model into a deterministic recovery strategy. Subsequently, the ADPDM was expanded into the complex-valued time–frequency joint domain, in order to take full advantage of the multi-dimensional information representation brought about by the lofargram. Moreover, an accelerating inference algorithm was adopted and calibrated to be fully compatible with the ADPDM framework. In contrast to the prevailing frequency line trackers that predominantly concentrate on discerning the frequency positions of the line spectrum, the ADPDM is dedicated to unearthing and reconstructing the latent line spectrum components concealed within the observed signal. This, in turn, paves the way for more effective subsequent detection or estimation operations. Empirical results demonstrated that the frequency lines within the signal enhanced by the ADPDM can be detected with remarkable efficacy, even when a relatively less sophisticated tracker is employed. On the basis of these findings, the detection performance metrics of the ADPDM have been shown to outstrip those of the current state-of-the-art (SOTA) methods, both those founded on deep learning and the hidden Markov model (HMM), across the entire spectrum of experimental SNRs.}, archive = {J_EAAI}, author = {Qifan Shen and Xinwei Luo and Long Chen}, doi = {10.1016/j.engappai.2025.111601}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111601}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep learning based iterative denoising algorithm for multiple frequency lines recovery}, volume = {159}, year = {2025}, }
• (2025). Hyperspectral image denoising via group-sparsity constrained low-rank matrix triple factorization and spatial–spectral residual total variation. EAAI, 159, 111600. (www)

  @article{ , abstract = {Mixed noise, such as Gaussian noise, impulse noise, deadline noise, stripe noise, and many others, distorts the hyperspectral image (HSI), usually causing severe difficulties in subsequent applications. Due to the rise of artificial intelligence technology, matrix triple factorization is attached importance again in the field of HSI denoising. However, for convenient computations, these factor matrices are commonly imposed by the orthogonality, which is inconsistent with the physical meanings in practice. To address this issue, this article proposes a group-sparsity constrained triple factorization method to explore the shared sparse pattern and yields a tighter approximation to the low-rank prior. Specifically, the Casorati matrix of each local cube in HSIs, is firstly decomposed into a core matrix and two factor matrices. Then, the group-sparsity regularization is imposed on the factor matrices and the core matrix, simultaneously representing the low-rank and sparse prior in local cubes. Moreover, we also use the tensor group-sparsity based spatial–spectral residual total variation to globally explore the shared sparse pattern in both spatial and spectral difference images of HSIs. Ultimately, the group-sparsity constrained local low-rank matrix triple factorization and global spatial–spectral residual total variation model is proposed for HSI denoising. In the framework of the alternating direction method of multipliers, the proposed model can be solved efficiently. Simulated and real HSI experiments demonstrate the effectiveness of the proposed model. Across all datasets and noise conditions, our method achieves an average increase of nearly 1.93 decibels in overall peak signal-to-noise ratio compared to state-of-the-art HSI denoising methods.}, archive = {J_EAAI}, author = {Xiaozhen Xie and Yangyang Song}, doi = {10.1016/j.engappai.2025.111600}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111600}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hyperspectral image denoising via group-sparsity constrained low-rank matrix triple factorization and spatial–spectral residual total variation}, volume = {159}, year = {2025}, }
• (2025). Suspicious human activity detection model via improved deep joint segmentation and hybrid model. EAAI, 159, 111599. (www)

  @article{ , abstract = {In general, suspicious human activities do not occur in front of the cameras, making it more challenging to detect the activities via Closed Circuit Television (CCTV) footage. The problem of detecting suspicious activity in low-quality videos is not sufficiently addressed by state-of-the-art efforts. Unlike traditional models that rely on limited spatial or temporal features, the proposed framework introduces a multi-level, deeply integrated approach that enhances both feature representation and behavior classification. The proposed model has several innovations such as enhanced preprocessing, advanced feature extraction, hybrid deep learning architecture and improved score-level fusion. First, during the pre-processing phase, the video is altered to frames. Pre-processing is then carried out after enhanced Deep Joint (DJ) segmentation. Next comes feature extraction, which involves drawing the Shape Local Binary Texture (SLBT), enhanced Local Gradient Increasing Pattern (LGIP), and Hierarchy of Skeleton (HOS) features. The core Artificial Intelligence (AI) implementation utilizes Local Binary Pattern (LBP)-embedded Convolutional Neural Networks (CNNs) in integration with Long Short-Term Memory (LSTM) networks to effectively model and recognize temporal behavior patterns. Finally, the proposed work adopts improved Score level fusion for getting the ultimate result of suspicious behaviors. The application of AI in this work revolves around the use of deep learning models (CNNs and LSTMs) to analyze video data, extract meaningful patterns, and ultimately identify behaviors that deviate from the norm, making it a powerful tool for automated surveillance systems. The proposed method offers significant advantages with more reliable suspicious behavior identification through the fusion of spatial and temporal models.}, archive = {J_EAAI}, author = {Mohd Hanief Wani and Arman Rasool Faridi}, doi = {10.1016/j.engappai.2025.111599}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111599}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Suspicious human activity detection model via improved deep joint segmentation and hybrid model}, volume = {159}, year = {2025}, }
• (2025). Investigation of the dependence of fill factor and efficiency on gallium-doped silicon wafer characteristics using unsupervised learning. EAAI, 159, 111598. (www)

  @article{ , abstract = {This research analyses the performance parameters of bifacial silicon solar cells using the production line data of gallium-doped silicon wafers through unsupervised machine learning models and reports suitable input parameters. Gallium doping offers enhanced electronic properties, mitigating the light-induced degradation commonly seen in boron-doped silicon, thus making it a promising material for high-efficiency solar cells. The impact of silicon wafer thickness, resistivity, and carrier lifetime on fill factor, open-circuit voltage, and efficiency has been investigated. Employing unsupervised learning models, extensive production line data has been analysed to elucidate the complex dependencies between input parameters of silicon wafers (thickness, resistivity, and carrier lifetime) and performance parameters of fabricated solar cells using these silicon wafers (fill factor, open-circuit voltage, and efficiency). The findings indicate significant dependency of performance parameters on input parameters. Random forest classifier model demonstrated robust predictive capabilities, providing valuable insights for optimizing manufacturing processes. Based on analysis with unsupervised learning of data, following conclusions are drawn for input parameters of silicon wafers: (i) thickness range of 160–164 micro-meter gives the highest fill factor, (ii) resistivity range of 0.5–0.8 Ohm-centimetre is superior for higher fill factor, (iii) carrier lifetime of 0.9–1.05 micro-second results in better fill factor. A combination of all the above three parameters works in favour of high fill factor, it may not be a good idea to achieve one and deviate from the others. The results contribute to guiding future material and process optimizations, pushing the boundaries of solar cell efficiency.}, archive = {J_EAAI}, author = {Denish Hirpara and Paramsinh Zala and Meenakshi Bhaisare and Chandra Mauli Kumar and Mayank Gupta and Manoj Kumar and Brijesh Tripathi}, doi = {10.1016/j.engappai.2025.111598}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111598}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Investigation of the dependence of fill factor and efficiency on gallium-doped silicon wafer characteristics using unsupervised learning}, volume = {159}, year = {2025}, }
• (2025). VFT: A versatile fine-tuning scheme based on feature distribution-aware knowledge distillation for lightweight convolutional neural networks. EAAI, 159, 111597. (www)

  @article{ , abstract = {Various network compression techniques, such as pruning and quantization, are being actively researched in order to lighten convolutional neural networks (CNNs), which have increasingly deep and complex structures accompanied by the achievement of higher accuracy. Since most of these network compression techniques cause a decrease in accuracy, fine-tuning is essential to recover the performance of lightweight models; however, fine-tuning has received limited research attention compared to numerous compression techniques, and thus, performance recovery by fine-tuning has significant room for improvement. In this paper, we analyze the shortcomings of existing fine-tuning methods in terms of loss landscape and introduce a knowledge distillation (KD)-based fine-tuning approach that solves these problems. In particular, to overcome the limitation that KD can be adversely affected by the capacity difference between the teacher and student models or the defined knowledge to be transferred, we propose a feature distribution-aware knowledge distillation (FDKD) method, which defines appropriate supervision in the form of feature distribution to transfer the semantic information from teacher models. Moreover, we also propose a layer-wise FDKD method by exploiting the uniqueness of the lightweight model that the baseline ( i.e. , teacher) and compressed models ( i.e. , student) have the same architecture. Experiments on classification tasks demonstrate the superiority of the proposed method over existing fine-tuning methods, achieving up to 1.99% and 3.83% of accuracy improvement for pruned and quantized models, respectively. The source code for this implementation is available at [ https://github.com/IDSL-SeoulTech/VFT ].}, archive = {J_EAAI}, author = {Hyeonseok Hong and Hyun Kim}, doi = {10.1016/j.engappai.2025.111597}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111597}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {VFT: A versatile fine-tuning scheme based on feature distribution-aware knowledge distillation for lightweight convolutional neural networks}, volume = {159}, year = {2025}, }
• (2025). Automatic clustering by automatically generated algorithms. EAAI, 159, 111596. (www)

  @article{ , abstract = {Clustering data based on similarity becomes particularly challenging when the number of clusters is not known in advance. This case, known as the automatic clustering problem (ACP), corresponds to an optimization problem that aims to identify the best possible clustering among the many existing options. Although several effective ACP methods have been proposed, identifying optimal clusterings remains a difficult task, and the space of algorithmic solutions has yet to be thoroughly explored. Existing approaches suggest that better results can be achieved by appropriately combining and assembling different techniques. While some combinations have been explored, many others remain unexamined and could be evaluated through a more exhaustive exploration, such as the automatic generation of algorithms (AGA). This article considers the combinations arising from the automatic construction of algorithms for the ACP. To this end, an optimization meta-problem is defined to construct algorithms with the best computational performance. The search for the optimal solution to the meta-problem allows a computational exploration of the space defined by all possible combinations of elementary algorithmic components. We specifically explore the potential of AGA to generate ACP-specialized algorithms tailored to each dataset. Through extensive computational experiments, we evaluate the effectiveness of these specialized algorithms with general-purpose algorithms generated by AGA and six state-of-the-art ACP algorithms across well-established datasets. The results demonstrate that both AGA-generated algorithms outperform the state-of-the-art ACP algorithms, with statistically significant differences. Furthermore, the specialized algorithms exhibit superior effectiveness, highlighting their advantage over their general-purpose counterparts.}, archive = {J_EAAI}, author = {Moisés Silva-Muñoz and Jonnatan Oyarzún and Gustavo Semaan and Carlos Contreras-Bolton and Carlos Rey and Victor Parada}, doi = {10.1016/j.engappai.2025.111596}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111596}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic clustering by automatically generated algorithms}, volume = {159}, year = {2025}, }
• (2025). Towards quantum audio steganalysis using synergy of quantum fourier transform and quantum neural network. EAAI, 159, 111595. (www)

  @article{ , abstract = {The proposed approach in this study introduces a comprehensive audio steganalysis scheme that integrates quantum signal processing with machine learning techniques. This method employs the quantum Fourier transform on the Quantum Representation of Digital Signals (QRDS) to extract statistical features from the second-order derivatives of the audio spectrum. These features are derived by analyzing the rate of change in the gradient of the quantum spectrum, providing valuable insights for identifying steganographic content, concealed within the audio data. The statistical analysis of these features includes the quantum spectral center (QSC), quantum spectral bandwidth (QSB), quantum spectral flatness measurement (QSFM), and quantum spectral crest factor (QSFC). The extracted features are then input into a multilayer quantum neural network that utilizes simple quantum gates, thereby reducing the algorithm's complexity and the time required for training and testing. The classification algorithm, applied by this neural network, can distinguish between clean and stego audio datasets, with an accuracy exceeding 96 %. It outperforms existing methods in both efficiency and accuracy.}, archive = {J_EAAI}, author = {Sanaz Norouzi Larki and Mohammad Mosleh and Mohammad Kheyrandish}, doi = {10.1016/j.engappai.2025.111595}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111595}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards quantum audio steganalysis using synergy of quantum fourier transform and quantum neural network}, volume = {159}, year = {2025}, }
• (2025). A trustworthy and explainable deep learning framework for skin lesion detection in smart dermatology. EAAI, 159, 111594. (www)

  @article{ , abstract = {The rapid evolution of artificial intelligence (AI) and deep learning profoundly impacts medical imaging, where it significantly enhances diagnostic accuracy. However, the effective deployment of AI systems in clinical settings, especially for skin lesion detection and diagnosis, requires not only high accuracy but also transparency and robustness to gain the trust of healthcare professionals. This is particularly crucial considering the challenges posed by varying sensor quality, lighting conditions, and lesion diversity. In this paper, we introduce a novel framework based on the You Only Look Once (YOLO) model that addresses these critical needs by enhancing both the explainability and performance of skin lesion detection models. Early and accurate identification of skin lesions is essential for the timely treatment and management of dermatological conditions. Traditional diagnostic methods, such as visual assessments by dermatologists, are often labor-intensive, subject to interpretative variability, and prone to inaccuracies, especially in cases involving atypical or subtle lesions. Our approach incorporates advanced data augmentation techniques to improve the model’s generalization capabilities across diverse clinical conditions. Additionally, we integrate saliency maps to provide visual explanations of the model’s predictions, allowing clinicians to understand the decision-making process and ensuring alignment with established clinical knowledge. Comparative analyses with the state-of-the-art models highlight the superior performance of our proposed framework, with significant improvements in the harmonic mean of precision and recall (F1-Score), and the Mean Average Precision (mAP50). The results underscore the effectiveness of our framework and how it advances the application of trustworthy AI in dermatology, paving the way for more reliable and informed clinical decisions in the diagnosis and treatment of skin conditions.}, archive = {J_EAAI}, author = {Mohammad A. Eita and Hamada Rizk}, doi = {10.1016/j.engappai.2025.111594}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111594}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A trustworthy and explainable deep learning framework for skin lesion detection in smart dermatology}, volume = {159}, year = {2025}, }
• (2025). Comparing attention-based methods with long short-term memory for state encoding in reinforcement learning-based separation management. EAAI, 159, 111592. (www)

  @article{ , abstract = {Reinforcement learning (RL) is a method that has been studied extensively for the task of conflict-resolution and separation management within air traffic control, offering advantages over analytical methods. One key challenge associated with RL for this task is the construction of the input vector. Because the number of agents in the airspace varies, methods that can handle dynamic number of agents are required. Various methods exist, for example, selecting a fixed number of aircraft, or using methods such as recurrent neural networks or attention to encode the information. Multiple studies have shown promising results using these encoder methods, however, studies comparing these methods are limited and the results remain inconclusive on which method works better. To address this issue, this paper compares different input encoding methods: three different attention methods – scaled dot-product, additive and context aware attention – and long short-term memory (LSTM) with three different sorting strategies. These methods are used as input encoders for different models trained with the Soft Actor–Critic algorithm for separation management in high traffic density scenarios. It is found that additive attention is the most effective at increasing the total safety and maximizing path efficiency, outperforming the commonly used scaled dot-product attention and LSTM. Additionally, it is shown that the order of the input sequence significantly impacts the performance of the LSTM based input encoder. This is in contrast with the attention methods, which are sequence-independent and therefore do not suffer from biases introduced by the order of the input sequence.}, archive = {J_EAAI}, author = {D.J. Groot and J. Ellerbroek and J.M. Hoekstra}, doi = {10.1016/j.engappai.2025.111592}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111592}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comparing attention-based methods with long short-term memory for state encoding in reinforcement learning-based separation management}, volume = {159}, year = {2025}, }
• (2025). DAMSLNet: Dual-attention multi-scale lightweight network for plant disease classification. EAAI, 159, 111590. (www)

  @article{ , abstract = {Accurately identifying crop diseases plays a crucial role in advancing intelligent and modern agricultural production. Deep learning techniques have performed robust performance in classifying plant disease images. However, current studies face the challenge that many plant disease datasets are generated in controlled environments, leading to reduced model performance in real-world agricultural settings. This paper aims to provide a lightweight model that can accurately classify plant diseases in natural environments. Specifically, this paper investigates the Dual-Attention Multi-Scale Lightweight Network (DAMSLNet), which combines dual-attention-based multi-scale feature extraction and deep information fusion, to classify plant diseases. At the front end, the model employs depth-wise separable convolution (DSC) to extract shallow features from images, followed by the improved InceptRes module for multi-scale deep feature extraction. The back end further focuses the network’s attention on the diseased part through the dual attention module (DA Module), while ignoring unimportant features. The proposed DAMSLNet has been trained and verified on four datasets, achieving a classification accuracy of 98.19%, 99.96%, 97.08%, 99.99% respectively. The complexity of the DAMSLNet is about 16% of InceptionV4, and the parameter volume is also reduced to 37.5% of InceptionV4. In addition, this paper used technologies such as Gradient-weighted Class Activation Mapping (GradCAM) to perform interpretability analysis on the performance of different modules in the network. The DAMSLNet was compared with state-of-the-art disease classification networks. The experimental results show that our proposed network has better classification performance compared to existing advanced methods.}, archive = {J_EAAI}, author = {Linfan Deng and Juan Qin and Kun Li and Jinhua Zhu and Zhaoxia Wang}, doi = {10.1016/j.engappai.2025.111590}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111590}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DAMSLNet: Dual-attention multi-scale lightweight network for plant disease classification}, volume = {159}, year = {2025}, }
• (2025). Bending analysis of cantilever microbeams with three porosity distributions using physics-informed neural network and modified couple stress theory. EAAI, 159, 111589. (www)

  @article{ , abstract = {This study explores the use of a Physics-Informed Neural Network (PINN) framework to investigate the bending behavior of a cantilever microbeam made of porous material. PINN is a powerful approach that combines machine learning with physics principles to address the challenges of limited training data and enforce domain knowledge into the learning process, making them effective surrogate solvers for Partial Differential Equations (PDEs). In this work, a cantilever microbeam subjected to a uniformly distributed transverse load is examined, considering three different pore distributions including homogeneous, symmetric, and non-symmetric. The bending analysis incorporates the size effect by integrating the modified couple stress theory with the Euler-Bernoulli beam theory. First, the bending equation based on the modified couple stress theory is extended to include porous material properties. The governing equation is then solved using the Laplace transform. The PINN model is trained to approximate the solution by minimizing a loss function that accounts for residual errors at collocation points, as well as initial and boundary conditions. To enhance computational efficiency, the optimal hyperparameters of the PINN model are determined using a combination of Taguchi design of experiments and the Grey Relational Method. Taguchi-Grey approach effectively captures the trade-off between these objectives by normalizing and aggregating them into a single value to reflect the overall performance. The results are validated against analytical solutions based on Laplace transform, and the influence of key parameters such as microbeam length, length scale parameter, and porosity is systematically investigated.}, archive = {J_EAAI}, author = {Aiman Tariq and Büşra Uzun and Babür Deliktaş and Mustafa Özgür Yaylı}, doi = {10.1016/j.engappai.2025.111589}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111589}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bending analysis of cantilever microbeams with three porosity distributions using physics-informed neural network and modified couple stress theory}, volume = {159}, year = {2025}, }
• (2025). Point cloud-based intelligent collision avoidance for stacker cranes in stereoscopic warehouses. EAAI, 159, 111588. (www)

  @article{ , abstract = {In stereoscopic warehouses, stacker cranes might collide with obstacles, resulting in substantial economic losses. Therefore, autonomous collision avoidance is essential for the safe operation of stacker cranes. Point cloud-based three-dimensional(3D) object detection networks can provide precise environmental information for devices, among which pillar-based networks are more suitable for scenarios requiring real-time response. However, current pillar-based networks suffer from insufficient pillar feature representation capability and perform poorly in situations where multiple obstacles are located at the same bird’s-eye view(BEV) coordinate. In this work, we propose a point cloud-based intelligent collision avoidance system. We first design and collect the Stereoscopic Warehouse Obstacle Dataset(SWOD), a point cloud dataset containing obstacles. Then, we propose a pillar-based detection network that includes two innovative improvements: the density-aware pillar encoder(DAPE) and the Z-axis Repositioning Module(ZRM). In the DAPE, to better extract pillar features, the estimated point cloud density is used for feature augmentation and is also used to predict the weight of point-wise feature for density control. The ZRM readjusts the z-axis positions of proposals during post-processing to identify multiple obstacles at the same BEV coordinate. Lastly, if the perception results indicate a risk of collision, an emergency braking signal will be sent to the stacker crane. On the SWOD test set, our proposed network achieves state-of-the-art detection accuracy while maintaining real-time detection speed. Additionally, we verify the feasibility of the collision avoidance system in a real-world scenario.}, archive = {J_EAAI}, author = {Guokai Yan and Chuanjun Chen and Yanchang Niu and Haonan Yin and Biqing Huang}, doi = {10.1016/j.engappai.2025.111588}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111588}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Point cloud-based intelligent collision avoidance for stacker cranes in stereoscopic warehouses}, volume = {159}, year = {2025}, }
• (2025). A dual spatial temporal neural network for bottleneck prediction in manufacturing systems. EAAI, 159, 111586. (www)

  @article{ , abstract = {In manufacturing systems, bottlenecks act as constraints that limit system throughput. Extensive efforts have been made to detect and predict bottlenecks. Traditional bottleneck prediction methods predominantly utilize time-series feature analysis, which is limited in capturing the dynamic spatial dependencies introduced by production material flow. To address these limitations, we proposed a dual spatial temporal neural network for dynamic bottlenecks (Dual-BDSTN) to learn the dependencies of temporal and spatial features dynamically. In the temporal module, a gated recurrent unit combined with a self-attention mechanism is employed to capture the time-evolving dynamics of temporal features related to machine status. In the spatial module, a dynamic graph neural network is employed to learn spatial information affected by dynamic production material flow and a cross-attention mechanism captures the effect of temporal features on spatial features. Finally, gated recurrent neural networks are applied to capture the temporal trends of the temporal and spatial features to predict future starvation and blockage for identifying bottleneck locations. Experimental results demonstrate that the proposed model outperforms the best benchmark, achieving a 5.95 and 2.95 reduction in root mean square error for predicting starvation and blockage times of overall machines in the production system respectively (over 10%), with a 2.85% improvement in bottleneck prediction accuracy.}, archive = {J_EAAI}, author = {Weihong Cen and Chupeng Su and Kainuo Cen and Lie Yang and Gang Chen and Longhan Xie}, doi = {10.1016/j.engappai.2025.111586}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111586}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual spatial temporal neural network for bottleneck prediction in manufacturing systems}, volume = {159}, year = {2025}, }
• (2025). Inter-layer explainable variational autoencoder model for multivariate time series anomaly detection. EAAI, 159, 111585. (www)

  @article{ , abstract = {The rapid expansion of big data and artificial intelligence has significantly increased the dimensionality and quantity of monitoring data across various industries. Multivariate time series data, one of the most frequently used data types in various industries, often contains anomalies caused by human error or electromagnetic interference. Detecting these anomalies is essential, the volume of data makes manual labeling impractical, and supervised methods are limited by their inability to address unknown anomalies. Additionally, existing deep learning detection methods often function as black-box models, reducing trust in the results and obscuring the testing process. To address these challenges, we propose Shapley Attention Variational Autoencoder (SHAPAttenVAE), an inter-layer explainable variational autoencoder (VAE) model that combines shapley values with a multi-head attention mechanism. SHAPAttenVAE leverages shapley values to quantify the contribution of nodes from previous layers to their successors, helping to identify root causes of anomalies across dimensions. The model is optimized using a two-phase training method inspired by adversarial network, where random Gaussian noise is introduced to reduce noise interference within the data. Experimental results on several large datasets demonstrate that SHAPAttenVAE outperforms existing models in both accuracy and robustness. These findings establish SHAPAttenVAE as an effective and interpretable framework for anomaly detection in multivariate time series, making it a promising tool for widespread industry application.}, archive = {J_EAAI}, author = {Xiaoxia Zhang and Guangyao Wang and Yi Chen and Wenzhi Yang and Guoyin Wang}, doi = {10.1016/j.engappai.2025.111585}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111585}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Inter-layer explainable variational autoencoder model for multivariate time series anomaly detection}, volume = {159}, year = {2025}, }
• (2025). A ground-based cloud image classification method for photovoltaic power prediction based on convolutional neural networks and vision transformer. EAAI, 159, 111582. (www)

  @article{ , abstract = {Cloud image classification plays a crucial role in accurately predicting solar radiation attenuation, which significantly impacts ultra-short-term photovoltaic power predictions. Despite recent advancements in cloud image classification using Transformer and convolutional neural networks, challenges remain, particularly in handling rapidly evolving cumuliform clouds. To address this, we propose CloudMViT, an improved model derived from a model that combines Convolutional Neural Networks and Vision Transformer (MobileViT). CloudMViT introduces the CloudMobileNetV2 (CMV2) Block, building upon the Mobile Network architecture (MobileNetV2). This block incorporates a triple-branch inverted residual structure consisting of depthwise separable convolution, standard convolution, and shortcut branches, along with a New Multi-scale Channel Attention Module (NMS-CAM). Simultaneously, CloudMViT proposes the CloudMViT Block, which integrates adaptive convolution to enhance local feature representation. CloudMViT improves local and global feature extraction, boosting cloud image classification accuracy. The proposed model was evaluated through ablation studies and comparative experiments on the Tianjin-Normal-University-Ground-based-Cloud-Dataset (GCD) and Tianjin-Normal-University-Ground-based-Remote-Sensing-Cloud-Database (GRSCD). The ablation experiments demonstrate that the modules introduced in CloudMViT significantly improve cloud image classification accuracy. Comparative results show that CloudMViT achieves higher accuracy than other state-of-the-art methods, reaching 91.40% on GCD and 98.25% on GRSCD datasets. Finally, CloudMViT was validated through experiments in conjunction with a photovoltaic power prediction model, achieving a prediction accuracy of 96.70%, which surpasses that of the original model. This further demonstrates that utilizing CloudMViT cloud image classification results enhances the accuracy of photovoltaic power prediction. Moreover, it verifies that improving cloud image classification accuracy can effectively enhance the precision of photovoltaic power forecasting.}, archive = {J_EAAI}, author = {Chaojun Shi and Mengyu Zhang and Hongyin Xiang and Ke Zhang and Sihao Ju and Xiaoyun Zhang and Leile Han}, doi = {10.1016/j.engappai.2025.111582}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111582}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A ground-based cloud image classification method for photovoltaic power prediction based on convolutional neural networks and vision transformer}, volume = {159}, year = {2025}, }
• (2025). Comprehensive-detail synergy with multi-level dynamic interaction for enhanced salient object detection. EAAI, 159, 111579. (www)

  @article{ , abstract = {Salient Object Detection involves identifying and segmenting the most visually distinctive objects in an image. A key challenge is distinguishing Salient objects from complex backgrounds while preserving global features and minimizing local detail loss. To address this issue, we introduce a Comprehensive-Detail Synergy with Multi-Level Dynamic Interaction for Enhanced Salient Object Detection aimed at enhancing salient object features. Initially, a Multi-Scale Pooling Self-Attention Module is introduced to capture global contextual information of salient objects by combining multi-scale max pooling across spatial dimensions with self-attention. Additionally, to better preserve local details, an Adaptive Channel Enhancement Block is proposed, utilizing an adaptive weighting strategy to prioritize salient channels and enhance the model’s ability to capture intricate local features. Furthermore, to enhance the interaction between features at different levels, a Multi-Level Diffusive Synergy Block is introduced. With the integration of the cross-attention and dynamic diffusion refinement mechanism, it enables deep features to guide shallow features in focusing on salient regions. To alleviate the loss of local details due to excessive deep feature guidance, a Dual-Domain Fusion Attention Module is proposed, which integrates global self-attention with locally enhanced feature extraction units, thereby balancing global context modeling and local detail preservation. The experimental results conducted on six challenging publicly available datasets demonstrate that the proposed method outperforms the state of the art, achieving improvements of 4.9%, 3.3%, 2.4%, 2.3%, 1.2%, and 7.2% in the Weighted Harmonic Mean of Precision and Recall. These results demonstrate that the method improves accuracy and boundary detail.}, archive = {J_EAAI}, author = {Bingfeng Li and Boxiang Lv and Qingshan Chen and Xinxin Duan and Xinwei Li}, doi = {10.1016/j.engappai.2025.111579}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111579}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comprehensive-detail synergy with multi-level dynamic interaction for enhanced salient object detection}, volume = {159}, year = {2025}, }
• (2025). Digital twin based intelligent control system on gas extraction from boreholes and experimental research. EAAI, 159, 111578. (www)

  @article{ , abstract = {Intelligent gas extraction in mines is a critical enabler for the realization of smart mine development. As an essential component of this process, the intelligent control of borehole gas extraction relies on real-time monitoring data from numerous extraction parameters, integrated with next-generation information technologies, to achieve objectives such as intelligent deployment of negative pressure, control of inefficient boreholes, and evaluate extraction effectiveness. This paper constructs an intelligent control system for borehole gas extraction based on a digital twin “Four-Dimensional” framework, enabling bidirectional mapping between physical entities and virtual digital twins through the integration of physical entities (PE) as foundational carriers, virtual entities (VE) as three-dimensional models, digital twin data (DD) as the control core, and services (SS) and connectivity (CN) as the methodological framework. Key technologies include developing a control model for the gas flow process from coal seam to borehole, processing multi-source extraction data using data fusion methods, constructing virtual twins with SolidWorks, and designing control schemes based on Model Predictive Control (MPC) algorithm. In order to verify the rationality and feasibility of the system, a pilot study on the digital twin for intelligent control of borehole gas extraction was carried out. The results show that the concentration of gas extraction after control rises significantly, and the gas extraction concentration of borehole 1# reaches the optimal range when the valve is opened to 75 %, while that of borehole 2# reaches the maximum concentration of the adjustable range when the valve is opened to 70 %, which meets the control target.}, archive = {J_EAAI}, author = {Suinan He and Hongyu Pan and Shuang Song and Tianjun Zhang and Juntao Chen and Guoying Liu and Xinshuang Cao and Yilun Xue}, doi = {10.1016/j.engappai.2025.111578}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111578}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Digital twin based intelligent control system on gas extraction from boreholes and experimental research}, volume = {159}, year = {2025}, }
• (2025). Medical artificial intelligence for early detection of lung cancer: A survey. EAAI, 159, 111577. (www)

  @article{ , abstract = {Lung cancer remains one of the leading causes of morbidity and mortality worldwide, making early diagnosis critical for improving therapeutic outcomes and patient prognosis. Computer-aided diagnosis systems, which analyze computed tomography images, have proven effective in detecting and classifying pulmonary nodules, significantly enhancing the detection rate of early-stage lung cancer. Although traditional machine learning algorithms have been valuable, they exhibit limitations in handling complex sample data. The recent emergence of deep learning has revolutionized medical image analysis, driving substantial advancements in this field. This review focuses on recent progress in deep learning for pulmonary nodule detection, segmentation, and classification. Traditional machine learning methods, such as support vector machines and k-nearest neighbors, have shown limitations, paving the way for advanced approaches like Convolutional Neural Networks, Recurrent Neural Networks, and Generative Adversarial Networks. The integration of ensemble models and novel techniques is also discussed, emphasizing the latest developments in lung cancer diagnosis. Deep learning algorithms, combined with various analytical techniques, have markedly improved the accuracy and efficiency of pulmonary nodule analysis, surpassing traditional methods, particularly in nodule classification. Although challenges remain, continuous technological advancements are expected to further strengthen the role of deep learning in medical diagnostics, especially for early lung cancer detection and diagnosis. A comprehensive list of lung cancer detection models reviewed in this work is available at https://github.com/CaiGuoHui123/Awesome-Lung-Cancer-Detection .}, archive = {J_EAAI}, author = {Guohui Cai and Ying Cai and Zeyu Zhang and Yuanzhouhan Cao and Lin Wu and Daji Ergu and Zhibin Liao and Yang Zhao}, doi = {10.1016/j.engappai.2025.111577}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111577}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Medical artificial intelligence for early detection of lung cancer: A survey}, volume = {159}, year = {2025}, }
• (2025). Adaptive graph attention networks with interactive learning for attributed graph clustering. EAAI, 159, 111574. (www)

  @article{ , abstract = {Attributed graph clustering aims to classify nodes into distinct clusters by learning graph embeddings from both graph structures and node attributes. Recently, methods based on graph neural networks have demonstrated their effectiveness through a two-stage process, i.e., first obtaining graph embeddings and then performing clustering. However, due to the independence of each stage, the graph embeddings obtained by these methods are not specifically designed for attributed graph clustering. To address this issue, we propose a new artificial intelligence scheme, the Adaptive Graph Attention Networks with Interactive Learning (AGAT-IL). In our scheme, it effectively integrates graph feature learning with clustering in an interactive framework. In detail, we design Laplacian Smoothing Filters to alleviate the high-frequency disturbances in node attributes. Then, an Adaptive Graph Attention Encoder with modularity maximization is proposed to obtain high-quality graph embeddings tailored for clustering. Furthermore, we design an Interactive Learning Unit based on the Kullback–Leibler divergence to further optimize graph embeddings according to the clustering results. To our best knowledge, the proposed AGAT-IL is the first work to enhance graph feature learning through interactive learning. The extensive experiments on four benchmark datasets show the superiority of our AGAT-IL over state-of-the-art methods, almost 3% higher than the best F1 on each dataset. Our codes are available at https://github.com/MrDec/AGAT-IL .}, archive = {J_EAAI}, author = {Weiwei Duan and Luping Ji and Lijun Wu and Qi Deng and Zhiyuan Li}, doi = {10.1016/j.engappai.2025.111574}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111574}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive graph attention networks with interactive learning for attributed graph clustering}, volume = {159}, year = {2025}, }
• (2025). Joint depth-segmentation learning with segment priors for non-contact seedling height and stem thickness estimation. EAAI, 159, 111572. (www)

  @article{ , abstract = {To achieve precise and rapid computation of seedling height and stem diameter — key phenotypic traits for monitoring seedling growth and selecting superior varieties — this study proposes a SAM-Integrated Adaptive Fusion Depth Network (SAFD-Net). SAFD-Net integrates segmentation masks generated by Segment Anything Model (SAM) with an Adaptive Prior Extraction (APE) module to produce priors focused on individual seedling characteristics, and it fuses these priors with deep features through an Adaptive Attention Fusion (AAF) module. A Local Depth Generation (LDG) module refines depth details to improve estimation accuracy, and an Adaptive Multi-scale Fusion (AMF) module merges LDG outputs at different scales to produce high-precision depth maps. From these maps, seedling region depth, pixel height, and pixel stem diameter are extracted to compute actual seedling height and stem diameter. Comparisons with various depth estimation networks demonstrate that SAFD-Net outperforms existing models in both depth estimation and seedling measurement. Experimental evaluations on seedlings from three crops with distinct phenotypic characteristics further show that the method maintains high accuracy under varying shooting distances, lighting conditions, multiple targets, and tilt angles, offering a novel approach for phenotypic monitoring during seedling cultivation. Code is released at https://github.com/Songlei7664/SAFD-Net .}, archive = {J_EAAI}, author = {Lei Song and Bo Jiang and Huaibo Song}, doi = {10.1016/j.engappai.2025.111572}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111572}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Joint depth-segmentation learning with segment priors for non-contact seedling height and stem thickness estimation}, volume = {159}, year = {2025}, }
• (2025). A fine-tuning framework based on question, context, and answer relationships for enhancing legal information retrieval. EAAI, 159, 111570. (www)

  @article{ , abstract = {Legal information retrieval is a critical task that requires identifying and retrieving relevant legal texts in response to specific queries. This task is particularly challenging due to the inherent complexity of legal language and the advanced level of comprehension that it requires. These challenges intensify in low-resource languages and specialized domains, where limited data availability and subtle linguistic variations further hinder effective retrieval. To address these issues, we introduce a fine-tuning framework based on the relationships between questions, context, and answers. This framework proposes two approaches: the first is fine-tuning based on question-context and question–answer relationships, and the second extends this by also incorporating answer-context relationships. Both approaches aim to improve the performance of embedding models for legal document retrieval tasks. Using synthetic training data generated by large language models, we demonstrate the effectiveness of our framework through experiments conducted on three Vietnamese legal datasets and ten widely used embedding models. Our proposed framework achieved an average improvement of 3.9% and 4.8% in Mean Average Precision at K for the first and second approaches, respectively. In particular, our framework outperformed the traditional question-context-based fine-tuning method even with only half of the available training data, showing its efficiency in scenarios with limited data. Furthermore, variations in parameters had minimal impact on model performance, showcasing the robustness and scalability of our solution, which can be adapted to other low-resource languages and specialized domains.}, archive = {J_EAAI}, author = {Nhu Hai Phung and Chi Thanh Nguyen and Minh-Tien Nguyen and Thu Ha Nguyen and Huu Loi Le and Truong-Phuc Nguyen}, doi = {10.1016/j.engappai.2025.111570}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111570}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A fine-tuning framework based on question, context, and answer relationships for enhancing legal information retrieval}, volume = {159}, year = {2025}, }
• (2025). Multi-objective weighted average algorithm: A novel algorithm for multi-objective optimization problems and its application in engineering problems. EAAI, 159, 111569. (www)

  @article{ , abstract = {Numerous meta-heuristic algorithms struggle with degraded performance when addressing multi-objective optimization problems due to the challenge of balancing two goals: accurately estimating Pareto-optimal solutions and ensuring their broad distribution across objectives. While the Weighted Average Algorithm (WAA) excels in single-objective optimization, its scalarization-based mechanism fundamentally conflicts with multi-objective requirements. To bridge this gap, we propose the Multi-Objective Weighted Average Algorithm (MOWAA) with three key innovations: (1) a hybrid exploration-exploitation mechanism integrating adaptive mutation and crossover operations; (2) an elitist archive management system using efficient non-dominated sorting across three critical solution sets; and (3) a novel roulette-wheel-based leader selection strategy that dynamically balances convergence and diversity. To verify the performance of the developed MOWAA, the numerical benchmark test functions (CEC2009, ZDT and DTLZ) and four engineering problems (the Binh and Korn (BNH), Constraint (CONSTR), Srinivas and Deb (SRN), and 10-bar Truss (BAR TRUSS)) are used in comparison with three multi-objective optimization algorithms. The results show that MOWAA achieves better optimization performance than comparative algorithms, with Pareto-optimal solutions exhibiting excellent convergence and coverage. Finally, applying MOWAA to an Artificial Neural Network (ANN) model using an experimental dataset on surface waviness (in mm) of Wire Arc Additive Manufacturing (WAAM) components enhances predictive accuracy by balancing optimization of prediction error and variance. Compared to single-objective optimization methods, the MOWAA approach effectively captures the complex relationships between process parameters and waviness in the WAAM process.}, archive = {J_EAAI}, author = {Jun Cheng and Wim De Waele}, doi = {10.1016/j.engappai.2025.111569}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111569}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective weighted average algorithm: A novel algorithm for multi-objective optimization problems and its application in engineering problems}, volume = {159}, year = {2025}, }
• (2025). Physics-informed partitioned coupled neural operator for complex networks. EAAI, 159, 111567. (www)

  @article{ , abstract = {Physics-Informed Neural Operators provide efficient, high-fidelity simulations for systems governed by partial differential equations. However, most existing studies focus only on multi-scale, multi-physics systems within a single spatial region, neglecting the case with multiple interconnected sub-regions, such as gas and thermal network systems. To address this, this paper proposes a Physics-Informed Partitioned Coupled Neural Operator to enhance the simulation performance of such networks. Compared to the existing Fourier Neural Operator, this method designs a joint convolution operator within the Fourier layers, enabling global integration capturing all sub-regions. Additionally, grid alignment layers are introduced outside the Fourier layers to help the joint convolution operator accurately learn the coupling relationship between sub-regions in the frequency domain. Experiments on gas, petroleum and transportation networks demonstrate that the proposed operator not only accurately simulates these complex networks but also shows good generalization and low model complexity.}, archive = {J_EAAI}, author = {Weidong Wu and Yong Zhang and Lili Hao and Yang Chen and Xiaoyan Sun and Dunwei Gong}, doi = {10.1016/j.engappai.2025.111567}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111567}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed partitioned coupled neural operator for complex networks}, volume = {159}, year = {2025}, }
• (2025). A hybrid approach for static hand gesture recognition: Integrating directional adaptive patterns with multi-scale feature extraction and aggregation. EAAI, 159, 111566. (www)

  @article{ , abstract = {This research introduces a hybrid model that combines the strengths of the Directional Adaptive Pattern (DAP) descriptor and the Multi-Scale Feature Extraction and Aggregation Network (MaXNet) to achieve robust and efficient gesture recognition. The primary objective of this study is to enhance accuracy and computational efficiency while ensuring robustness across diverse datasets. The directional adaptive pattern descriptor effectively captures intricate texture details and directional variations by leveraging directional feature analysis, adaptive neighborhood encoding, and multilevel pattern representation. To address the variable-size feature outputs of the proposed descriptor, agglomerative clustering is utilized to generate compact, fixed-size representations, reducing noise while preserving essential texture information. Multi-scale feature extraction and aggregation network further enhances multiscale feature extraction by integrating multi-kernel convolutional layers, depthwise and pointwise convolutions, and hierarchical feature aggregation. Its lightweight and modular design allows for efficient extraction of fine-grained and large-scale patterns while maintaining computational efficiency. The effectiveness of the proposed model is evaluated based on accuracy, precision, recall, and F1-score across ten benchmark datasets. Experimental results show that the proposed model achieves superior accuracy compared to the current state-of-the-art methods.}, archive = {J_EAAI}, author = {Arti Bahuguna and Gopa Bhaumik and Bam Bahadur Sinha and Mahesh Chandra Govil}, doi = {10.1016/j.engappai.2025.111566}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111566}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid approach for static hand gesture recognition: Integrating directional adaptive patterns with multi-scale feature extraction and aggregation}, volume = {159}, year = {2025}, }
• (2025). A novel feature extraction and deep spiking MobileNet for brain stroke detection using MRI image. EAAI, 159, 111565. (www)

  @article{ , abstract = {The brain is considered a complex organ that controls vision, motor skills, emotion, memory, and so on of the body activities. Brain stroke is a cerebrovascular injury occurring due to haemorrhage or brain artery ischemia and mainly causes motor damages that risk various functionalities of the human brain. More effective therapy is provided to patients by detecting brain strokes at the initial stages. Brain stroke is easily diagnosed using a commonly available imaging modality called Magnetic Resonance Imaging (MRI). Meanwhile, highly accurate and immediate interpretation remains laborious and time-consuming. In this research, Deep Spiking MobileNet (DSpM-Net) is introduced to detect brain stroke from MRI images. Here, the MRI image undergoes image denoising using a high boost filter, and the skin lesion segmentation using a Recurrent Residual Convolutional Neural Network based on U-Net (R2U-Net) is performed from the denoised image. Later, image augmentation of the segmented image is performed, and the features are extracted by utilizing a Dual Tree- Local Directional Number Pattern (DT-LDNP). After that, the brain stroke is detected using DSpM-Net from the extracted feature. Moreover, DSpM-Net attains high performance with a True Negative Rate (TNR), True Positive Rate (TPR), accuracy, and F1-score of 91.87 %, 94.76 %, 92.85 %, and 93.80 %.}, archive = {J_EAAI}, author = {Yanda Sailaja and Velmurugan P}, doi = {10.1016/j.engappai.2025.111565}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111565}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel feature extraction and deep spiking MobileNet for brain stroke detection using MRI image}, volume = {159}, year = {2025}, }
• (2025). Efficient unmanned aerial vehicle detection algorithm based on cross-weighted pixel reconstruction and feature selection network. EAAI, 159, 111562. (www)

  @article{ , abstract = {Unmanned Aerial Vehicle (UAV) detection is a key technology for Anti-UAV systems, which has high requirements for accuracy and real-time performance. The existing detection algorithms based on You Only Look Once (YOLO) struggle to strike a balance between detection accuracy and speed, making them unsuitable for practical Anti-UAV systems. To address these problems, a UAV detection algorithm based on the cross-weighted pixel reconstruction and feature selection network is proposed. First, a novel feature reconstruction module for parallel processing of multi-dimensional information is proposed, namely Cross-weighted Pixel Reconstruction module. This module designs a spatial segmentation strategy, through which the pixels of feature maps are reconstructed from three dimensions. The purpose is to highlight the object details and eliminate the interference from background and redundant information. Second, an innovative feature selection mechanism guided by semantic information is developed. This is the adjacent feature selection mechanism, which makes full use of the semantic information contained in the high-level feature maps to guide the screening of important information in the low-level feature maps. Furthermore, based on the adjacent feature selection mechanism, an improved feature fusion method, feature selection fusion, is constructed. This method can exploit semantic information and multi-scale information, enhancing the network's detection ability of small objects and similar objects. The experimental results show that the proposed algorithm has better comprehensive performance than YOLOv5, YOLOv8, YOLOv9 and YOLOv10 in UAV detection. The video of the UAV detection results based on the proposed method have been uploaded to https://github.com/mengmeng1023/CFS-YOLO .}, archive = {J_EAAI}, author = {Mengchu Tian and Jun Wang and Shaohua Yu and Meiji Cui and Huanyu Yang}, doi = {10.1016/j.engappai.2025.111562}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111562}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient unmanned aerial vehicle detection algorithm based on cross-weighted pixel reconstruction and feature selection network}, volume = {159}, year = {2025}, }
• (2025). MRTI-CR: A model based on multi-relationship and time-aware interest for personalized course recommendation. EAAI, 159, 111560. (www)

  @article{ , abstract = {With the advancement of information technology, Massive Open Online Course (MOOC) platforms offer students a diverse selection of courses but also introduce the challenge of “course overload”. Most existing course recommendation methods primarily model students’ interactions with courses implicitly, overlooking the rich multi-relationships between different entities and also failing to account for the impact of students’ evolving learning interests, particularly the influence of time on course selection behavior. To address these limitations, we propose a model based on M ulti- R elationship and T ime-aware I nterest for personalized C ourse R ecommendation(MRTI-CR), which effectively integrates heterogeneous relationships and dynamic interest evolution. Our approach extracts global features of users and courses by constructing a heterogeneous information network and leveraging a meta-path-guided graph convolutional network, such as prerequisite relationship meta-paths. Furthermore, to enhance the utilization of temporal information, we design a dynamic interest extraction module based on a time-aware Transformer. This module incorporates time-interval-aware positional encoding and optimizes multi-head attention using temporal weights, enabling the dynamic modeling of students’ learning interests. Experiments conducted on the MOOCCube public dataset demonstrate that MRTI-CR outperforms existing baseline models across multiple evaluation metrics in the course recommendation task.}, archive = {J_EAAI}, author = {Shiyi Huang and Yongquan Dong and Ziyin Wang and Nan Zhou and Yuchao Ping}, doi = {10.1016/j.engappai.2025.111560}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111560}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MRTI-CR: A model based on multi-relationship and time-aware interest for personalized course recommendation}, volume = {159}, year = {2025}, }
• (2025). A robotic grasping method of box-shaped objects based on dual-stream you only look once framework. EAAI, 159, 111559. (www)

  @article{ , abstract = {In the application area of sorting and placement, efficient sorting and precise placement of box-shaped objects have always been a key task, traditionally relying heavily on manual operation. With the advancement of industrial automation and artificial intelligence (AI), the application of AI in automation solutions offers a more effective approach to replacing manual labor, thereby enhancing production efficiency. At present, there is still a lack of box-shaped object classification and grasping methods for cluttered stacking scenes. To address this issue, this paper presents an AI-based approach that improves upon the You Only Look Once (YOLO) version 8 framework, enabling it to perform Red, Green, Blue, and Depth (RGBD) instance segmentation, called as Dual-Stream YOLO (DS-YOLO), and then combined with feature point matching algorithm, we can achieve precise recognition, grasping, and orderly placement of box-shaped objects in complex stacking environments. We created a synthetic dataset, referred to as Snack Box. Compared to state-of-the-art methods, the model trained with DS-YOLO on the Snack Box dataset achieves a 0.5% improvement in both mean average precision(mAP) at 50% intersection over union (IoU) and mAP from 50% to 90% IoU metrics, with a 93.3% grasp success rate. The average center point error is 6.20 millimeter, with plane normal vectors deviating by 4.27°and object angles by 6.26°on average. Additionally, our method outperforms state-of-the-art approaches on the Low-light Vision Visible-infrared Paired (LLVIP) dataset, with 15,488 aligned image pairs and pedestrian annotations. Our code and dataset are available at: https://github.com/DHR0703/YOLOv8_dual_Stream .}, archive = {J_EAAI}, author = {Haoran Deng and Jinlong Shi and Jiale Cui and Suqin Bai and Xin Zuo}, doi = {10.1016/j.engappai.2025.111559}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111559}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robotic grasping method of box-shaped objects based on dual-stream you only look once framework}, volume = {159}, year = {2025}, }
• (2025). An intelligent fuzzy logic based-trust system in underwater acoustic sensor networks. EAAI, 159, 111558. (www)

  @article{ , abstract = {Due to the ongoing progress in ocean exploration, underwater acoustic sensor networks (UASNs) have become a significant focus of research. However, the inherent openness and lack of supervision in these networks expose them to various security threats. Thus, efficient and reliable security systems are very necessary to keep the normal performance of these networks. Nonetheless, in the trust evaluation process, dishonest nodes likely broadcast incorrect recommendations in the network. This decreases the accuracy of the trust value and affects the normal operation of the trust process. To solve this challenge, this paper presents an intelligent fuzzy logic-based trust system (IFTS) in UASNs. The proposed scheme employs a fuzzy trust mechanism to assess direct trust. To design this mechanism, energy evidence, data evidence, and communication evidence are considered as inputs in this fuzzy system, and direct trust is extracted as the fuzzy output. Energy evidence is obtained from the remaining energy and the energy change rate. Data evidence is obtained from the packet loss rate and data consistency, and communication evidence is calculated based on three link-related parameters, namely link reliability, link delay, and link stability. Likewise, recommendation trust depends on the recommendations offered by the recommenders. The trustor node evaluates each recommender and calculates its merit by using the root mean square (RMS) error and the trust value of the trustor relative to the recommender. Furthermore, IFTS computes indirect trust based on the trust chain, i.e., a set of recommender nodes. This trust chain is built using the greedy strategy based on the closest and most reliable recommender nodes. Further, IFTS uses a sliding time window for refreshing trust values. Finally, the simulation and evaluation process of IFTS is carried out in comparison with a recommendation management trust mechanism based on collaborative filtering and variable weight fuzzy algorithm (CFFTM), an adaptive trust model based on long short-term memory (LTrust), and a trust model based on cloud theory (TMC) under three attacks, namely bad/good mouthing attack, collusion attack, and hybrid attack, and its results are compared in terms of two criteria, i.e., diagnosis accuracy rate and false diagnosis rate. Hence, in the bad/good mouthing attack, IFTS improves the indirect trust level of honest nodes, accuracy, and the false diagnosis rate by 2.24%, 1.97%, and 12.68%, respectively. In the collusion attack, IFTS upgrades the indirect trust level of abnormal nodes, accuracy, and the false diagnosis rate by 7.2%, 1.17%, and 0.69%, respectively. In a hybrid attack, IFTS optimizes accuracy and the false diagnosis rate by 2.30% and 29.27%, respectively.}, archive = {J_EAAI}, author = {Parisa Khoshvaght and Musaed Alhussein and Khursheed Aurangzeb and Mohammad Sadegh Yousefpoor and Jan Lansky and Mehdi Hosseinzadeh}, doi = {10.1016/j.engappai.2025.111558}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111558}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An intelligent fuzzy logic based-trust system in underwater acoustic sensor networks}, volume = {159}, year = {2025}, }
• (2025). Stability analysis of a homogeneous rock slope under steady-state seepage using artificial neural networks. EAAI, 159, 111556. (www)

  @article{ , abstract = {An accurate and effective procedure to determine the stability of rock slopes is paramount during the design and, even more so, during the management and maintenance of transportation infrastructures, when the rock slopes should be continuously monitored and checked against changing environmental conditions. The stability of rock slopes may be assessed using stability charts or limit equilibrium methods that assume an equivalent linear Mohr-Coulomb failure criterion for the rock or numerical models that can deal with complex configurations and rock behavior but demand more computational resources, expertise, and advanced software. In this study, an artificial neural network (ANN) was used to predict the safety factor of a two-dimensional, homogeneous rock slope incorporating several critical factors simultaneously, such as the rock's non-linear failure behavior, rock dilatancy, and different levels of the groundwater table associated with steady-state seepage flow through the rock. This greatly extends previous contributions in the field. A two-hidden layers ANN was trained using thousands of numerical simulations applying the Discontinuity Layout Optimization model, considering the rock mass toughness coefficient, non-dimensional height of the slope, artificial slope angle, dilatancy, and non-dimensional water level position. The ANN predictions closely matched the numerical results, demonstrating its potential as an easier and more accessible method for accurately evaluating the safety factor of rock slopes and as a reliable alternative to traditional numerical methods within specified input ranges. Its implementation is straightforward, using simple equations provided in the article, and it is easily scalable to perform intensive tasks and help complex engineering decision-making.}, archive = {J_EAAI}, author = {M.A. Millán and R. Galindo}, doi = {10.1016/j.engappai.2025.111556}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111556}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Stability analysis of a homogeneous rock slope under steady-state seepage using artificial neural networks}, volume = {159}, year = {2025}, }
• (2025). A novel fitness-based particle swarm optimization and radial basis function neural network for predicting liquid sloshing in tanks under seismic excitations. EAAI, 159, 111555. (www)

  @article{ , abstract = {Accurately predicting liquid sloshing in tanks under seismic excitations is of great significance in engineering applications. The establishment of physical models is a time-consuming and costly process, numerical simulations are computationally complex and exhibit limited flexibility, and analytical methods rely on linear assumptions and neglect boundary effects, while traditional artificial neural networks (ANNs) also struggle with predicting nonlinear phenomena. To address these challenges, a novel fitness-based particle swarm optimization (PSO) and radial basis function neural network (RBFNN) is proposed in this paper to intelligently predict the wave height and pressure of liquid sloshing. Specifically, a novel velocity formula incorporating a fitness function is designed and calculated to improve the global search capability of PSO. The proposed model employs earthquake and tank parameters as input data, with sloshing height and pressure as the output. The dominant parameters contributing mostly to the sloshing height/pressure are identified and quantified through the combined principal component analysis (PCA) and sensitivity analysis. The proposed model is validated against available data in literature, whose superiority is then verified by comparing it with the RBFNN and conventional PSO-RBFNN. Furthermore, comparisons with existing semi-theoretical and empirical formulas further confirm its effectiveness in evaluating the seismic responses of liquid in storage tanks.}, archive = {J_EAAI}, author = {Xin Jin and Yuan Tao and Xin-chi Zhou and Ying-zhao Yang and Min Luo and Jiu-zhou Huang and Chang-yuan Fan}, doi = {10.1016/j.engappai.2025.111555}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111555}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel fitness-based particle swarm optimization and radial basis function neural network for predicting liquid sloshing in tanks under seismic excitations}, volume = {159}, year = {2025}, }
• (2025). A multi-agent deep reinforcement learning method for fully noisy observations. EAAI, 159, 111553. (www)

  @article{ , abstract = {Multi-agent reinforcement learning (MARL) algorithms have achieved great breakthroughs in many aspects. The MARL algorithms can learn effective policies in ideal simulation environments. But different from the ideal simulation environments, noise is unavoidable in the real world. MARL algorithms need to learn effective policies in unavoidable fully noisy environments. In this paper, we consider a challenging multi-agent reinforcement learning problem: All agents cannot observe any noiseless observations from environments during the whole training process and MARL algorithms cannot learn effective policies in these fully noisy observation environments. To solve this problem, we propose a method called Robust P olicy L earning under Fully Noisy Observation vi A De N oising R E presentation Ne T work (PLANET), which enables MARL algorithms learning effective policies in fully noisy observation environments. The PLANET method learns the effective policy through two steps. (1) Extracting the noise characteristics and motion laws to obtain clean observations information from fully noisy observation histories. (2) Making MARL algorithms extract information from the noise characteristics and motion laws information, and learn effective policies. The results of a series of exhaustive experiments show that our method can mitigate the effects of noise and learn effective policies in fully noisy observation environments. Our Artificial Intelligence contribution lies in introducing the denoising representation network that learns noise characteristics and motion dynamics to recover clean observations from fully noisy observations. The proposed PLANET framework could be applied to real-world multi-agent robotic and sensor network systems, potentially improving policy robustness under fully noisy observation.}, archive = {J_EAAI}, author = {Kaiyu Wang and Danni Wang and Bohao Qu and Menglin Zhang and Xianchang Wang and Ximing Li}, doi = {10.1016/j.engappai.2025.111553}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111553}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-agent deep reinforcement learning method for fully noisy observations}, volume = {159}, year = {2025}, }
• (2025). Exploring multi-label feature selection via feature and label information supplementation. EAAI, 159, 111552. (www)

  @article{ , abstract = {Multi-label feature selection has garnered significant attention for its ability to identify informative features while capturing complex dependencies between labels and features. Therefore, it is essential for dealing with the high dimensionality of multi-label problems. Traditional information-theoretic-based feature selection methods ignore the dynamic nature of the relationship between label relationships and selected features with the labels while assessing candidate features. To address these limitations, we propose a novel method called Feature and Label Information Supplementation (FLIS), which integrates label information supplementation and the mechanisms of selected features with labels. In the FLIS method, we consider dynamic changes in selected features based on conditional mutual information and the mutual information among selected features, candidate features, and labels. Additionally, we distinguish label relationships that provide label information supplementation. Therefore, FLIS can dynamically evaluate the impact of different features on the label relationships. Experimental results on multiple benchmark datasets demonstrate that FLIS captures supplementary information from dynamic relationships between features and labels and significantly improves classification accuracy compared to traditional multi-label feature selection methods.}, archive = {J_EAAI}, author = {Suqi Zhang and Yonghao Li and Ping Zhang and Wanfu Gao}, doi = {10.1016/j.engappai.2025.111552}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111552}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploring multi-label feature selection via feature and label information supplementation}, volume = {159}, year = {2025}, }
• (2025). An oscillation based simulated annealing algorithm for the single row facility layout problem. EAAI, 159, 111551. (www)

  @article{ , abstract = {The single row facility layout problem aims to position a set of facilities of given lengths on a single line so as to minimize the weighted sum of the distances between all the pairs of facilities, which has wide real-world applications in planning areas. In this paper, we propose an oscillation based simulated annealing algorithm for solving the single row facility layout problem. Our algorithm dynamically oscillates between two simulated annealing algorithms. One employs an exponential descent insertion strategy to capture effective movements, which increases the search efficiency, while the other adopts a radius-constrained neighborhood structure to reduce search space, which significantly enhances the intensification of the search. Besides, a new fast incremental evaluation method based on decomposition and recombination is adopted to speed up the search. Experiments on 110 instances demonstrate the competitiveness of our algorithm. In specific, for all the 110 instances, our algorithm discovers new upper bounds in 32 cases and matches the best known results for other 75 instances, only remaining 3 worse results.}, archive = {J_EAAI}, author = {Baiyu Chen and Zhipeng Lü and Zhouxing Su and Junwen Ding}, doi = {10.1016/j.engappai.2025.111551}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111551}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An oscillation based simulated annealing algorithm for the single row facility layout problem}, volume = {159}, year = {2025}, }
• (2025). Research on efficient and fast extraction of vineyard navigation path based on key point detection. EAAI, 159, 111549. (www)

  @article{ , abstract = {This study proposes an improved model based on key point detection, the YOLOv8-KN (You Only Look Once version 8-Keypoint Detection Navigation) model, which is used for autonomous navigation path extraction of agricultural robots in vineyards. The model comprehensively optimizes the original network structure by introducing FasterNet Block, Efficient Multi-Scale Attention (EMA), Universal Inverted Bottleneck (UIB) module, and Upsampling by Dynamic (DySample) dynamic upsampler. The improved network can accurately locate the key points of grapevine rhizomes directly from the image, and use the least squares method to achieve fast straight-line fitting of the key points on the left and right sides, thereby generating a high-precision navigation path. In the experiment, the model achieved an average precision of 87.1% and a key point detection precision of 91.2% on the grapevine rhizome detection task. At the same time, the model parameters are reduced by 25.8% compared with the original structure, and the computational complexity is controlled within 6.7 Giga Floating-point Operations Per Second (GFLOPs). For navigation path extraction, the evaluation results show that the average yaw angle of the method is only 0.75°, the maximum yaw angle is 1.47°, the average pixel offset error is 7.94 pixels, the maximum offset error is 14.04 pixels, and the average path fitting time is only 1.66 milliseconds (ms). The experimental results thoroughly verify the efficiency and precision of the proposed model in vineyards and provide a lightweight and high-performance solution for the autonomous navigation of agricultural robots in unstructured environments such as vineyards.}, archive = {J_EAAI}, author = {Quan Chen and Jiqing Chen and Zhiwu Jiang and Lixiang Huang and Jingyao Gai and Peilin Li and Mingchang Zhang}, doi = {10.1016/j.engappai.2025.111549}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111549}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on efficient and fast extraction of vineyard navigation path based on key point detection}, volume = {159}, year = {2025}, }
• (2025). ESRPCB: An edge guided super-resolution model and ensemble learning for tiny printed circuit board defect detection. EAAI, 159, 111547. (www)

  @article{ , abstract = {Printed Circuit Boards (PCBs) are critical components in modern electronics, which require stringent quality control to ensure proper functionality. However, the detection of defects in small-scale PCBs images poses significant challenges as a result of the low resolution of the captured images, leading to potential confusion between defects and noise. To overcome these challenges, this paper proposes a novel framework, named ESRPCB (edge-guided super-resolution for PCBs defect detection), which combines edge-guided super-resolution with ensemble learning to enhance PCBs defect detection. The framework leverages the edge information to guide the EDSR (Enhanced Deep Super-Resolution) model with a novel ResCat (Residual Concatenation) structure, enabling it to reconstruct high-resolution images from small PCBs inputs. By incorporating edge features, the super-resolution process preserves critical structural details, ensuring that tiny defects remain distinguishable in the enhanced image. Following this, a multi-modal defect detection model employs ensemble learning to analyze the super-resolved image, improving the accuracy of defect identification. Experimental results demonstrate that ESRPCB achieves superior performance compared to State-of-the-Art (SOTA) methods, achieving an average Peak Signal to Noise Ratio (PSNR) of 30.54 d B ( d e c i b e l ) , surpassing EDSR by 0 . 42 d B . In defect detection, ESRPCB achieves a mAP50(mean average precision at an Intersection over Union threshold of 0.50) of 0.965, surpassing EDSR (0.905) and traditional super-resolution models by over 5%. Furthermore, the ensemble-based detection approach further enhances performance, achieving a mAP50 of 0.977. These results highlight the effectiveness of ESRPCB in enhancing both image quality and defect detection accuracy, particularly in challenging low-resolution scenarios.}, archive = {J_EAAI}, author = {Xiem HoangVan and Dang Bui Dinh and Thanh Nguyen Canh and Van-Truong Nguyen}, doi = {10.1016/j.engappai.2025.111547}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111547}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ESRPCB: An edge guided super-resolution model and ensemble learning for tiny printed circuit board defect detection}, volume = {159}, year = {2025}, }
• (2025). Dual-channel batch constrained deep Q-learning for sepsis treatment. EAAI, 159, 111545. (www)

  @article{ , abstract = {Sepsis is a highly complex and heterogeneous critical illness, requiring dynamic adjustments to treatment strategies based on individual patient characteristics. However, existing reinforcement learning (RL) methods for personalized treatment face several challenges, such as incomplete modeling of patient states, limited generalization capability of policies, and distributional shift issues in offline learning. To address these challenges, we propose a novel Dual-Channel Batch-Constrained Deep Q-Learning (DBCDQ) method to enable more precise personalized sepsis treatment. Specifically, we design a dual-channel mechanism that integrates the patient’s current physiological state with their historical treatment responses, enabling comprehensive modeling of dynamic patient characteristics and improving responsiveness to individualized treatment needs. Additionally, we introduce a batch-constrained mechanism into the policy network, which enforces consistency between the learned policy and the actual clinical data distribution. This mitigates distributional shift issues in offline reinforcement learning (ORL). We evaluate our approach on the sepsis data from the MIMIC-III dataset, and experimental results show that our method outperforms state-of-the-art methods and can reduce patient clinical mortality by 3.85%.}, archive = {J_EAAI}, author = {Liu Huidong and Zhang Xiangfei and Yu Hang and Zhang Qingchen}, doi = {10.1016/j.engappai.2025.111545}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111545}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-channel batch constrained deep Q-learning for sepsis treatment}, volume = {159}, year = {2025}, }
• (2025). A lightweight algorithm for wildlife detection in outdoor environments based on you only look once version 8 network. EAAI, 159, 111544. (www)

  @article{ , abstract = {To address the challenges posed by outdoor environments, such as obstructions, background noise, and the limited computational resources of detection devices that hinder accurate wildlife detection, we propose a lightweight wildlife detection algorithm named Wildlife-You Only Look Once (WL-YOLO). First, the High-level Screening-feature Pyramid Networks (HS-FPN) is introduced to enhance the processing of multi-scale wildlife features. This architecture captures comprehensive feature information while significantly reducing computational complexity. Next, the Depthwise Adaptive Spatial Attention (DASA) module is designed to improve the adaptability of the model to various environmental backgrounds, effectively addressing challenges such as object occlusion and complex backgrounds in wildlife images. Additionally, the Scylla Intersection over Union (SIoU) loss is utilized to optimize detection accuracy. Finally, a pruning method based on layer-adaptive magnitude-based pruning (LAMP) is applied to the model to trim redundant parameters. Experimental results indicate that WL-YOLO achieves comparable detection accuracy to the original YOLO model (version 8) while reducing parameters by 66.7% and computational load by 57.5%. It processes images at a speed of 344 frames per second. Furthermore, detection results indicate that WL-YOLO performs better in handling occlusions and complex environments. Compared to mainstream object detection algorithms, WL-YOLO achieves a favorable balance among detection accuracy, inference speed, and model complexity. The research suggests that WL-YOLO is better applicable in complex outdoor environments, offering a new pathway for research on wildlife diversity. The code will be available on GitHub ( https://github.com/xust-9527/WL-YOLO ).}, archive = {J_EAAI}, author = {Pengtao Jia and Yu Zhang}, doi = {10.1016/j.engappai.2025.111544}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111544}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight algorithm for wildlife detection in outdoor environments based on you only look once version 8 network}, volume = {159}, year = {2025}, }
• (2025). A comprehensive multimodal benchmark of neuromorphic training frameworks for spiking neural networks. EAAI, 159, 111543. (www)

  @article{ , abstract = {Spiking neural networks (SNNs) represent a promising paradigm for energy-efficient, event-driven artificial intelligence, owing to their biological plausibility and unique temporal processing capabilities. Despite the rapid growth of neuromorphic training frameworks, the lack of standardized benchmarks hinders both the effective comparison of these tools and the broader advancement of SNN-based solutions for real-world applications. In this work, we address this critical gap by conducting a comprehensive, multimodal benchmark of five leading SNN frameworks—SpikingJelly, BrainCog, Sinabs, SNNGrow, and Lava. Our evaluation system integrates quantitative performance metrics – including accuracy, latency, energy consumption, and noise immunity – across diverse datasets (image, text, and neuromorphic event data), along with qualitative assessments of framework adaptability, model complexity, neuromorphic features, and community engagement. Our results indicate that SpikingJelly excels in overall performance, particularly in energy efficiency, while BrainCog demonstrates robust performance on complex tasks. Sinabs and SNNGrow offer balanced performance in latency and stability, though SNNGrow shows limitations in advanced training support and neuromorphic features, and Lava appears less adaptable to large-scale datasets. Additionally, we investigate the effects of varying time steps, training methods, and data encoding strategies on performance. This benchmark not only provides actionable guidance for selecting and optimizing SNN solutions but also lays the foundation for future research on advanced architectures and training techniques, ultimately accelerating the adoption of energy-efficient, brain-inspired computing in practical artificial intelligence engineering.}, archive = {J_EAAI}, author = {Ying-Chao Cheng and Wang-Xin Hu and Yu-Lin He and Joshua Zhexue Huang}, doi = {10.1016/j.engappai.2025.111543}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111543}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A comprehensive multimodal benchmark of neuromorphic training frameworks for spiking neural networks}, volume = {159}, year = {2025}, }
• (2025). A lightweight image segmentation network leveraging inception and squeeze-excitation modules for efficient skin lesion analysis. EAAI, 159, 111541. (www)

  @article{ , abstract = {The U-shaped network (U-Net) and its derivatives are widely regarded as the cornerstone of medical image segmentation, with performance often improved by increasing model depth and complexity. However, this results in a greater computational burden and slower inference, limiting practical deployment. To address these issues, we propose a lightweight image segmentation based on the convolutional multilayer perceptron (MLP)-based network with U-Net (IS-UNeXt) model, a lightweight segmentation model based on an MLP framework that incorporates Inception-inspired multi-scale fusion blocks and squeeze-and-excitation (SE) modules to mitigate key limitations of existing models, such as high computational complexity, excessive parameter size, and high inference time. Evaluated on the international skin imaging collaboration 2018 (ISIC2018) and the dermoscopic image database acquired at the dermatology service of Hospital Pedro Hispano, Portugal (PH2) datasets, IS-UNeXt reduces inference time by 58.7%, parameters by 37.7%, and computational complexity by 48.4% compared to the convolutional MLP-based network with U-Net (UNeXt), while reaching an intersection over union (IoU) of 81.1% and a dice coefficient (DC) of 88.9% on ISIC2018 and IoU of 90.34% and DC of 94.42% on PH2. These results demonstrate IS-UNeXt’s effectiveness and efficiency in skin lesion segmentation, rendering it highly suitable for real-time medical applications on resource-constrained devices.}, archive = {J_EAAI}, author = {Woei-Hwa Tarn and Chi Hou Chong and Lei Wang and Chang-Fu Kuo and Jenhui Chen}, doi = {10.1016/j.engappai.2025.111541}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111541}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight image segmentation network leveraging inception and squeeze-excitation modules for efficient skin lesion analysis}, volume = {159}, year = {2025}, }
• (2025). Double stage network inference system using k means clustering and fuzzy cognitive maps for cardiovascular disease diagnosis. EAAI, 159, 111540. (www)

  @article{ , abstract = {The diagnosis of cardiovascular disease (CVD) is crucial as it stands as a prominent cause of mortality and its risk factors are often modifiable. Early diagnosis is an essential component of overall CVD health management and prevention strategies. When diagnosing CVD, numerous uncertainties arise, including variations in symptoms among individuals and symptom overlap with other diseases. To deal such uncertainties in the diagnosis of CVD, a fuzzy logic based diagnostic system is required. While designing fuzzy inference system (FIS), optimizing the rule is challenging task. When the rules are optimized in the system, the time complexity can be minimized and performance of the system may also be effective. Since optimized inference system will be more efficient, accurate, interpretable and easier to maintain, the objective of the study is to design a novel hybrid double stage network inference system using k means clustering, fuzzy cognitive maps (FCM) and Mamdani fuzzy inference system (MFIS) for diagnosing CVD. Initially, risk factors of CVD are categorized into modifiable and non-modifiable factors through k means clustering. Then, most influencing risk factors are identified from the modifiable risk factors using FCMs. The rules are created by implementing double stage network by incorporating sub factors of most influencing factors and biological factors. Then, the obtained rules from the double stage network are integrated into a MFIS to determine the level of CVD. The system’s effectiveness is assessed using a real-time clinical dataset comprising 1250 CVD risks, employing performance metrics, receiver operating curve (ROC) analysis and statistical evaluations. The proposed system demonstrates an exceptional performance, achieving a 99.23% accuracy, 98.99% sensitivity, 95.76% specificity and 99.46% precision in detecting CVD. This piece of work suggested that the proposed technique serves as a valuable tool for diagnosing CVD risks.}, archive = {J_EAAI}, author = {Stephen Mariadoss and Felix Augustin}, doi = {10.1016/j.engappai.2025.111540}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111540}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Double stage network inference system using k means clustering and fuzzy cognitive maps for cardiovascular disease diagnosis}, volume = {159}, year = {2025}, }
• (2025). A multi-sensor fusion and multi-source domain adaptive fault diagnosis method for rotating machinery. EAAI, 159, 111538. (www)

  @article{ , abstract = {The multi-sensor information fusion method is widely used in rotating machinery fault diagnosis for its ability to provide more comprehensive information, while multi-source domain adaptation improves cross-domain fault diagnosis by extracting shared features. However, how to extract common fused features from multiple source domains for multi-sensors to enhance fault diagnosis performance remains a challenge. This study proposes a multi-sensor fusion and multi-source domain adaptive fault diagnosis method. A Kronecker product-based fusion network extracts comprehensive features from multiple sensors, addressing the limitations of individual sensors. Due to distribution variations in fusion features, maximum mean discrepancy is used to capture common domain information, reducing inter-domain discrepancies and improving feature transferability. Additionally, a domain adaptive transfer learning framework employs domain anchor adapters to mitigate multi-source distribution gaps and align source-target domains, ensuring effective multi-source adaptation while aligning source and target domains. Experiments show that the proposed method achieves 99.80 % and 99.62 % on two datasets. These results validate its effectiveness in enhancing fault diagnosis accuracy and stability for rotating machinery.}, archive = {J_EAAI}, author = {Qibin Wang and Jie Xia and Mingqi Li and Lei Yin and Xinming Xie}, doi = {10.1016/j.engappai.2025.111538}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111538}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-sensor fusion and multi-source domain adaptive fault diagnosis method for rotating machinery}, volume = {159}, year = {2025}, }
• (2025). Contrastive deep graph clustering via higher-order heuristic augmentation and propagation. EAAI, 159, 111536. (www)

  @article{ , abstract = {Deep graph clustering is a critical technique for applications such as social network analysis, bioinformatics, and recommender systems, aimed at dividing nodes in a graph into disjoint clusters by leveraging the learning capabilities of deep neural networks. Current clustering methods face issues like poor alignment with tasks, reliance on fixed augmentations (causing semantic drift and info loss), and real-world data sparsity. These methods also neglect higher-order structure in sparse data, hampering their ability to capture node relations. To address these issues, we propose a Contrastive deep Graph Clustering method via Higher-order heuristic Augmentation and Propagation (CGC-HAP). Our higher-order heuristic augmentation strategy enriches intra-cluster connections by uncovering hidden structural and semantic information from local to global perspectives. Additionally, our higher-order information propagation ensures the dissemination of detailed information throughout the graph. Extensive experiments on five benchmark datasets demonstrate that CGC-HAP significantly outperforms state-of-the-art methods in clustering accuracy and robustness. Our implementations will be publicly available at: https://github.com/jianruichen/CGC-HAP .}, archive = {J_EAAI}, author = {Zheyu Zheng and Jianrui Chen and Junjie Huang and Junbo Wang}, doi = {10.1016/j.engappai.2025.111536}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111536}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Contrastive deep graph clustering via higher-order heuristic augmentation and propagation}, volume = {159}, year = {2025}, }
• (2025). Interpolation mesh-graph net using disconnected spatial graphs of higher order element based meshes. EAAI, 159, 111535. (www)

  @article{ , abstract = {The problem of disconnectivity in graph neural networks poses a significant obstacle in the accurate modeling of physical systems. Conventional simulation techniques used for complex problems often require higher order meshes containing many nodes without direct element-edge connections, which cannot be appropriately utilized by the state of the art mesh-graphs models. The existing literature has addressed the problem of isolated nodes in small to medium sized graphs to some extent, but a notable gap remains regarding higher order element based meshes in graph neural networks. This is a crucial issue, as the disconnectivity disrupts the information propagation and compromises the prediction quality, meanwhile the naive solutions are likely to fail on these challenging problems. To close this divide, we propose the Interpolation Mesh-Graph Network, the first mesh-graph neural network model that natively uses higher order element based meshes. The model utilizes the disconnected graphs derived from these meshes in a way that maintains the nodal resolution, extends the message passing distance and efficiently uses the computation resources, thereby offering a powerful solution to the disconnectivity challenge in the higher order mesh-graph based modeling of complex physical systems.}, archive = {J_EAAI}, author = {Jozsef Adam Sebestyen and Do-Nyun Kim}, doi = {10.1016/j.engappai.2025.111535}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111535}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpolation mesh-graph net using disconnected spatial graphs of higher order element based meshes}, volume = {159}, year = {2025}, }
• (2025). Real-time point-cloud-based vehicle fleet monitoring system on the edge, edge computing, and deep learning technique. EAAI, 159, 111534. (www)

  @article{ , abstract = {Intelligent Transportation Systems (ITS) increasingly rely on real-time 3-Dimensional (3D) data for vehicle dynamics, making edge computing crucial for timely and scalable 3D object detection, vehicle tracking, and counting. This paper introduces a real-time vehicle fleet monitoring system operating at 10 Hertz with a latency of 62.71 ms using advanced hardware and software. The primary objective of “fleet monitoring” is accurate vehicle counting and classification across diverse vehicle types. The hardware consists of a multi-beam flash Light Detection and Ranging (LiDAR), and an edge computing device. The software architecture is comprised of four modules: (1) a Graphics Processing Unit (GPU)-accelerated sensor interface module, which effectively processes high-density LiDAR point clouds by filtering out irrelevant background data and extracting regions of interest (RoIs); (2) Dynamic Voxelization Detector (DV-Det), a 3D object detection model which identifies and categorizes various vehicle types. It demonstrated exceptional performance, achieving 75 Hertz on the Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) dataset and surpassing bird’s eye view (BEV)-based methods in all evaluation metrics; (3) A-Baseline-for-3D-Multi-Object-Tracking (AB3DMOT) algorithm, a robust multi-vehicle tracking module; (4) a specialized multi-vehicle counting algorithm, tailored for accurate vehicle enumeration in 3D environments. Field experiments validate the system’s capability to perform fine-grain vehicle classification with high precision and recall, achieving at least 80 Mean Average Precision (mAP) at a 3D Intersection-over-Union (IoU) threshold of 0.5 for most vehicle classes, along with a recall of 100% and a precision of at least 80% for the majority of the 10 fine-grain vehicle classes.}, archive = {J_EAAI}, author = {Tun Jian Tan and Zhaoyu Su and Jun Kang Chow and Tin Long Leung and Pin Siang Tan and Mei Ling Leung and Wai Yin Gavin Wu and Hai Yang and Dasa Gu and Yu-Hsing Wang}, doi = {10.1016/j.engappai.2025.111534}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111534}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time point-cloud-based vehicle fleet monitoring system on the edge, edge computing, and deep learning technique}, volume = {159}, year = {2025}, }
• (2025). Critical area prediction in train stations using interactive object detection with detection transformer. EAAI, 159, 111532. (www)

  @article{ , abstract = {With increasing global passenger numbers, train stations face challenges in accommodating higher volumes while ensuring safety. Urbanization exacerbates this issue in densely populated, built-up areas with limited space. This poses increased safety risks due to potential overcrowding, especially during emergency evacuations. Therefore, integrating pedestrian simulators into early planning stages is crucial, yet time-consuming and manual, hindering thorough exploration of building design variants. In this article, we propose an interactive solution using a Detection Transformer to immediately assess the impact of design decisions on pedestrian flow and safety. Specifically, we present a digital train station model utilizing parametric Building Information Modeling (BIM) to generate many hundreds of floorplan variants to train our neural network. We then connect the trained network to a BIM authoring software, which allows for immediate evaluation of design changes. To address the diverse design parameter space, we present distinct customization options for the network to integrate supplementary inputs, merging simulator and floorplan data for accurate predictions. Notably, our approach conforms to official train station design guidelines and congestion standards, demonstrating its applicability in practice and ensuring compliance with safety regulations. To showcase the effectiveness and versatility of our methodology, we apply it to a real-world use case.}, archive = {J_EAAI}, author = {Patrick Berggold and Simon Brunner and Angelika Kneidl and Stavros Nousias and André Borrmann}, doi = {10.1016/j.engappai.2025.111532}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111532}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Critical area prediction in train stations using interactive object detection with detection transformer}, volume = {159}, year = {2025}, }
• (2025). Rapid rock mass classification framework using machine learning and semantic segmentation for underground mining. EAAI, 159, 111530. (www)

  @article{ , abstract = {Conventional rock mass classification still heavily relies on manual field investigations and laboratory testing. Despite advancements in Artificial Intelligence (AI)-enhanced methods, integrating machine learning for reliable classification remains challenging. This study proposes an optimized AI-powered rock mass classification framework. First, three types of machine learning models – tree-based models, a stacking ensemble model, and a Graph Neural Network (GNN) – were employed to predict uniaxial compressive strength (UCS) based on field data. Next, three semantic segmentation architectures – U-shaped Network (U-Net), Fully Convolutional Network-8s (FCN-8s), and Segmentation Network (SegNet) – were comparatively employed to detect joint fractures from field-captured rock surface images. Then, a multi-stage clustering-fitting algorithm was developed to extract key joint parameters from the segmented images. Finally, The framework was applied to a real mining case for ore rock mass classification. Results show that the GNN exhibits superior capability in modeling nonlinear parameter interactions for uniaxial compression strength prediction, while U-Net delivers the highest segmentation accuracy for joint detection under complex geological conditions. The AI-powered method achieves higher classification accuracy in ore body zones compared to surrounding rock and aligns well with the Rock Mass Rating (RMR) system. By automating parameter extraction and integrating rapid field testing, the framework reduces reliance on manual work and expensive equipment, advancing automation and reliability in geological applications.}, archive = {J_EAAI}, author = {Zilong Zhou and Jiakang Yu and Jingyao Gao and Xin Cai and Zhongkang Wang}, doi = {10.1016/j.engappai.2025.111530}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111530}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rapid rock mass classification framework using machine learning and semantic segmentation for underground mining}, volume = {159}, year = {2025}, }
• (2025). Coupled flows as guidance for model-based policy optimization. EAAI, 159, 111528. (www)

  @article{ , abstract = {Model-based reinforcement learning (MBRL) offers high sample efficiency but suffers from cumulative multi-step prediction errors that degrade long-term performance. To address this, we propose a coupled flows-guided policy optimization framework, where two coupled flows quantify and minimize the discrepancy between the true and learned state–action distributions. By reducing this divergence, the loss functions serve as both a discriminator, selecting more accurate rollouts for policy learning, and a reward signal, refining the dynamics model to mitigate multi-step errors. Theoretical analysis establishes a bound on the expected return discrepancy. Empirical evaluations demonstrate that our method achieves higher cumulative rewards than the representative model-based approaches across diverse control tasks. This highlights its applicability in data-scarce domains such as robotics, recommendation systems, and autonomous driving.}, archive = {J_EAAI}, author = {Shengrong Gong and Yi Wang and Xin Du and Yuya Sun and Lifan Zhou and Shan Zhong}, doi = {10.1016/j.engappai.2025.111528}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111528}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Coupled flows as guidance for model-based policy optimization}, volume = {159}, year = {2025}, }
• (2025). Advancing the safety of intelligent rail transit systems: A segmentation network for efficient end-of-track degradation feature extraction. EAAI, 159, 111527. (www)

  @article{ , abstract = {Accurate segmentation of rail track is crucial for the safe autonomous driving of intelligent trains. Current train operations struggle with insufficient precision in rail track segmentation, primarily due to poor end-of-track segmentation performance caused by degradation of track-end features. To address these challenges, we propose Rail Track End Wise Network (RTEW-Net), an effective rail track end wise segmentation method. This network utilizes Full-Transformer Module (FTM) for effective track feature extraction and integrates the Global Response Normalization (GRN) module to handle drastic lighting changes. Additionally, we designed the Wise Weigh Maintain (WWM) method to enhance feature learning and retain track features. To validate its effectiveness, we constructed the RailMixed2024 (RM2024) dataset. Our model achieves high-precision global rail track segmentation and optimizes end detection. Experimental results demonstrate that RTEW-Net exhibits outstanding performance on the RM2024 and RailSem19 datasets, establishing it as the state-of-the-art (SOTA) in this field.}, archive = {J_EAAI}, author = {Tao Ye and Haoran Chen and Guopeng Liu and Liu Liu and Hongbin Ren and Xiaosong Li and Xi Zhang}, doi = {10.1016/j.engappai.2025.111527}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111527}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advancing the safety of intelligent rail transit systems: A segmentation network for efficient end-of-track degradation feature extraction}, volume = {159}, year = {2025}, }
• (2025). A physics-informed global-local neural operator for parametric stochastic differential equations. EAAI, 159, 111523. (www)

  @article{ , abstract = {Deep operator network (DeepONet) learning the mapping between infinite-dimensional functions has provided a promising paradigm for learning parametric stochastic differential equations (PSDE). As an extension of DeepONet, Multi-input operator network (MIONet) allows handling multiple input functions effectively. In this study, inspired by MIONet, a physics-informed global-local neural operator (PIGLNO) is established for learning PSDE considering the uncertainties induced by multiple stochastic processes and uncertain parameters. Compared with the vanilla MIONet, our proposed PIGLNO considers both global functions evaluated at fixed grids and local functions evaluated at any discretized location simultaneously, enabling the proposed approach's efficient adaptability to varying-resolution inputs without re-training. Besides, our proposed PIGLNO does not require any conventional solver and is purely physics-informed at training time. Numerical studies demonstrate that our PIGLNO outperforms the physics-informed MIONet and achieves a strong gain in accuracy and efficiency for uncertainty quantification of PSDEs solutions. Our PIGLNO also provides useful information for rare event estimation of stochastic physical systems.}, archive = {J_EAAI}, author = {Yi Gong and Sai Hung Cheung}, doi = {10.1016/j.engappai.2025.111523}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111523}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A physics-informed global-local neural operator for parametric stochastic differential equations}, volume = {159}, year = {2025}, }
• (2025). Dynamic sequential transfer optimization for municipal solid waste incineration process under multiple operating conditions. EAAI, 159, 111519. (www)

  @article{ , abstract = {In order to comply with increasingly rigorous emission regulations as well as maintain high combustion efficiency, it is practically significant to enhance the operation performance of the municipal solid waste incineration (MSWI) process. However, it is challenging to guarantee optimal operation due to the nonstationary dynamics and multiple operating conditions of the MSWI process. To address this issue, a dynamic sequential transfer optimization method (DSTO-MOC) is proposed in this paper. First, the optimization objective oriented operating conditions are identified based on the prediction of the pollutants. Accordingly, the changeable optimization objectives are formulated under multiple conditions to match different requirements. Second, the data stream ensemble learning is employed to establish optimization objective models, where incremental learning and selective ensemble strategies are used to adapt to changing conditions. Third, a dynamic sequential transfer optimization algorithm is designed to derive the optimal solutions of the manipulated variables. Specifically, in accordance with condition variations, the homogeneous transfer optimization and the heterogeneous transfer optimization are switched adaptively, aiming to efficiently optimize the changeable objectives. Finally, the proposed method is verified by the real industrial data. The experimental results demonstrate the applicability and superiority of the proposed DSTO-MOC.}, archive = {J_EAAI}, author = {Yingying Cui and Xi Meng and Junfei Qiao}, doi = {10.1016/j.engappai.2025.111519}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111519}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic sequential transfer optimization for municipal solid waste incineration process under multiple operating conditions}, volume = {159}, year = {2025}, }
• (2025). An efficient deep learning strategy for real-time semantic segmentation of trees for embedded systems. EAAI, 159, 111516. (www)

  @article{ , abstract = {Real-time segmentation plays a critical role in semantic simultaneous localization and mapping (SLAM) and autonomous navigation, where speed of inference is often prioritized over pixel-level accuracy. Existing segmentation models, such as “You Only Look Once” version 8 (YOLOv8) or tree detection and diameter estimation algorithm based on deep learning (known as “Perceptree”) are designed for generic use cases, leading to unnecessary computational overhead in structured environments such as managed pine forests. In this paper, we propose a lightweight and optimized method for real-time tree segmentation using red, green, blue, and depth channels (RGB-D) data. Our contribution is threefold. The first contribution focuses on depth-guided region proposal: we extract candidate regions from the depth map using mathematical filtering techniques, thus reducing the search space of the supervised model. The second one deals with the development of an embedded-friendly backbone: we simplify the YOLOv8 backbone while integrating depth information, improving inference speed without compromising key features for similarly shaped and sized objects. The last one focuses on the development of a compact segmentation head: instead of pixel-wise classification, we estimate polynomial coefficients to represent object contours, drastically reducing the number of parameters and accelerating inference. Our model achieves 53 frames per second on a ray tracing 2060 super graphics processing unit (GPU), which is 2.7 times faster than YOLOv8 and 10.8 times faster than Perceptree, while achieving a mean average precision score of 78.13% on real forest data.}, archive = {J_EAAI}, author = {Pierre Leroy and Emmanuelle Abisset-Chavanne and Régis Pommier and Marco Montemurro}, doi = {10.1016/j.engappai.2025.111516}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111516}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient deep learning strategy for real-time semantic segmentation of trees for embedded systems}, volume = {159}, year = {2025}, }
• (2025). Detection and classification of anomalies in oil well production using open-world learning. EAAI, 159, 111514. (www)

  @article{ , abstract = {Accurate anomaly detection and classification are critical for operational safety and efficiency in oil well production. While machine learning methods can identify known anomalies, detecting and classifying previously unseen anomalies remains a challenge. This paper presents the first Open-World Learning strategy applied to anomalies in oil well production data. The strategy detects anomalous behavior, classifies whether it belongs to a known labeled anomaly, and, if not, clusters it into newly proposed anomaly classes and learns to classify them. The approach integrates autoencoder reconstruction error for anomaly detection, autoencoder-based dimensionality reduction to extract latent features, binary classifiers to classify known anomalies, and clustering methods to group similar unseen anomalies. If an anomaly is detected via reconstruction error, the binary classifiers determine whether it belongs to a known class. If it does not, the clustering method groups similar events into new classes, which are validated by human experts. This validation enables the training of specific binary classifiers for the new classes and updates existing ones. Experiments on real anomalous oil well production data demonstrate that the discovered clusters align well with ground-truth labels. The clustering methodology achieves 81% accuracy overall, exceeding 95% for certain anomalies, while updated binary classifiers reach up to 99% accuracy. These findings demonstrate the proposed method’s effectiveness in dynamically adapting to novel anomalies, improving classification accuracy, and enhancing oil well monitoring.}, archive = {J_EAAI}, author = {Lucas Gouveia Omena Lopes and Thales Miranda de Almeida Vieira and Pedro Esteves Aranha and Eduardo Toledo de Lima Junior and William Wagner Matos Lira}, doi = {10.1016/j.engappai.2025.111514}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111514}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detection and classification of anomalies in oil well production using open-world learning}, volume = {159}, year = {2025}, }
• (2025). Sim-to-real transfer via a style-identified cycle consistent generative adversarial network: Zero-shot deployment on robotic manipulators through visual domain adaptation. EAAI, 159, 111510. (www)

  @article{ , abstract = {The sample efficiency challenge in Deep Reinforcement Learning (DRL) compromises its industrial adoption due to the high cost and time demands of real-world training. Virtual environments offer a cost-effective alternative for training DRL agents, but the transfer of learned policies to real setups is hindered by the sim-to-real gap. Achieving zero-shot transfer, where agents perform directly in real environments without additional tuning, is particularly desirable for its efficiency and practical value. This work proposes a novel domain adaptation approach relying on a Style-Identified Cycle Consistent Generative Adversarial Network (StyleID-CycleGAN or SICGAN), an original Cycle Consistent Generative Adversarial Network (CycleGAN) based model. SICGAN translates raw virtual observations into real-synthetic images, creating a hybrid domain for training DRL agents that combines virtual dynamics with real-like visual inputs. Following virtual training, the agent can be directly deployed, bypassing the need for real-world training. The pipeline is validated with two distinct industrial robots in the approaching phase of a pick-and-place operation. In virtual environments agents achieve success rates of 90 to 100%, and real-world deployment confirms robust zero-shot transfer (i.e., without additional training in the physical environment) with accuracies above 95% for most workspace regions. We use augmented reality targets to improve the evaluation process efficiency, and experimentally demonstrate that the agent successfully generalizes to real objects of varying colors and shapes, including LEGO® cubes and a mug. These results establish the proposed pipeline as an efficient, scalable solution to the sim-to-real problem.}, archive = {J_EAAI}, author = {Lucía Güitta-López and Lionel Güitta-López and Jaime Boal and Álvaro J. López-López}, doi = {10.1016/j.engappai.2025.111510}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111510}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sim-to-real transfer via a style-identified cycle consistent generative adversarial network: Zero-shot deployment on robotic manipulators through visual domain adaptation}, volume = {159}, year = {2025}, }
• (2025). Topic models for decision-support systems, part II: Expert-in-the-loop hierarchical topic models. EAAI, 159, 111508. (www)

  @article{ , abstract = {Topic modeling has been extensively used across disciplines to extract knowledge from text collections for policy design, implementation, and monitoring. However, flat topic models are limited in their ability to group overly broad topics with more specific ones. Hierarchical topic models ( htm ) address this limitation by offering thematic analysis at different granularity levels, but existing htm s rely on complex implementations, complicating hyperparameter tuning and domain expert-guided topic curation. This paper introduces two novel algorithms for hierarchical topic modeling: htm-ws ( htm with word selection) and htm-ds ( htm with document selection). These expert-in-the-loop methods enable domain experts to refine topic hierarchies by selecting subtopics for further partitioning. They are model-agnostic—compatible with both Bayesian and neural topic models—and require minimal hyperparameter tuning, comparable to flat topic models. Both algorithms are integrated within a tool for the training and exploitation of topic models, developed in the context of the IntelComp project. Experiments on three scientific paper datasets demonstrate the effectiveness of the algorithms through quantitative evaluations with automatic metrics and qualitative assessments via human judgment, including comparisons both between the two methods and against existing hierarchical topic models from the literature.}, archive = {J_EAAI}, author = {Lorena Calvo-Bartolomé and Jerónimo Arenas-García}, doi = {10.1016/j.engappai.2025.111508}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111508}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Topic models for decision-support systems, part II: Expert-in-the-loop hierarchical topic models}, volume = {159}, year = {2025}, }
• (2025). Enhancing road crack detection with neural architecture seeks large neural network: Leveraging deep learning and augmented minority over-sampling technique on public and custom developed datasets. EAAI, 159, 111507. (www)

  @article{ , abstract = {Deep neural networks for identifying road cracks have emerged as a crucial field of study, marking a significant advancement in infrastructural maintenance. The proposed research presents a novel Neural Architecture Seeks Large Neural Network for detecting road cracks, featuring 27 convolutional layers and ten modules, leveraging Softmax for classification. Initially, the custom developed dataset contained 30,283 images, expanded to 218,073 images using the Augmented Minority Over-Sampling Technique. For better comparison, only 30,350 images are utilized from this expanded data set. Similarly, the Karlsruhe Institute of Technology and Toyota Technological Institute dataset grew from 30,274 to 217,972 images after Augmented Minority Over-Sampling Technique processing. However, 30,274 images from the original dataset and 30,327 images from the Augmented Minority Over-Sampling Technique dataset have been processed. This normalization process aimed to ensure a balanced comparative study between the original and augmented datasets, minimizing differences and enhancing the reliability of results across the datasets. Utilizing a 70/30 train-test split, the network effectively classifies seven types of crack anomalies. The model achieves 83.7% and 89.8% accuracy on the original and augmented Karlsruhe Institute of Technology and Toyota Technological Institute datasets. The custom dataset reaches up to 91.0% accuracy for post-augmentation, while the pre-augmentation accuracy is 90.7%.}, archive = {J_EAAI}, author = {Asad Ullah and Sanam Shahla Rizvi and Shengjun Xu and Amna Khatoon and Se Jin Kwon}, doi = {10.1016/j.engappai.2025.111507}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111507}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing road crack detection with neural architecture seeks large neural network: Leveraging deep learning and augmented minority over-sampling technique on public and custom developed datasets}, volume = {159}, year = {2025}, }
• (2025). Three-dimensional deep shape optimization with a limited dataset. EAAI, 159, 111504. (www)

  @article{ , abstract = {Generative models have attracted considerable attention for their ability to produce novel shapes. However, their application in mechanical design remains constrained due to the limited size and variability of available datasets. This study proposes a deep learning-based optimization framework specifically tailored for shape optimization with limited datasets, leveraging positional encoding and a Lipschitz regularization term to robustly learn geometric characteristics and maintain a meaningful latent space. Through extensive experiments, the proposed approach demonstrates robustness, generalizability and effectiveness in addressing typical limitations of conventional optimization frameworks. The validity of the methodology is confirmed through multi-objective shape optimization experiments conducted on diverse three-dimensional datasets, including wheels and cars, highlighting the model’s versatility in producing practical and high-quality design outcomes even under data-constrained conditions.}, archive = {J_EAAI}, author = {Yongmin Kwon and Namwoo Kang}, doi = {10.1016/j.engappai.2025.111504}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111504}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-dimensional deep shape optimization with a limited dataset}, volume = {159}, year = {2025}, }
• (2025). Low-light image enhancement using dual cross attention. EAAI, 159, 111501. (www)

  @article{ , abstract = {Low-light image enhancement (LLIE) aims to improve the perceptibility and interpretability of images captured in poorly illuminated environments. Existing LLIE methods often fail to capture the local self-similarity and long-range dependencies at the same time, causing the loss of complementary information between multiple modules or network layers, ultimately resulting in the loss of image details. To alleviate this issue, we design a hierarchical mutual Enhancement via a dual cross-attention transformer (ECAFormer), which introduces an architecture that enables concurrent propagation and interaction of multiple disentangling features. To capture the local self-similarity, we design a Dual Multi-head Self-Attention (DMSA), which leverages the disentangled visual and semantic features across different scales, allowing them to guide and complement each other. Further, a cross-scale DMSA block is incorporated to capture residual connections, thereby integrating cross-layer information and capturing the long-range dependencies. Experimental results show that the ECAFormer reaches competitive performance across multiple benchmarks, yielding nearly a 3.7% improvement in Peak Signal-to-Noise Ratio (PSNR) over the suboptimal method, demonstrating the effectiveness of information interaction in LLIE. For facilitating the efforts to replicate our results, our implementation is available on GitHub 1}, archive = {J_EAAI}, author = {Yudi Ruan and Hao Ma and Di Ma and Weikai Li and Xiao Wang}, doi = {10.1016/j.engappai.2025.111501}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111501}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Low-light image enhancement using dual cross attention}, volume = {159}, year = {2025}, }
• (2025). Leveraging falling acceleration and body part clustering for physics-based human fall detection with millimeter wave radar. EAAI, 159, 111500. (www)

  @article{ , abstract = {Human fall (HF) detection is critical for ensuring the health of elderly people in preventing accidents and enhancing healthcare outcomes. To address enhanced fall detection, we introduce acceleration and body part clustering for human fall detection (ABC-HF), a novel deep-learning methodology constructed for accurate HF detection using millimeter wave (mmWave) radar sensors. However, challenges arise due to data ambiguities from mmWave measurements, resulting in an incomplete representation of complex HF motions. The proposed ABC-HF framework is constructed as an anomaly detection framework based on a variational autoencoder (VAE) with a reverse frame estimation model. This model is equipped with a human fall descriptor (HF-descriptor) designed to encapsulate the latent features of HF, thereby facilitating an understanding of HF. This descriptor incorporates physical characteristics of HF. It includes a temporal differential layer specifically designed to extract acceleration features, which are critically correlated with falls. Furthermore, to reflect the composite structure of the human body, an attention-based soft clustering module has been integrated. This dual approach ensures a comprehensive representation and analysis of the dynamics involved in HF. Through rigorous quantitative and qualitative evaluations, we have validated the performance of our model. In quantitative analyses, ABC-HF stands out by achieving superior F-scores with over 94% across a range of HF scenarios, encompassing diverse fall and non-fall movements in settings such as indoor office spaces and industrial environments with complex terrain features including varied floor surfaces, elevation changes, and obstacles. During qualitative assessments, we examined the operations of the HF-descriptor in detail.}, archive = {J_EAAI}, author = {Hyunsuk Huh and Iljoo Jeong and Anna Lee and Seungchul Lee and Young-Sik Shin}, doi = {10.1016/j.engappai.2025.111500}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111500}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Leveraging falling acceleration and body part clustering for physics-based human fall detection with millimeter wave radar}, volume = {159}, year = {2025}, }
• (2025). Multi-scale feature refinement via perspective scaling and adaptive regularization for text-based person search. EAAI, 159, 111496. (www)

  @article{ , abstract = {Text-based person search, a crucial cross-modal task, aims to retrieve matching person images from a gallery given a textual query. A key challenge in this domain lies in effectively extracting and aligning multi-granular features across image and text modalities. In this paper, we propose the Multi-scale Feature Refinement Network (MSFR) to address the cross-modal feature matching challenge in text-based person search. Our specific contributions to advancing text-based person search within artificial intelligence are as follows: First, Perspective Scaling transforms person feature maps into multi-scale global feature maps, enhancing global feature diversity through multi-scale fusion and feature space expansion. Second, Multi-scale Textual Attention captures multi-scale textual features using various convolution kernel sizes, enhancing the model’s comprehension of multi-level information in text. Third, Adaptive Regularization dynamically adjusts the weight coefficients of local feature losses for image-text pairs during model training, assigning higher importance to critical local features. The application of our MSFR in engineering is demonstrated through extensive experiments on three text-based person search datasets has consistently demonstrated superior performance.}, archive = {J_EAAI}, author = {Sheng Xie and Canlong Zhang and Runcong Ma and Zhixin Li and Zhiwen Wang and Chunrong Wei}, doi = {10.1016/j.engappai.2025.111496}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111496}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale feature refinement via perspective scaling and adaptive regularization for text-based person search}, volume = {159}, year = {2025}, }
• (2025). A self-attention enhanced generative adversarial autoencoder for anomaly detection of self-discharge of lithium-ion batteries. EAAI, 159, 111495. (www)

  @article{ , abstract = {Self-discharge is a crucial parameter affecting the reliability and lifespan of lithium-ion batteries (LIBs). However, traditional methods for detecting self-discharge rely heavily on time-consuming experimental testing. To address this limitation, we propose a novel model, self-attention enhanced generative adversarial autoencoder (SAE-GAAE), for rapid and accurate LIB self-discharge anomaly detection. SAE-GAAE integrates recent advances in artificial intelligence into battery production scenarios by combining a dot-product self-attention mechanism within the encoder-which captures inter-feature dependencies and highlights key indicators-with a generative adversarial component in the latent space, which enhances generalization and robustness by regularizing feature representations. This end-to-end deep learning framework enables automatic extraction of informative representations from raw input data without relying on manual feature engineering. Moreover, a tree-structured Parzen estimator (TPE)-based Bayesian optimization algorithm is employed to efficiently fine-tune model hyperparameters, improving detection performance. Applied to the capacity grading stage in LIB production, the model uses 40 features – including voltage, current, capacity, and temperature – extracted from charge–discharge curves. Experimental evaluation on real-world production data demonstrates that SAE-GAAE achieves a 26. 27% improvement in the average area under the receiver operating characteristic curve (AUC-ROC) in four models based on the autoencoder, with a detection accuracy of 99.05% and a recall rate of 100%. These results highlight the model’s practical value in enhancing battery screening efficiency while reducing reliance on long-duration standing tests.}, archive = {J_EAAI}, author = {Zhengyu Liu and Chuanqing Wu and Tong Wu and Chao Sun and Yining Liu}, doi = {10.1016/j.engappai.2025.111495}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111495}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A self-attention enhanced generative adversarial autoencoder for anomaly detection of self-discharge of lithium-ion batteries}, volume = {159}, year = {2025}, }
• (2025). A decomposed fuzzy based fusion of decision-making and metaheuristic algorithm to select best unmanned aerial vehicle in agriculture 4.0 era. EAAI, 159, 111491. (www)

  @article{ , abstract = {As the world embraces sustainable and smart solutions, agriculture is evolving through rapid technological advancements. Unmanned Aerial Vehicles (UAVs) are transforming smart farming, particularly for smallholder farmers, by reducing costs, saving time, and improving efficiency of agricultural tasks. This study aims to introduce a comprehensive group decision-making framework for selecting the most suitable UAV for agricultural purposes. Traditional Multi-Criteria Decision-Making (MCDM) methods face challenges with intricacies, non-linearity, limited exploration of solution space and weight distortion during defuzzification. To address these issues, this study introduces a novel Decomposed Fuzzy-based Non-Linear (DFNL) optimization model within Analytical Hierarchy Process (AHP), which directly extracts subjective crisp weights from DF-decisions. A hybrid metaheuristic algorithm is then proposed to solve this model efficiently. Additionally, objective weights are calculated using the CRiteria Importance Through Inter-criteria Correlation (CRITIC) method and qualitative data, enhancing the accuracy of the decision-making process. For ranking the UAV alternatives, the full Multiplicative form of the Multi-Objective Optimization by Ratio Analysis (MULTIMOORA) method is applied. The effectiveness of the proposed methodology is demonstrated through two extensive examples and validated via a case study focusing on the Indian subcontinent. Sensitivity analysis confirms its robustness and stability. The findings and novelties are supported by comparing with other extant models. This fusion of group decision-making methods and metaheuristic algorithms improves weight accuracy, reduces manual complexity, and adapts to uncertainty, offering policymakers actionable insights and a tailored approach for UAV selection.}, archive = {J_EAAI}, author = {Rishabh Rishabh and Kedar Nath Das}, doi = {10.1016/j.engappai.2025.111491}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111491}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A decomposed fuzzy based fusion of decision-making and metaheuristic algorithm to select best unmanned aerial vehicle in agriculture 4.0 era}, volume = {159}, year = {2025}, }
• (2025). A review on data-driven prognostics and health management for wind turbine systems. EAAI, 159, 111484. (www)

  @article{ , abstract = {The wind power industry has developed rapidly due to the transformation of the global energy matrix. Prognostics and health management have attracted significant attention from industries and academia to ensure the reliability and safety of wind turbines, reduce maintenance costs, and increase productivity. Currently, most wind farms use supervisory control and data acquisition systems to collect, record, and store wind turbine operating data. Data-driven prognostics and health management of wind turbine systems have become the most commonly used methods for real-time monitoring and fault alarm prediction. Although some reviews of studies on data-driven prognostics and health management for wind turbine systems are available, they are structured mainly based on the processing tasks or key components of wind turbine systems, and have overlooked the challenges of data quality behind these processes. Different from previous works, this paper reviews the current developments of data-driven prognostics and health management for wind turbine systems from the perspective of data challenges, which include data availability, data labeling, data scarcity, data imbalance, data inconsistency, dynamic data and fleet-based data. In this paper, we discuss some open datasets and current data challenges in data-driven prognostics and health management for wind turbine systems, and review the related methods proposed for addressing these data issues. We then provide practical applications to help engineers understand data-driven prognostics and health management better. Finally, future research directions are suggested for further work on data-driven prognostics and health management for wind turbine systems.}, archive = {J_EAAI}, author = {Mi Yan and Siu Cheung Hui and Na Jiang and Ning Li}, doi = {10.1016/j.engappai.2025.111484}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111484}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A review on data-driven prognostics and health management for wind turbine systems}, volume = {159}, year = {2025}, }
• (2025). A cross dual branch guidance network for salient object detection. EAAI, 159, 111480. (www)

  @article{ , abstract = {The effective integration of multi-level contextual information is crucial for deep learning-based salient object detection. However, most existing approaches either adopt the parallel structure or the progressive structure to predict salient objects, which still face challenges in consistently and accurately detecting salient objects of varying scales. In this paper, we propose a novel cross dual branch guidance network to effectively extract the rich semantic features and gradually enhance the saliency map scale-by-scale. Concretely, the parallel branch is guided by the progressive branch to obtain coarse location information of salient objects. In turn, the progressive branch is able to obtain uniform semantics and rich details to enhance saliency map with the guidance of the parallel branch. To obtain the dynamic receptive field, a dynamic sampling module (DSM) is introduced, which can dynamically adjust the sampling positions such that the spatial details of salient objects in complex scenes can be well recognized. In addition, we design a global context module (GCM) to explore the correlation between different parts of salient object or different salient objects, which is favorable for improving the completeness of saliency map. Experiments on five released benchmark datasets demonstrate the effectiveness and superiority of our proposed approach against other state-of-the-art methods.}, archive = {J_EAAI}, author = {Yiru Wei and Zhiliang Zhu and Hai Yu and Wei Zhang}, doi = {10.1016/j.engappai.2025.111480}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111480}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cross dual branch guidance network for salient object detection}, volume = {159}, year = {2025}, }
• (2025). Polyhedral representations with high-frequency for three-dimensional point cloud classification. EAAI, 159, 111474. (www)

  @article{ , abstract = {Point clouds have become increasingly important due to their wide applications in fields such as autonomous driving and robotic navigation. However, the disordered and irregular nature of point clouds poses a challenge to traditional coordinate-based processing methods. These methods often struggle to accurately capture the geometric features of point clouds obtained in real environments, and the high-frequency structural information inherent in point clouds is often overlooked. In this study, a novel feature extraction method called Polyhedral Representations with High-Frequency (HF-Poly) is developed for point cloud classification. The HF-Poly method constructs tangent plane features for each point and creates polyhedral representations by neighboring tangent plane features. This approach enhances the capacity of the model to capture local features. Secondly, we design a novel high-frequency coding function that enhances the edge structure features using a high-frequency mapping strategy. This approach effectively delays the decay of these features in the deep network. In addition, we propose a general Local Channel Attention module (LCA), which assigns varying attention weights to different channels of the fused high-frequency polyhedral features to extract global information from the point cloud more efficiently. Extensive experiments on the ModelNet40 and ScanObjectNN datasets demonstrate that HF-Poly maintains superior classification accuracy while reducing computational costs compared to traditional methods, particularly achieving an impressive 88.7 % accuracy on the ScanObjectNN dataset, with a 10.8 % increase compared to PointNet++.}, archive = {J_EAAI}, author = {Xiaoxin Mao and Xue Li and Puming Wang and Xin Jin and Shengfa Miao and Peng Liu and Shaowen Yao and Siwang Yang}, doi = {10.1016/j.engappai.2025.111474}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111474}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Polyhedral representations with high-frequency for three-dimensional point cloud classification}, volume = {159}, year = {2025}, }
• (2025). Cluster-aggregated transformer: Enhancing lightweight parameter models. EAAI, 159, 111468. (www)

  @article{ , abstract = {Despite the advancements made by Natural Language Processing (NLP) models in handling complex tasks, capturing long-range dependencies remains a significant challenge, particularly in applications that require long input tokens, such as document and book summarization, machine translation, and sentiment analysis of lengthy user comments. This is because NLP models have fixed but limited input token sizes (typically 512 or 1024 tokens). Hence, they struggle with the complexity and dependencies of extended text sequence sizes. Although many research efforts strive to address this problem and enhance the NLP model’s performance, two significant issues remain unresolved: the high memory and computational costs, as well as the performance degradation when the model is compressed for deployment in downstream tasks. This paper introduces the cluster aggregation transformer that uses an attention clustering mechanism to replace the classic transformer with the O ( n 2 ⋅ d ) time complexity. This study not only reduces the high computational cost of processing long text input but also maintains high performance when the model is compressed for downstream tasks, making the model deployment more convenient and faster. The model’s efficiency is demonstrated on two public language summarization datasets, namely the Government Report (GovReport) with a token size of 9616 and Book Summarization (BookSum) with a token size of 143,301, in which our method significantly improved the Bidirectional and Auto-Regressive Transformers (BART) model performance. Specifically, our approach achieved 24.5%, 53.6%, and 33.3% improvements on the Recall-Oriented Understudy for Gisting Evaluation (ROUGE) 1, 2 & L scores, respectively, and this performance is maintained up to a 0.7 compression ratio of the BART model, making it suitable for downstream tasks and model deployment. In addition, our approach significantly improves Unlimiformer (the current state-of-the-art), achieving 17.1%, 12.2%, and 5.2% improvements on the ROUGE 1, 2 & L scores, respectively.}, archive = {J_EAAI}, author = {Zikun Guo and Adeyinka P. Adedigba and Rammohan Mallipeddi}, doi = {10.1016/j.engappai.2025.111468}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111468}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cluster-aggregated transformer: Enhancing lightweight parameter models}, volume = {159}, year = {2025}, }
• (2025). Sampling-tailored two-pronged network for long-tailed class imbalance learning. EAAI, 159, 111466. (www)

  @article{ , abstract = {A long-tail class imbalanced learning problem is a scenario where the rare or minority classes, representing infrequent events or categories, make up the long tail of the class distribution and have disproportionately few examples compared to the dominant classes. The resulting imbalance makes it challenging to train models effectively for these underrepresented classes. We introduce a comprehensive solution - STTP-Net : S ampling- T ailored T wo- P ronged Net work for long-tail class-imbalanced learning, which aims to address this issue holistically. The study thoroughly examines mixed sample data augmentation techniques in conjunction with various sampling strategies to identify the most effective approaches for handling long-tail imbalance. Based on this analysis, a hybrid mixup strategy tailored explicitly for data augmentation in long-tail imbalanced settings is proposed. The core of the proposed approach comprises a two-pronged network consisting of two classification heads designed to handle long-tail imbalanced datasets. One head specializes in learning the head and median classes in this design. In contrast, the other head becomes an expert in tail classes, striking a balance between accurate prediction of tail classes without compromising accuracy for the head classes. Additionally, we address the label distribution shifts in long-tail imbalance by introducing an Effective Balanced Softmax (EBS) function. The presented method achieves state-of-the-art performance in several benchmark categories for long-tail visual recognition datasets, surpassing the most prominent and pertinent end-to-end and dual-branch approaches.}, archive = {J_EAAI}, author = {Faizanuddin Ansari and Abhranta Panigrahi and Swagatam Das}, doi = {10.1016/j.engappai.2025.111466}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111466}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sampling-tailored two-pronged network for long-tailed class imbalance learning}, volume = {159}, year = {2025}, }
• (2025). Adaptive prediction of rate of penetration while oil-well drilling: A hoeffding tree based approach. EAAI, 159, 111465. (www)

  @article{ , abstract = {Oil well drilling is an expensive process that needs a particular focus. For this reason, Rate Of Penetration (ROP) has been widely approved as a measure of drilling efficiency and adequate configuration parameters. Our aim in this work consists in the elaboration of a smart system using Hoeffding trees for predicting the Rate of Penetration (ROP) in oilfield drilling. The choice of Hoeffding trees to build our model is motivated by their adaptive learning capability and drift detection. They offer continuous, fast, and efficient learning both online on data streams and offline on batch data. To validate our approach, we used real drilling data from the “Hassi-Terfa” oilfield located in Southeast Algeria. The obtained results show in comparison to the eXtreme Gradient Boosting (XGBoost) algorithm that Hoeffding trees maintain their learning capacity and produce more accurate predictions even in the presence of drifts. This is thanks to the combination of the Adaptive Windowing (ADWIN) algorithm to manage drifts and least mean squares (LMS) filters to reduce noise. This observation highlights the effectiveness of our approach to predict the ROP while oil-well drilling. The proposed smart system offers more efficient solution to predict the ROP, whether in real-time or offline. By leveraging its adaptability to changes in data distribution, our approach ensures more accurate and adaptive predictions, facilitating drilling operations optimization and boosting the overall efficiency of the process.}, archive = {J_EAAI}, author = {Djamil Rezki and Leïla-Hayet Mouss and Abdelkader Baaziz and Toufik Bentrcia}, doi = {10.1016/j.engappai.2025.111465}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111465}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive prediction of rate of penetration while oil-well drilling: A hoeffding tree based approach}, volume = {159}, year = {2025}, }
• (2025). State estimation of lithium-ion batteries with state space model. EAAI, 159, 111463. (www)

  @article{ , abstract = {For fault prediction and health management of lithium-ion batteries, numerous data-driven models leveraging Recurrent Neural Network and Transformer have been developed to characterize their degradation processes. However, the high computational cost poses significant challenges for their effective deployment on mobile devices. Recently, state space models like Mamba have mitigated these issues by managing sequence dependencies with near-linear computational complexity. Nonetheless, Mamba is constrained by its recurrent structure, resulting in limitations in modeling long-term dependencies and leveraging global contextual information. To address these limitations, we propose a simple and efficient model based on Mamba, named SE-Mamba. Specifically, SE-Mamba utilizes Mamba for sequence modeling and incorporates a Frequency Adaptive Filtering token mixer to transform latent representations into the frequency domain. This approach leverages element-wise multiplication to perform semantically adaptive frequency filtering, thereby enhancing the capacity to model long-term dependencies and exploit global contextual information. Additionally, the Particle Swarm Optimization algorithm is used to find the best hyperparameters. Experimental results across multiple datasets demonstrate that SE-Mamba maintains low computational overhead while achieving state-of-the-art performance.}, archive = {J_EAAI}, author = {Zihao Lv and Yi Song and Yu Xue and Shijie Xu and Chunlin He and Liming Xu}, doi = {10.1016/j.engappai.2025.111463}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111463}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {State estimation of lithium-ion batteries with state space model}, volume = {159}, year = {2025}, }
• (2025). Distribution line inspection method using multi-scale information augmentation and ensemble learning. EAAI, 159, 111461. (www)

  @article{ , abstract = {This inspection of power distribution lines is evolving from manual to digitalized and intelligent methods with the advancement of Unmanned Aerial Vehicle (UAV) and artificial intelligence algorithms. However, existing universal or those focusing on specific defect detectors struggle to handle the complexity of real-world inspection tasks. To bridge this gap, we proposed a comprehensive distribution line inspection method based on You Only Look Once Exceeding (YOLOX), capable of accurately detecting 14 key defects across 6 critical components. Through large-scale UAV data analysis, we identified two major challenges in practical scenarios: Small Object Detection and Long-tailed Distribution. These insights guide the development of more robust and generalizable inspection methods. And to address these two challenges, we introduced a Multi-Scale Information Augmentation module to enhance the detection of small defects in high-resolution images, and a Dual-Branch Ensemble Classifier structure to mitigate the impact of imbalanced defect distribution. Extensive experiments demonstrated the practical effectiveness of our method, achieving a R e c a l l of 76.2%, P r e c i s i o n of 59.7%, and A v e r a g e P r e c i s i o n at 50% Intersection over Union Threshold ( A P 50 ) of 52.6%. Notably, our method delivered a 51.9% improvement in small-defect R e c a l l and a 24.3% increase in tail-category performance compared to the baseline YOLOX. A fully implemented inspection system built on our framework has been deployed to assist field engineers, significantly reducing manual workload and improving inspection efficiency. These results demonstrated the robustness and adaptability of our method, offering a valuable contribution to the intelligent operation and maintenance of modern power distribution networks.}, archive = {J_EAAI}, author = {Yihao Liang and Liangwu Wei and Yanzhi Song and Zhouwang Yang}, doi = {10.1016/j.engappai.2025.111461}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111461}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Distribution line inspection method using multi-scale information augmentation and ensemble learning}, volume = {159}, year = {2025}, }
• (2025). CorNet: A deep learning method based on physics-guided and attention mechanism for predicting flow field of coronary arterial tree. EAAI, 159, 111460. (www)

  @article{ , abstract = {Replacing traditional computational fluid dynamics (CFD) with deep learning techniques has become a prevalent approach for studying blood flow and diagnosing diseases. Nevertheless, no neural network has been specifically designed for tree-like blood vessel structures that can effectively capture their bifurcation patterns and branching dependencies. Coronary neural network (CorNet) was proposed for predicting the pressure field and fractional flow reserve (FFR CorNet ) of the coronary artery tree. The novelty of this framework lies in utilizing self-attention mechanisms to address long-term spatial dependencies within the vascular tree. Our dataset comprised 295 coronary arterial trees from 273 patients, each including point clouds reconstructed from medical images and fractional flow reserve values calculated by CFD (FFR CT ). Physical constraints were incorporated to mitigate data sparsity and enhance the interpretability of the neural network. The pressure results predicted by CorNet are consistent with the pressure calculated by CFD (mean relative error = 3.96%). There is also a good consistency between FFR CT and FFR CorNet .Compared to invasive fractional flow reserve, which is considered the “gold standard”, FFR CorNet demonstrates accuracy comparable to FFR CT (88% vs. 90%) while reducing computation time by several thousand-fold. Compared to previous studies, CorNet eliminates the need to identify specific lesion sites or manually extract geometric parameters of stenotic segments. This is the first computational method to predict hemodynamics in three-dimensional vascular tree structures using attention mechanisms within a deep learning model. We foresee that this framework will enable near-real-time flow field predictions for arterial trees and offer valuable insights for cardiovascular disease treatment.}, archive = {J_EAAI}, author = {Xiaoyu Liu and Shengze Cai and Hongtao Lin and Xingli Liu and Xiuhua Hu and Longjiang Zhang and Qi Gao}, doi = {10.1016/j.engappai.2025.111460}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111460}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {CorNet: A deep learning method based on physics-guided and attention mechanism for predicting flow field of coronary arterial tree}, volume = {159}, year = {2025}, }
• (2025). Physics-informed surrogate for cardiovascular flow extrapolation through transductive learning. EAAI, 159, 111458. (www)

  @article{ , abstract = {We consider learning surrogate models that directly predict cardiovascular flow fields by mapping geometry and/or fluid properties to hemodynamic parameters. Various machine learning approaches have been developed, but they generally do not extrapolate well to problems beyond the range covered by the training data. We propose a transductive physics informed neural network (T-PINN) approach to improve the extrapolation performance. Our approach builds on the standard PINN approach, which uses governing partial differential equations (PDEs) and labeled data for problems in the training regime to guide the training of neural network surrogate, but we additionally incorporate the governing PDEs for test problems from the extrapolation regimes. T-PINN demonstrates improved extrapolation performance on three synthetic cardiovascular flow problems as compared to purely data-driven neural network surrogates and standard PINNs. Additionally, we perform experiments to investigate how T-PINN’s performance varies when the physical constraints are softened, with hard boundary constraints replaced by soft ones, or simplified PDEs by full PDEs. Our results indicate that these two variants result in similar equation residuals as the original T-PINN but lead to less accurate velocity and pressure predictions. T-PINN’s enhanced extrapolation performance can be particularly significant for cardiovascular flow predictions in clinical settings, where patient morphologies and fluid properties often exhibit variations outside the collected data.}, archive = {J_EAAI}, author = {Yuchen Wang and Nan Ye and Zhiyong Li}, doi = {10.1016/j.engappai.2025.111458}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111458}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed surrogate for cardiovascular flow extrapolation through transductive learning}, volume = {159}, year = {2025}, }
• (2025). Open-pit mine occlusion object detection for unmanned transport vehicles. EAAI, 159, 111436. (www)

  @article{ , abstract = {Accurate object recognition in open-pit mine environments is crucial for the safety of autonomous transport vehicles. Existing autonomous driving perception mostly focuses on urban structured road traffic, and it is hard to adapt to the challenging open-pit mine environment. Lacking of datasets further limits the development of the specific work. In this paper, we propose an object detection dataset for open-pit mine autonomous driving applications. This dataset encompasses data from several mines and includes different periods such as day, dusk, and night. It provides detailed annotations for diverse objects in the open-pit mines and incorporates additional attributes for evaluating occlusion detection. In addition, to address the multi-scale changes of objects in open-pit mines and the occlusion problems caused by dust, we propose a novel occlusion mine object general distribution detection method, utilizing soft labels and vehicle attribute location to reduce the positioning ambiguity in difficult backgrounds and achieve specific object detection in harsh open-pit mine environments. Our work explores the benchmark for open-pit mine object recognition involving occlusion. Comparison with mainstream techniques on the benchmark demonstrates that our approach outperforms existing state-of-the-art methods and can achieve 82.2%, 81.7%, and 76.7% average precision in easy, moderate, and hard modes, respectively.}, archive = {J_EAAI}, author = {Chao Zheng and Guoxing Bai and Yu Meng and Lu Wang and Xianyao Jiang and Li Liu}, doi = {10.1016/j.engappai.2025.111436}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111436}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Open-pit mine occlusion object detection for unmanned transport vehicles}, volume = {159}, year = {2025}, }
• (2025). Topic models for decision-support systems. part i: Training, representation and exploitation of topic models. EAAI, 159, 111424. (www)

  @article{ , abstract = {Analyzing document collections in Science, Technology, and Innovation (STI) is essential for informed policy-making. Advances in topic modeling offer powerful tools to uncover key themes within large, heterogeneous STI datasets that can inform decision-making. However, challenges such as aligning model outputs with expert knowledge, determining the desired topic granularity, and addressing model variability, limit their widespread adoption. This paper presents the Topic Analysis and Search Engine ( tase ), a platform developed within the European IntelComp project to address these limitations. tase combines Bayesian and Neural topic modeling techniques within a unified framework that incorporates an expert-in-the-loop approach for training and curating models. This framework is further enhanced with a Solr-based exploitation tool, featuring an innovative indexing method and a novel criterion for document retrieval. These features enable efficient semantic similarity calculations and seamless integration into decision-support systems. We demonstrate tase ’s scalability and effectiveness through two real-world STI use cases, highlighting its potential for broader applications. The software is freely available as open-source under the MIT license.}, archive = {J_EAAI}, author = {Lorena Calvo-Bartolomé and Diamantis Tziotzios and Saúl Blanco-Fortes and David Pérez-Fernández and Jerónimo Arenas-García}, doi = {10.1016/j.engappai.2025.111424}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111424}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Topic models for decision-support systems. part i: Training, representation and exploitation of topic models}, volume = {159}, year = {2025}, }
• (2025). Parameterized-action based deep reinforcement learning for intelligent traffic signal control. EAAI, 159, 111422. (www)

  @article{ , abstract = {Traffic Signal Control (TSC) is a crucial component in Intelligent Transportation Systems (ITS) for optimizing traffic flow. Deep Reinforcement Learning (DRL) techniques have emerged as leading approaches for TSC due to their promising performance. Most existing DRL-based approaches typically use discrete action spaces to predict the next action phase, without specifying the signal duration. In contrast, some studies employ continuous action spaces to determine signal phase timing within a fixed light cycle. To address the limitations of both approaches, we propose a flexible framework that predicts both the appropriate traffic light phase along with its associated duration. Our approach utilizes a Parameterized-action based deep reinforcement learning architecture to handle the combination of discrete-continuous actions. We evaluate our method using the Simulation of Urban MObility (SUMO) environment, comparing its efficiency against state-of-the-art techniques. Results demonstrate that our approach significantly outperforms traditional and learning-based methods.}, archive = {J_EAAI}, author = {Salah Bouktif and Abderraouf Cheniki and Ali Ouni and Hesham El-Sayed}, doi = {10.1016/j.engappai.2025.111422}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111422}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Parameterized-action based deep reinforcement learning for intelligent traffic signal control}, volume = {159}, year = {2025}, }
• (2025). A diffusion model and knowledge distillation framework for robust coral detection in complex underwater environments. EAAI, 159, 111414. (www)

  @article{ , abstract = {Coral reefs play a crucial role in marine ecosystems, but their sustainability is increasingly threatened by climate change and human activities. To aid in the protection and monitoring of these ecosystems, developing advanced artificial intelligence (AI)-based automated detection technologies is essential. This paper introduces the MambaCoral-Diffusion Detection framework (MambaCoral-Diffusion Detection, MCDD), an AI-driven approach for robust coral detection, designed to enhance performance in complex underwater environments—a critical challenge in marine engineering. Key AI contributions include integrating a diffusion model to generate realistic and diverse training data from limited and challenging underwater coral datasets, effectively alleviating the issue of data imbalance. Secondly, we adopted an innovative spatial sensing detection mechanism that enhances the accuracy of feature extraction in complex underwater environments. Finally, we introduced an efficient knowledge distillation technique that successfully transfers knowledge from complex models to more lightweight counterparts, thereby reducing computational resource requirements while maintaining efficiency and facilitating practical deployment. Experimental results show that MCDD achieves high performance on the Soft Coral dataset, reaching 91 frames per second (FPS), the mean average precision at 50% Intersection over Union (IoU) threshold of 0.843, and the mean average precision averaged across IoU thresholds from 50% to 95% of 0.566, with only 6.5 million parameters and 13.6 billion floating point operations per second (GFLOPs). These results demonstrate MCDD’s reliability and efficiency in detecting corals under complex underwater conditions, highlighting its significant potential for advancing marine research and conservation efforts. The code and dataset are available at https://github.com/RDXiaoLu/MambaCoral-DiffDet.git .}, archive = {J_EAAI}, author = {Zhaoxuan Lu and Lyuchao Liao and Chuang Li and Xingang Xie and Hui Yuan}, doi = {10.1016/j.engappai.2025.111414}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111414}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A diffusion model and knowledge distillation framework for robust coral detection in complex underwater environments}, volume = {159}, year = {2025}, }
• (2025). Graph modeling of multi-scale energy entropy for intelligent fault detection. EAAI, 159, 111389. (www)

  @article{ , abstract = {Bearing is a key component of rotating machinery, and the fault detection of bearing is of great significance to ensure the normal operation of machinery. How to extract fault features with rich information from the time-frequency domain after signal decomposition becomes the key to fault detection. However, the undershoot and overshoot phenomena affect the result of signal decomposition, and the defined fault features do not consider the correlation between local scales after decomposition, which hinders the accuracy of fault detection. To address this issue, this study proposes an intelligent fault detection algorithm, named multi-scale energy entropy-graph model fault detection algorithm. The algorithm first proposes a local envelope mode decomposition method to obtain the multi-scale components of the signal, effectively addressing the issues of undershoot and overshoot present in traditional decomposition methods while reducing signal decomposition errors. Subsequently, a multi-scale energy entropy-graph model is constructed to capture more fault information considering the correlation between local scales. Furthermore, the graph state similarity index is defined to describe the machine state, and fault detection is realized using hypothesis testing. The experimental results were compared with some benchmark methods reported in the existing literature. The experimental results show that the detection algorithm achieves the best performance in Precision of 83 %, Recall of 100 %, and F1 score of 91 %, and the detection results are satisfactory.}, archive = {J_EAAI}, author = {Jianfeng Wei and Faping Zhang and Jiping Lu and Jianing Man}, doi = {10.1016/j.engappai.2025.111389}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111389}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph modeling of multi-scale energy entropy for intelligent fault detection}, volume = {159}, year = {2025}, }
• (2025). Outcome-oriented local explanation using context-aware process patterns. EAAI, 159, 111387. (www)

  @article{ , abstract = {Predictive Process Monitoring leverages machine learning models to predict the future states of business processes. Most Predictive Process Monitoring approaches rely on black-box models which, while powerful, lack interpretability, limiting their applicability in critical decision-making scenarios. Explainable Predictive Process Monitoring has emerged to address this gap, focusing on delivering actionable and transparent insights into predictions. Current approaches, however, often fail to incorporate multiple process perspectives and granularity levels in their explanations, overlooking crucial factors that influence process outcomes. This paper proposes a novel Explainable Predictive Process Monitoring approach to deliver explanations, integrating multiple process perspectives at various levels of granularity. The proposed approach addresses the limitations of existing methods, providing comprehensive, process context-aware explanations. The effectiveness of the proposed method is assessed through experimental evaluations on real-life event logs, showing how the integration of diverse process perspectives improves the interpretability and predictive insights of local explanations with improvements ranging from 2%–3% to 15%–20% depending upon the event log under analysis.}, archive = {J_EAAI}, author = {Mozhgan Vazifehdoostirani and Andrei Buliga and Laura Genga and Chiara Di Francescomarino and Massimiliano Ronzani and Chiara Ghidini and Remco Dijkman}, doi = {10.1016/j.engappai.2025.111387}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111387}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Outcome-oriented local explanation using context-aware process patterns}, volume = {159}, year = {2025}, }
• (2025). Automated imaging and machine learning for soil bacteria classification: Challenges and insights. EAAI, 159, 111369. (www)

  @article{ , abstract = {Soil bacteria play a vital role in biofertilizer production, making accurate classification of bacteria genera essential for optimizing industrial processes. While previous studies report high classification accuracy, they often overlook the challenges of deploying such systems in real-world production environments, where image acquisition must be automated rather than conducted by the microbiologists manually. This study addresses these practical issues by investigating the automated classification of bacteria images under varying conditions of lighting and glass type. Both classical machine learning approaches and advanced neural networks, including modifications of Extreme Learning Machines - Radial Basis Function and deep convolutional architectures of Residual Neural Networks, were employed. Unforeseen challenges impacting the reliability of automated classification were identified, highlighting novel factors that could significantly affect classification accuracy. This work not only emphasizes the limitations of existing approaches in production settings but also proposes possible directions for future research in automating bacteria classification for industrial biofertilizer development. Importantly, it brings attention to a broader issue: the critical need for rigorous dataset preparation and validation when applying artificial intelligence to microscopic image analysis across biological domains.}, archive = {J_EAAI}, author = {Aleksandra Konopka and Ryszard Kozera and Lidia Sas-Paszt and Paweł Trzciński}, doi = {10.1016/j.engappai.2025.111369}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111369}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated imaging and machine learning for soil bacteria classification: Challenges and insights}, volume = {159}, year = {2025}, }
• (2025). Integrating artificial intelligence and optimization tools in bed assignment to enhance patient satisfaction under uncertainty. EAAI, 159, 111363. (www)

  @article{ , abstract = {Efficient patient bed assignment is critical for optimizing hospital resource management and enhancing care quality. Hospitals encounter challenges in managing bed capacity due to patient surges, limited bed availability, and staffing shortages, which cause delays in care and are intensified by seasonal fluctuations. This study introduces a novel two-phase methodology for hospital bed allocation that tailors artificial intelligence techniques, combining machine learning models with a mathematical optimization framework. In the first phase, several artificial intelligence-driven models estimate patients' length of stay using demographic and clinical attributes. The gradient boosting model outperformed the others, achieving 96.58 % accuracy, 96.57 % F-score, and 0.965 area under the curve. Accurate prediction enables hospitals to allocate beds more effectively, identify patients at high risk of prolonged stays, and ensure that patients receive appropriate care. In the second phase, the predicted length of stay is incorporated into an integer linear programming model to optimize bed assignment, considering factors such as minimizing patient transfers, ensuring medical isolation, and adhering to age-based room assignment policies. To address the uncertainty in the length of stay predictions, chance-constrained stochastic programming is employed, making this study one of the first to integrate length of stay forecasting with robust bed assignments. The method's applicability and effectiveness are demonstrated through a real-world case study at a pediatric hospital in Gorgan City, Iran. The results indicate the proposed model's potential to enhance bed management through no bed shortage, significant improvement in patient transfers, and approximately 20 % reduction in hospitalization delays compared to current practices.}, archive = {J_EAAI}, author = {Fatemeh Farzanegan and Mohammad Mehdi Sepehri and Mohammad Reza Ghatreh Samani and Leila Barati}, doi = {10.1016/j.engappai.2025.111363}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111363}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating artificial intelligence and optimization tools in bed assignment to enhance patient satisfaction under uncertainty}, volume = {159}, year = {2025}, }
• (2025). Real-time inner wall surface defect detection based on multi-morphological feature fusion network. EAAI, 159, 111331. (www)

  @article{ , abstract = {In industrial manufacturing, defects on the inner wall surface are crucial for quality and safety assessment. However, existing detection methods are limited by low resolution and glare interference. This study presents a Multi-morphological Feature Fusion Network for Object Detection (MFFN-OD) for 360°detection of inner wall image defects. First, it cleverly integrates panoramic imaging with conventional features through a dual-branch backbone and annular features, ensuring rotation invariance and holistic feature preservation. Second, we develop an Adaptive Multi-morphological Feature Alignment Module (AMFAM) that combats centrally polarized defects by automatically adjusting feature alignment, reducing noise, and increasing accuracy, as well as a feature interaction module with a focus on strengthening multiscale feature fusion. Third, we introduce an Asymptotic Feature Pyramid Network with Auxiliary Features (AFPN-AF) to further refine fusion, close semantic gaps, and improve performance. Experimental results show that MFFN-OD achieves 96.1% mean Average Precision (mAP) and 94.3% Average Precision (AP) for demanding faults with fast detection of 17 milliseconds per frame, meeting industrial requirements for accuracy and real-time performance.}, archive = {J_EAAI}, author = {Zhenshen Qu and Xinxu Cai and Tianyi Zhang and Jiazheng Xu and Xintong Jiang and Chuan Lin}, doi = {10.1016/j.engappai.2025.111331}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111331}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time inner wall surface defect detection based on multi-morphological feature fusion network}, volume = {159}, year = {2025}, }
• (2025). A novel framework for assessing determinant risk factors on cyber (dis)trust behaviors of netizens in deepfakes. EAAI, 159, 111319. (www)

  @article{ , abstract = {Nowadays, Generative Artificial Intelligence (GenAI) tools or trainable agents can craft synthetic media (hereafter referred to as deepfakes) in the form of realistic texts, images, videos, and audios, incorporating events or things that never occurred in real life. These GenAI tools empower marketers and malicious actors to create deepfakes, both authorized and weaponized multimedia, which allows them to include celebrities without appearing in front of cameras or creating seductive phishing scams. Although GenAI tools can reduce the cost of content construction, they enable new risky opportunities (e.g., deepfake phishing and cyberbullying) that negatively impact netizens’ learning and (dis)trust behaviors in cyberspace. To address such risks, this study proposes a Multi-Criteria-Multi-Decision-Makers (MCMDM)-based Deepfake Risk Assessment Framework (DeepFakeR-MF) to evaluate determinant factors that impact the cyber (dis)trust behaviors of netizens in deepfakes. Moreover, DeepFakeR-MF deploys a combination of a novel optimized spherical fuzzy analytic hierarchy process method and a game theory-based MCMDM approach to prioritize and recommend alternative strategies that can be taken by five management sectors (e.g., industrial enterprises, governmental organizations, media outlets, social non-profit, and educational institutes) to mitigate GenAI-associated risks. Then, we collect 100 experts’ judgments by analyzing their responses to our questionnaire and prioritize the importance of determinant factors considering their preferences. To validate the prioritized factors on the performance of DeepFakeR-MF, we conduct a sensitivity analysis applying Monte Carlo statistical modeling. Finally, our results confirm that DeepFakeR-MF provides effective strategic alternatives for policymakers, educators, media professionals, engineers, and netizens, hopefully reducing the socio-economic risks of deepfakes.}, archive = {J_EAAI}, author = {Milad Taleby Ahvanooey and Wojciech Mazurczyk and Zefan Wang and Jun Zhao}, doi = {10.1016/j.engappai.2025.111319}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111319}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel framework for assessing determinant risk factors on cyber (dis)trust behaviors of netizens in deepfakes}, volume = {159}, year = {2025}, }
• (2025). Mamba-360: Survey of state space models as transformer alternative for long sequence modelling: Methods, applications, and challenges. EAAI, 159, 111279. (www)

  @article{ , abstract = {Sequence modeling is a crucial area across various domains, including Natural Language Processing (NLP), speech recognition, time series forecasting, music generation, and bioinformatics. Recurrent Neural Networks (RNNs) and Long Short Term Memory Networks (LSTMs) have historically dominated sequence modeling tasks like Machine Translation, Named Entity Recognition (NER), etc. However, the advancement of transformers has led to a shift in this paradigm, given their superior performance. Yet, transformers suffer from O ( N 2 ) attention complexity and challenges in handling inductive bias. Several variations have been proposed to address these issues which use spectral networks or convolutions and have performed well on a range of tasks. However, they continue to struggle with managing long sequences. State Space Models(SSMs) have emerged as promising alternatives for sequence modeling paradigms in existing state space models. In this survey, we categorize the foundational SSMs based on three paradigms namely, Gating architectures, Structural architectures, and Recurrent architectures. This survey also highlights diverse applications of SSMs across domains such as vision, video, audio, speech, language (especially long sequence modeling), medical (including genomics), chemical (like drug design), recommendation systems, and time series analysis, including tabular data. Moreover, we consolidate the performance of SSMs on benchmark datasets like Long Range Arena (LRA), Text datasets, Image datasets, as well as video datasets and various time series datasets. The project page for Mamba-360 work is available on this webpage. https://github.com/badripatro/mamba360 .}, archive = {J_EAAI}, author = {Badri Narayana Patro and Vijay Srinivas Agneeswaran}, doi = {10.1016/j.engappai.2025.111279}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111279}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mamba-360: Survey of state space models as transformer alternative for long sequence modelling: Methods, applications, and challenges}, volume = {159}, year = {2025}, }
• (2025). Dynamic boundary guided adversarial robust distillation to improve robustness. EAAI, 159, 111228. (www)

  @article{ , abstract = {The vulnerability of deep neural networks to adversarial sample attacks seriously affects their reliability and safety in practical applications. By introducing finely designed, imperceptible perturbations to normal samples, the model can be induced to make incorrect judgements, potentially resulting in serious consequences. Existing adversarial defense methods are typically confined to defending against a single specific adversarial attack, and the generated adversarial samples often lack sufficient robustness features. Moreover, enhancing adversarial robustness while maintaining standard accuracy remains a significant challenge. This study meticulously examined the relationship between the state of the decision boundary and model performance. The results revealed that the samples located near the decision boundary are crucial for improving the robustness of the model. On this basis, a novel decision boundary-guided adversarial training method was developed. Further, a dynamic boundary-guided adversarial robust distillation framework was designed based on dynamic boundary-guided distillation. The framework adopts the standard knowledge distillation technique to train the student network at the initial stage, and introduces high-quality boundary samples at the later stage to improve the model’s adversarial robustness and accelerate convergence. Meanwhile, an adaptive strategy for adjusting the decision boundary was devised to dynamically adjust the weight proportion of different loss functions to balance the standard accuracy and adversarial robustness of the deep learning model. Experimental results show that the proposed method improved the average adversarial robustness by 6.17% on multiple datasets and models. Under the comprehensive AutoAttack method, the proposed method achieved a significant improvement of 5.98%, demonstrating its superiority in complex scenarios.}, archive = {J_EAAI}, author = {Fengyu Yang and Hong Yan and Yu Xiong and Xiaohui Wei and Yihui Zhong}, doi = {10.1016/j.engappai.2025.111228}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111228}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic boundary guided adversarial robust distillation to improve robustness}, volume = {159}, year = {2025}, }
• (2025). A survey on state-of-the-art deep learning applications and challenges. EAAI, 159, 111225. (www)

  @article{ , abstract = {Deep learning, a branch of artificial intelligence, is a data-driven method that uses multiple layers of interconnected units or neurons to learn intricate patterns and representations directly from raw input data. Empowered by this learning capability, it has become a powerful tool for solving complex problems and is the core driver of many groundbreaking technologies and innovations. Building a deep learning model is challenging due to the algorithm’s complexity and the dynamic nature of real-world problems. Several studies have reviewed deep learning concepts and applications. However, the studies mostly focused on the types of deep learning models and convolutional neural network architectures, offering limited coverage of the state-of-the-art deep learning models and their applications in solving complex problems across different domains. Therefore, motivated by the limitations, this study aims to comprehensively review the state-of-the-art deep learning models in computer vision, natural language processing, time series analysis and pervasive computing, and robotics. We highlight the key features of the models and their effectiveness in solving the problems within each domain. Furthermore, this study presents the fundamentals of deep learning, various deep learning model types and prominent convolutional neural network architectures. Finally, challenges and future directions in deep learning research are discussed to offer a broader perspective for future researchers.}, archive = {J_EAAI}, author = {Mohd Halim Mohd Noor and Ayokunle Olalekan Ige}, doi = {10.1016/j.engappai.2025.111225}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111225}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A survey on state-of-the-art deep learning applications and challenges}, volume = {159}, year = {2025}, }
• (2025). Addressing anomalies in smart grids: Methods for detecting and localizing high-current loads. EAAI, 159, 111223. (www)

  @article{ , abstract = {The concept of a smart grid encompasses a wide range of advanced technologies that surpass the capabilities of traditional power grids, enabling enhanced monitoring, control and efficiency. In this context, the term “anomaly” denotes unusual or unexpected occurrences, such as abnormal consumption patterns, infrastructure failures, power outages, cyber attacks, or energy theft. Anomaly detection is a critical aspect of improving the reliability and operational efficiency of Smart grid systems. In this study, we focus on detecting high current loads as anomalies in low voltage networks and present a system designed for this purpose. The system architecture includes distribution transformers and customer meters equipped with current sensors, radio frequency (RF) modems for wireless data transmission, and a central management server for data analysis and storage. The system employs machine learning (ML) algorithms for real-time anomaly detection and localization. Therefore, in the feature extraction phase, three distinct methods are utilized: Pattern Analysis, Discrete Wavelet Transform (DWT), and Fast Walsh–Hadamard Transform (FHWT). The Pattern Analysis method achieved the best performance using the Matern Gaussian Process Regression (MGPR) approach, whereas the DWT and FHWT methods yielded the most accurate results with the Optimizable Boosting Method. The results of the Pattern Analysis method indicate a test root mean square error (RMSE) of 69.36, a coefficient of determination (R 2 ) of 0.97, and a mean absolute error (MAE) of 45.40. The DWT method, employing a five-level wavelet transform with Daubechies 18 as the main wavelet, achieved a test RMSE of 53.44, an R 2 of 0.98, and an MAE of 37.25. The FHWT method resulted in a test RMSE of 83.99, an R 2 of 0.95, and an MAE of 53.71. Among these methods, the DWT method demonstrated the highest accuracy, with an average error rate of 3.02%, while the Pattern Analysis method and the FHWT method exhibited error rates of 4.09% and 4.86%, respectively. This innovative anomaly detection and localization approach provides a robust foundation for future research and development in power distribution networks, with the potential to reduce energy losses and improve grid stability. This system holds significant promise for reducing energy losses, preventing equipment damage, and enhancing grid stability by enabling rapid identification of unauthorized consumption and fault conditions. Its real-time monitoring capability empowers utilities to mitigate risks associated with energy theft and transient overloads, directly contributing to operational cost savings and improved service reliability.}, archive = {J_EAAI}, author = {Onder Civelek and Sedat Gormus and H.İbrahim Okumus and Hasan Yilmaz}, doi = {10.1016/j.engappai.2025.111223}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111223}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Addressing anomalies in smart grids: Methods for detecting and localizing high-current loads}, volume = {159}, year = {2025}, }
• (2025). Agent deception via polynomial path planning. EAAI, 159, 111205. (www)

  @article{ , abstract = {Deceptive behavior involves an intelligent agent creating plans that conceal its true intentions while appearing to pursue different goals. It is crucial for inducing confusion in various applications, including security, military, and competitive environments, where the ability to conceal true intentions can lead to significant strategic advantages. Developing better artificial intelligence (AI) for adversarial environments or strategic decision-making scenarios (e.g., more realistic testing of human or AI decision-making capabilities in games and simulations) requires effective deceptive planning. In this paper, we propose a novel polynomial path planner that enables an agent to deceive its observers. Our contributions include the following: (i) we develop a framework for obtaining ambiguous functions; (ii) we introduce new deception metrics; (iii) we present a method for standardizing trajectories to enable shape-based comparisons independent of speed; (iv) we conduct a human survey to evaluate deception and its relation to goal recognition; and (v) we outperform the state of the art on a multitude of deception metrics. Furthermore, our findings show that using more complex functions and increasing the level of misdirection greatly enhances agent deception effectiveness.}, archive = {J_EAAI}, author = {Nolan B. Gutierrez and Brian M. Sadler and William J. Beksi}, doi = {10.1016/j.engappai.2025.111205}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111205}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Agent deception via polynomial path planning}, volume = {159}, year = {2025}, }
• (2025). A disc-spherical fuzzy combined compromise solution for efficient smart grid energy management. EAAI, 159, 111196. (www)

  @article{ , abstract = {The imperative inclusion of sustainable and renewable energy sources into smart grid networks underscores the need for resilient and eco-friendly energy infrastructures. The problem addressed in this paper is the selection of the most effective technique for integrating renewable energy into smart grids using decision-making frameworks. Even with the progress made in integrating renewable energy, the complexity and unpredictability of choosing the optimum energy sources are still difficult for current decision-making frameworks to adequately represent. To solve this, the study aims to develop an integrated multiple attribute decision-making (MADM) methodology by utilizing a combined compromise solution (CoCoSo) and disc spherical fuzzy (D-SF) approach. The introduction of innovative logarithmic-based operations enhances the efficiency of D-SF sets. Novel operational laws for D-SF numbers are established, and corresponding aggregation operations as specified by these laws are proposed, exploring their excellent properties. The D-SF method employs the criteria importance through intercriteria correlation (CRITIC) method for attribute weight computation. Lastly, a MADM approach is presented, namely the CRITIC-based D-SF CoCoSo method, to rank the best energy sources for smart grid network integration. A comparative study is carried out among the suggested approach and current techniques to assess its feasibility and practicability. Furthermore, a sensitivity investigation shows the resilience of the proposed methodology. Our findings reinforce the critical role of renewable resources in achieving sustainable energy goals and demonstrate how smart energy grids enhance efficiency, reliability, and adaptability in managing diverse energy sources through the proposed decision-making methodology.}, archive = {J_EAAI}, author = {Shahzaib Ashraf and Wania Iqbal and Muhammad Naeem and Vladimir Simic and Bushra Batool and Dragan Pamucar}, doi = {10.1016/j.engappai.2025.111196}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111196}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A disc-spherical fuzzy combined compromise solution for efficient smart grid energy management}, volume = {159}, year = {2025}, }
• (2025). Automated strabismus diagnosis: A deep learning approach using cover test video analysis. EAAI, 159, 111161. (www)

  @article{ , abstract = {A condition known as strabismus occurs when one eye’s line of sight cannot fix on the target item that both eyes are picturing; when one eye fixes its gaze on the object, the other eye’s line of sight is in a different direction. Considering the characteristics of the disturb, the number of specialized doctors available to the public, and recent computer advancements recently, this study presents a computational method, applying artificial intelligence, for detecting and diagnosing strabismus using a convolutional neural network (CNN) You Only Look Once Version 5 (YOLOv5) for eyes and occluder detection in digital exam videos of the Cover Test. We conducted the study using two video databases: Video Dataset Cover Test I (CV-I), which includes 13 volunteer videos, and Video Dataset Cover Test II (CV-II), which comprises 57 volunteer videos. In measuring horizontal deviations, the method showed a maximum error of 7.28 Δ , below the literature-defined error limit of 8 Δ . For the CV-I dataset, the method achieved an accuracy of 100%, with both specificity and sensitivity also at 100% for horizontal strabismus. In the CV-II dataset, the method produced a sensitivity of 66.67%, a specificity of 100%, and an accuracy of 99.33% when focusing solely on horizontal measurements. When we considered both horizontal and vertical measurements, we found a sensitivity of 66.67% (with a 95% confidence interval [0.63; 0.99]), a specificity of 94.44% (with a 95% confidence interval [0.84; 0.98]), and an overall accuracy of 93.89% (with a 95% confidence interval [0.84; 0.98]).}, archive = {J_EAAI}, author = {Robert Douglas de Araújo Santos and João Dallyson Sousa de Almeida and Jorge Antonio Meireles Teixeira and Thales Levi Azevedo Valente and Geraldo Braz Jr. and Anselmo Cardoso de Paiva}, doi = {10.1016/j.engappai.2025.111161}, journal = {Engineering Applications of Artificial Intelligence}, month = {11}, pages = {111161}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated strabismus diagnosis: A deep learning approach using cover test video analysis}, volume = {159}, year = {2025}, }
• (2025). Cost-focused intelligent reservoir logging interpretation: Evaluation of cost-sensitive active learning strategies. EAAI, 158, 111630. (www)

  @article{ , abstract = {Geophysical logging interpretation directly affects the reliability of hydrocarbon reservoir evaluation and development decisions. Machine learning and data mining have significantly advanced the progress of intelligent logging interpretation. However, existing interpretive models have two main limitations. Firstly, supervised models rely on numerous labeled samples, but limited interpreted core data are difficult to meet the demand. Then, the neglect of misclassification costs can have serious consequences, particularly when misidentifying key interpretation targets. Therefore, there is an urgent need to conduct research on low-cost interpretive models for reservoir evaluation. In this study, cost-sensitive active learning (CSAL) strategy is introduced and evaluated for its applicability and effectiveness in logging interpretation. The active learning paradigm queries high-value samples as the way to reduce the labeling requirements for core interpretation. By introducing the cost-sensitive matrix in query strategies, the iteration process aims to optimize the misclassification cost. In comprehensive experiments on two typical interpretation tasks including gas reservoir identification and lithology identification, CSAL performs well and significantly outperforms classical active learning and random sampling. In terms of specific methods, the iterative effects of cost-weighted minimum margins (CWMM) and maximum expected cost (MEC) are more stable in both tasks. In particular, CWMM successfully indicates the gas-water distribution with the lowest cost (2407) and the highest accuracy (87.56 %) in validation well tests for gas reservoir identification. The results show that CSAL is able to perform more rational decisions and provides a new strategy for constructing more applicable low-cost models for intelligent logging interpretation.}, archive = {J_EAAI}, author = {Xinyi Zhu and Hongbing Zhang and Quan Ren and Lingyuan Zhang and Yihang Ge}, doi = {10.1016/j.engappai.2025.111630}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111630}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cost-focused intelligent reservoir logging interpretation: Evaluation of cost-sensitive active learning strategies}, volume = {158}, year = {2025}, }
• (2025). Task-adaptive learning with gradient weighting for bearing fault identification under small sample conditions. EAAI, 158, 111612. (www)

  @article{ , abstract = {Deep learning has demonstrated remarkable success in fault identification tasks; however, its performance often degrades under small sample conditions due to its dependence on large-scale labeled data. Meta-learning (defined as a learn-to-learn paradigm) offers a promising solution for small sample learning, yet it is challenged by task imbalance caused by disparities in task importance, complexity, and quantity, which impairs generalization. In this paper, a meta-gradient weighting meta-learning (MGWML) approach is proposed for small sample rolling bearing fault diagnosis. A meta-optimization framework is proposed to dynamically adjust task weights using meta-gradient information, with a dual-loop optimization strategy implemented to enhance convergence and adaptability across diverse task distributions. Concurrently, a base learner based on parallel attention is constructed to extract both local features and global dependencies from vibration signals. The effectiveness and robustness of MGWML is validated on two bearing fault datasets. In cross-working condition fault diagnosis, MGWML achieves a maximum accuracy of 99 %, and it maintains over 90 % accuracy under 5 dB noise interference.}, archive = {J_EAAI}, author = {Xianze Li and Qingtao Yao and Ling Xiang and Guopeng Zhu and Aijun Hu}, doi = {10.1016/j.engappai.2025.111612}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111612}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Task-adaptive learning with gradient weighting for bearing fault identification under small sample conditions}, volume = {158}, year = {2025}, }
• (2025). Hyperspectral imaging for rapid impurity detection in power system liquids. EAAI, 158, 111607. (www)

  @article{ , abstract = {The chemical integrity of power system liquids, such as coolants and transformer oils, is critical for the reliable operation of energy systems. Contaminants such as carbon, iron, copper, and tin can compromise cooling efficiency, increase failure risks, reduce equipment lifespan, and cause electrical malfunctions, thereby threatening the safety and stability of these systems. This study presents an innovative approach that integrates hyperspectral imaging (HSI) with machine learning (ML) algorithms to identify and quantify impurities in these liquids. A weighted ensemble model, referred to as the WeightedEnsemble_L2 model, has been developed and optimized. This model utilizes thirteen advanced machine-learning algorithms to identify impurities by analyzing spectral signatures across a broad wavelength range. The implemented artificial intelligence (AI) model demonstrates 90 % accuracy on the training set and 87.53 % on the validation set. This novel approach offers a robust solution for impurity detection in power system liquids, supporting predictive maintenance and enhancing the safety and stability of energy systems through the practical application of AI technology.}, archive = {J_EAAI}, author = {Liang Xue and Li Zhang and Zhuoyue Yang and Youhua Jiang and Chao Jiang and Haoyang Cui}, doi = {10.1016/j.engappai.2025.111607}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111607}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hyperspectral imaging for rapid impurity detection in power system liquids}, volume = {158}, year = {2025}, }
• (2025). Enhanced underwater acoustic target recognition using parallel dual-branch network with attention mechanism. EAAI, 158, 111603. (www)

  @article{ , abstract = {Ship-radiated noise serves as a crucial source of underwater acoustic signals for vessel classification, but its identification is often hindered by environmental variability and internal noise interference. To address these challenges, we propose a dual-branch Residual Attention-Long Short-Term Memory (ResA-LSTM) network for underwater acoustic target recognition. The proposed model integrates a Residual Attention (ResA) branch to extract spatial features and a Bidirectional Long Short-Term Memory (Bi-LSTM) branch to capture long-term temporal dependencies from Mel spectrograms. The ResA module incorporates attention mechanisms and residual connections to enhance feature selection and improve robustness in noisy environments. Evaluations conducted on two public datasets, ShipsEar and DeepShip, demonstrate the effectiveness of our approach, achieving classification accuracies of 98.55 % and 99.31 %, respectively. Sensitivity analysis further confirms the model's ability to handle long-duration acoustic sequences, highlighting its potential for practical deployment in real-world underwater recognition tasks.}, archive = {J_EAAI}, author = {Jingpu Xu and Xiaowei Li and Dan Zhang and Yaoran Chen and Yan Peng and Wenhu Liu}, doi = {10.1016/j.engappai.2025.111603}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111603}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced underwater acoustic target recognition using parallel dual-branch network with attention mechanism}, volume = {158}, year = {2025}, }
• (2025). A gated convolutional neural network-based tool remaining useful life prediction method. EAAI, 158, 111593. (www)

  @article{ , abstract = {Accurate and effective prediction of the remaining useful life (RUL) of the tools helps to rationalize the use and replacement of the tools, improve tools' utilization, and ensure the stable operation of the machining process, to enhance machining accuracy and productivity. Therefore, this article presents a method for the prediction of the RUL of cutting tools from vibration signals during the milling processing. Firstly, a new gated convolutional neural network is designed and constructed to establish a mapping connection from vibration signals to the tools' RUL. Additionally, a multi head attention mechanism is included to suppress the impact of weak correlation and redundant features, thus boosting the model's predictive ability. Further, the viability and efficacy of the proposed method are verified through the milling experiment. Compared with different deep learning models, this method can more accurately and reliably predict the tool RUL. Quantitative analysis of evaluation indicators shows that it has higher stability and generalization ability, thereby highlighting its superiority in predicting tools' RUL. According to the average calculation results of triple cross-validation, compared with other methods, the mean absolute error (MAE) of this method is reduced by 62.28 %, 38.05 %, and 20.72 %, the root mean square error (RMSE) reaches 62.31 %, 48.54 %, and 24.70 %, respectively, and the coefficient of determination (R 2 ) exceeds 0.98. Finally, this method is visualized through hybrid programming in LabVIEW and MATLAB. The research results provide the theoretical basis for the intelligent monitoring technology of the tools' RUL, which is essential for enhancing machining quality and productivity.}, archive = {J_EAAI}, author = {Shilong Zhou and Yaonan Cheng and Jing Xue and Rui Guan and Xingwei Zhou and Wenjie Zhai}, doi = {10.1016/j.engappai.2025.111593}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111593}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A gated convolutional neural network-based tool remaining useful life prediction method}, volume = {158}, year = {2025}, }
• (2025). An evacuation guidance approach based on multi-agent shared Q-learning. EAAI, 158, 111587. (www)

  @article{ , abstract = {Crowd evacuation dynamics have garnered significant research attention in recent years due to their critical impact on public safety. Guided evacuation systems have emerged as a promising solution for optimizing evacuation time and enhancing evacuation efficiency in emergency scenarios. To address these challenges, this study proposes a novel multi-agent collaborative evacuation guidance framework utilizing shared Q-learning. The proposed architecture employs coordinated agents to direct pedestrian flows toward exits through strategic cooperation. Specifically, we implement a group-based evacuation mechanism where only designated leaders receive optimized routing instructions, thereby minimizing haphazard movements among crowd members. To enable effective path planning, our system synthesizes localized evacuation data into comprehensive global information through multi-agent knowledge sharing, thereby generating optimized paths for leaders. To evaluate a path, we define a reward function incorporating four key factors: distance, density, traveling time, and selection frequency of route sections. A path-planning algorithm is developed to provide paths with a high reward for evacuees. Furthermore, to validate the proposed framework, a reciprocal velocity obstacle (RVO) model is employed to simulate crowd dynamics during evacuations. The experimental results demonstrate that the approach can shorten the total evacuation time and promote the balanced utilization of all available exits.}, archive = {J_EAAI}, author = {Yanbin Han and Hong Liu and Liang Li and Wenhao Li}, doi = {10.1016/j.engappai.2025.111587}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111587}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An evacuation guidance approach based on multi-agent shared Q-learning}, volume = {158}, year = {2025}, }
• (2025). A multi-sensor deep metric learning method for unknown anomaly detection in unmanned aerial vehicles. EAAI, 158, 111584. (www)

  @article{ , abstract = {In recent years, unmanned aerial vehicles (UAVs) have been widely adopted in various fields, and their safety issues have become increasingly critical. Additionally, the emergence of unknown anomalies can significantly heighten UAVs’ safety risks. However, existing anomaly detection methods cannot effectively handle unknown anomalies. Therefore, a multi-sensor deep metric learning method (MS-DML) is proposed here for the discovery and identification of unknown anomalies. MS-DML constructs a multi-sensor feature extraction network, which can extract features from multimodal heterogeneous data and map the inputs into a low-dimensional space. Subsequently, MS-DML adopts a deep metric learning (DML) framework and a metric learning algorithm is utilized to classify samples in the low-dimensional space. Meanwhile, MS-DML possesses the potential to address open-set problems with its DML architecture and an improved K-means algorithm is proposed to discover unknown anomalies. Afterwards, a model update algorithm is proposed based on K-nearest neighbor. It can update MS-DML with a small number of labeled unknown anomaly samples and without retraining. The updated MS-DML can accurately differentiate the remaining unlabeled unknown anomalies and the update cost is very low. Experimental results on five benchmark datasets show that MS-DML outperforms baselines in known anomaly detection, achieving improvements in accuracy by 1.30 %, 0.71 %, 0.10 %, 1.27 %, and 2.44 %. In unknown anomaly discovery, MS-DML achieves a maximum accuracy of 98.39 %, with recall and precision rates of 94.66 % and 97.68 %. In unknown anomaly classification, the detection accuracy exceeds 80 % for all unknown categories with only 5 % of the unknown anomalies labeled.}, archive = {J_EAAI}, author = {Jinbo Zhao and Xiaolong Xu}, doi = {10.1016/j.engappai.2025.111584}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111584}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-sensor deep metric learning method for unknown anomaly detection in unmanned aerial vehicles}, volume = {158}, year = {2025}, }
• (2025). Divide-and-rule and interpretable proactive threat analysis for mobile crowdsensing systems. EAAI, 158, 111583. (www)

  @article{ , abstract = {Mobile Crowdsensing is a new paradigm of sensing that utilizes mobile devices and crowd intelligence to collect, process and share environmental data. The large amount of high-value data contained in the sensing platform makes it a target for a variety of threats such as Witch attacks and Denial of Service attacks. It seriously interferes with the platform's sensing services and affects the quality of sensing data. Historical data containing attack information is modeled as a multivariate time series and a learnable and interpretable method is proposed for proactive forecasting of threats. At first, a “divide-and-rule” non-stationary factor processing method is proposed to normalize the attack data in slices. Learn the statistical characteristics of each slice individually and forecast the future distribution of these statistical characteristics through a designed statistical forecasting module. Subsequently, utilize the normalized data, which has removed non-stationary factors, as the standardized input for the threat forecasting module. Utilizing a contrastive learning framework, combined with the integration of a bidirectional cross-attention mechanism, learn the similarity coefficients between attack data and temporal bases. The integration of similarity coefficients with the future segments of the temporal bases results in the standardized output for this stage. Finally, based on the future forecast of the statistical characteristics by the statistical forecasting module, restore their non-stationary factors in the standardized output of the threat forecasting module, culminating in the result of threats forecasting. Through extensive experiments on six datasets, we demonstrate that the forecasting performance of our proposed model exceeds the previous state-of-the-art baseline methods by an average of 12.30 % and 7.12 % in Mean Square Error and Mean Absolute Error, respectively.}, archive = {J_EAAI}, author = {Delei Zhao and Jian Wang and Guosheng Zhao}, doi = {10.1016/j.engappai.2025.111583}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111583}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Divide-and-rule and interpretable proactive threat analysis for mobile crowdsensing systems}, volume = {158}, year = {2025}, }
• (2025). Context-aware emotion recognition through agent-scene interactions. EAAI, 158, 111581. (www)

  @article{ , abstract = {In real-world scenarios, context-aware emotion recognition (CAER) is a key problem in affective computing with broad application prospects. Most current CAER methods primarily rely on image-level contextual features. However, the interactive relationships between the agent and other objects within the scene are often overlooked or only partially modeled, which limits emotion recognition accuracy. To address this, we proposed a spatial interactive context-aware emotion network (ICENet) that consists of an agent feature extraction branch and a scene-context interaction branch. Specifically, the agent feature extraction branch aims to extract facial and posture features from the target agent and fuse them. In the facial feature extraction network named FaceNet, pure Convolutional Neural Network (ConvNeXt) is used as the backbone to extract global features, and a self-attention-based fine-grained feature extraction (FGFE) module is designed to capture more discriminative local features. In the posture feature extraction network, semantic segmentation is used to extract human silhouettes, which are then processed by Vision Transformer to obtain posture-related features. Meanwhile, the scene-context interaction branch named ObjNet integrates agent’s gaze angle and global depth maps to construct target agent-objects relationship in three-dimensional (TAR3D). Subsequently, a Graph Convolutional Network is employed to model the TAR3D and extract scene-context interaction features. Subsequently, a multiplicative fusion strategy is adopted to integrate agent features with scene-context interaction features, and emotion classification is performed based on the fused representation. Finally, experiments on EMOTIC and CAER-S datasets show that our approach outperforms current state-of-the-art methods in classification accuracy. The code is available at https://github.com/Cyx336/ICENet.git .}, archive = {J_EAAI}, author = {Yu-Xiang Chen and Hong-Mei Sun and Cheng-Yue Che and Shuo Feng and Rui-Sheng Jia}, doi = {10.1016/j.engappai.2025.111581}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111581}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Context-aware emotion recognition through agent-scene interactions}, volume = {158}, year = {2025}, }
• (2025). An asynchronous advantage learning-based novel defense method in microgrids compromised by hybrid cyber threats. EAAI, 158, 111580. (www)

  @article{ , abstract = {In this paper, a defense mechanism is developed to address the disruptive effects of cyber-attacks in an islanded isolated microgrids (μG) with various components. False data injection (FDI) and denial-of-service (DoS) attacks, as two prevalent cyber threats, are commonly implemented on cyber-physical systems to manipulate measurement signals transmitted during communication networks. To tackle cyber-attacks, a novel application of the machine-learning algorithm coordinated with Luenberger observer has been developed for dynamic frequency regulation in microgrids under FDI and DoS attacks. The proposed defense scheme is designed in two phases: i ) the Luenberger observer is utilized to estimate state parameters and residual signal of the test-system. The presence of cyber-attacks is identified by comparing the residual signal with a predefined threshold. ii ) an adaptive mitigation scheme based on the asynchronous advantage actor-critic (A3C) technique is also employed to block the falsified frequency signal. By interacting with the A3C-agent with the networked system, the parameters of the secondary controller are automatically adjusted to mitigate the effect of the anomaly on the measurement signal. The deep neural networks of A3C are trained to maximize a reward function in such a way that eliminates the destructive effects of cyber-attacks. Extensive examinations under various cyber security scenarios are conducted to appraise the feasibility of the proposed defense scheme. Comparative analysis reveals the supremacy of the proposed defense scheme over other state-of-the-art schemes. Comparative analysis reveals the supremacy of the proposed defense scheme over other state-of-the-art methods, which reflects an improvement of 94.96 % compared to Luenberger-based fractional order controller, 94.25 % compared to sliding mode control, 93.48 % compared to model predictive control, and 84.66 % compared to distributed economic model predictive control in terms of the integral of absolute error index under hybrid attacks.}, archive = {J_EAAI}, author = {Shuguang Li and Zhenxin Wen and Yanbo Ding}, doi = {10.1016/j.engappai.2025.111580}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111580}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An asynchronous advantage learning-based novel defense method in microgrids compromised by hybrid cyber threats}, volume = {158}, year = {2025}, }
• (2025). Detailed fault detection of industrial sensor based on semantic segmentation models. EAAI, 158, 111576. (www)

  @article{ , abstract = {Sensors play a critical role in modern condition monitoring systems. In harsh environments of industrial production, sensors are susceptible to faults, which affects the condition monitoring of production equipment. Detailed information of sensor faults, including location, range, and type, is crucial to the postprocessing tasks such as correction and attribution. However, only limited fault information can be figured out by existing methods. To address this issue, the semantic segmentation model is introduced into sensor fault detection for the first time. In this way, comprehensive fault information can be obtained by a single model. Moreover, the proposed semantic segmentation model is specially designed according to the unique data characteristics of this task. The wide information bottleneck and moderate multi-scale feature extraction enable the model to capture subtle features and reduce overhead. The proposed method has achieved points accuracy of 99.4 %, 98.4 %, and 97 % with less cost on three industrial datasets.}, archive = {J_EAAI}, author = {Xirui Chen and Hui Liu}, doi = {10.1016/j.engappai.2025.111576}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111576}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detailed fault detection of industrial sensor based on semantic segmentation models}, volume = {158}, year = {2025}, }
• (2025). Tuberculosis and pneumonia diagnosis in chest X-rays by large adaptive filter and aligning normalized network with report-guided multi-level alignment. EAAI, 158, 111575. (www)

  @article{ , abstract = {Tuberculosis (TB) and pneumonia remain major global public health challenges, necessitating accurate and efficient diagnostic tools. This study proposes a novel deep learning framework, Large Adaptive Filter and Aligning Normalized Network (LAFAN-Net), designed to improve chest X-ray (CXR) diagnosis by integrating visual and textual information. The framework comprises three key components: (1) a report-guided multi-level alignment mechanism that aligns CXR features with radiology reports at the token, sample, and disease levels; (2) a large adaptive filter block for capturing multi-scale visual patterns; and (3) AlignNorm, a new normalization technique that mitigates oversmoothing and enhances feature separation. LAFAN-Net is evaluated on three publicly available CXR datasets, achieving accuracies of 97.14 %, 95.35 %, and 89.39 %, and F1 scores of 90.77 %, 96.32 %, and 88.33 %, respectively. Extensive ablation studies confirm the model's robustness. The results underscore LAFAN-Net's ability to extract clinically meaningful features while maintaining interpretability, supported by singular value distributions and Gradient-weighted Class Activation Mapping visualizations. Future work will explore extending the model to broader disease categories and multi-class classification tasks to enhance clinical utility. In addition, improving computational efficiency and ensuring real-time applicability are essential for deployment in resource-limited settings.}, archive = {J_EAAI}, author = {Si-Yuan Lu and Ziquan Zhu and Yu-Dong Zhang and Yu-Dong Yao}, doi = {10.1016/j.engappai.2025.111575}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111575}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Tuberculosis and pneumonia diagnosis in chest X-rays by large adaptive filter and aligning normalized network with report-guided multi-level alignment}, volume = {158}, year = {2025}, }
• (2025). An effective feature enhancement detection network for building change detection in high-resolution remote sensing images. EAAI, 158, 111573. (www)

  @article{ , abstract = {Accurate building change detection (BCD) in high-resolution (HR) remote sensing imagery faces persistent challenges including edge misdetection, omission errors, and coarse segmentation. We propose an effective feature-enhanced detection network (DVM-Net) for BCD in HR remote sensing images. This framework addresses these limitations through three key innovations: a dual-channel structure for comprehensive detail extraction, a multi-scale down-sampling module to preserve features and reduce errors, and a visual attention mechanism for refined edge detection. Evaluations include ablation studies on the Wuhan University Building Change Detection Dataset (WHU_CD) and cross-dataset validation using the Google Earth Image Change Detection Dataset (Google_CD) and Large-Scale Earth Vision Infrastructure Change Detection Dataset (LEVIR_CD). DVM-Net achieves state-of-the-art performance, surpassing 11 existing methods with mean Intersection over Union (mIoU) scores of 81.71 %, 79.77 %, and 86.53 % on these benchmarks, respectively. Analysis on the SEmantic Change detectiON Dataset (SECOND) identifies limitations in complex urban environments. The open-source implementation ( https://github.com/AroundSeeYou/DVM_Net ) provides an advanced solution for HR BCD tasks.}, archive = {J_EAAI}, author = {Zhihuan Liu and Tingting Ren and Guoxiong Zhou and Xiaoli Cai and Yirui Zeng and Liujun Li and Shipeng Hu}, doi = {10.1016/j.engappai.2025.111573}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111573}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An effective feature enhancement detection network for building change detection in high-resolution remote sensing images}, volume = {158}, year = {2025}, }
• (2025). Interpretable and robust fault diagnosis of rotating machinery in noisy environments via improved high-order spatial interactions network. EAAI, 158, 111568. (www)

  @article{ , abstract = {According to the problems of the existing fault diagnosis (FD) model being affected by noise and lacking interpretability, this paper proposed an innovative FD model for rotating machinery, named noise critical layer adaptation (NCLA). By designing the module of noise robustness criticality (NRC), the model effectively focuses on layers most affected by noise, significantly improving classification accuracy and interpretability in noisy environments. Furthermore, this study designed an improved feature extraction framework based on the high-order spatial interactions with recursive gated wavelet convolution (WTConv) network, which enables the model to decompose and process signal components at different frequencies, enhancing its robustness and capability to capture fine-grained features. Unlike traditional models that rely on datasets with identical feature distributions, the proposed model was pre-trained on noise-free data and tested on noisy datasets, which aligns more closely with actual engineering applications. Experimental results of the two cases demonstrated that the model exhibits superior generalization and robustness across various noise conditions, outperforming conventional approaches. Additionally, by visualizing the impact of noise on critical layers, the proposed model addresses the limitations of the black box in deep learning methods.}, archive = {J_EAAI}, author = {Bin Wang and Pengfei Liang and Ying Li and Junhui Hu and Lijie Zhang}, doi = {10.1016/j.engappai.2025.111568}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111568}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpretable and robust fault diagnosis of rotating machinery in noisy environments via improved high-order spatial interactions network}, volume = {158}, year = {2025}, }
• (2025). Deeppipe: A physics-enhanced adaptive multi-modal fused neural network for predicting contamination length interval in multi-product pipeline. EAAI, 158, 111564. (www)

  @article{ , abstract = {During the operation of a multi-product pipeline, an accurate and effective prediction of contamination length interval is the central key to guiding the cutting plan formulation and improving the economic effect. However, the existing methods focus on extracting implicit principles and insufficient feature correlations in a data-driven pattern but overlook the potential knowledge in the scientific theory of contamination development, failing to provide practically useful results. Consequently, in this study, the holistic feature correlations and physical knowledge are extracted and integrated into the neural network to propose a physics-enhanced adaptive multi-modal fused neural network (PE-AMFNN) for contamination length interval prediction. In PE-AMFNN, a multi-modal adaptive feature fusion module is created to establish a comprehensive feature space with quantified feature importance, thus capturing sufficient feature correlations. Subsequently, a mechanism-coupled customized neural network is designed to incorporate the explicit scientific principle into the forward and backward propagation. Besides, a physics-embedded loss function, which introduces interval differences and interval correlation constraints, is established to unearth the latent physical knowledge in contamination development and force the model to draw physically unreasonable results. Validation on the real-world cases implies that the proposed model outperforms the start-of-the-art techniques and latest achievements, with Root Mean Squared Relative Errors reduced by 31 % and 36 % in lower and upper limit prediction. Furthermore, the sensitivity analysis of model modules suggests that both the multi-modal feature fusion and physical principles are crucial for model improvement.}, archive = {J_EAAI}, author = {Jian Du and Jianqin Zheng and Pengtao Niu and Shiyuan Pan and Qi Liao and Rui Qiu and Bin Xu and Ning Xu and Yongtu Liang}, doi = {10.1016/j.engappai.2025.111564}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111564}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deeppipe: A physics-enhanced adaptive multi-modal fused neural network for predicting contamination length interval in multi-product pipeline}, volume = {158}, year = {2025}, }
• (2025). An adaptive collaborative intrusion detection system for vehicular fog computing networks. EAAI, 158, 111563. (www)

  @article{ , abstract = {Vehicular fog computing networks (VFNs) are integral to modern intelligent transportation systems, providing critical infrastructure for real-time communication and data processing. However, these networks face growing security challenges, including time-sensitive attacks, location spoofing, and distributed denial-of-service (DDoS) attacks, which exploit the distributed nature of the network. Traditional collaborative intrusion detection systems (CIDS) struggle with issues such as low detection accuracy, high latency, and scalability, limiting their effectiveness in such dynamic environments. To address these challenges, we propose the adaptive collaborative intrusion detection system (A-CIDS), a novel approach that enhances both detection and real-time defense in VFNs. A-CIDS integrates a machine learning-driven detection unit with a modified probabilistic synchronous parallel (MPSP) module, enabling efficient real-time intelligence sharing and synchronization across fog nodes. The system incorporates advanced preprocessing techniques, variational autoencoders (VAE) for feature selection and dimensionality reduction, and a fine-grained extended isolation forest (FEIF) for accurate anomaly detection. A comprehensive evaluation using the vehicular reference misbehavior extension (VeReMi_Extension) and telemetry, operating systems, network – internet of things (ToN-IoT) datasets shows that A-CIDS outperforms existing systems, achieving a local detection accuracy of 99.2 %, the lowest false positive rate of 2 %, and a total latency of 150 ms. These results demonstrate the effectiveness of A-CIDS in securing VFNs against sophisticated threats, offering scalable and low-latency solutions for real-time intrusion detection in safety-critical vehicular environments.}, archive = {J_EAAI}, author = {Samuel Akwasi Frimpong and Mu Han and Wenyi Zheng and Ismatul Jannat Chowdhury}, doi = {10.1016/j.engappai.2025.111563}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111563}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive collaborative intrusion detection system for vehicular fog computing networks}, volume = {158}, year = {2025}, }
• (2025). Deep learning-based extension of gait segmentation to abnormal patterns using inertial measurement units. EAAI, 158, 111561. (www)

  @article{ , abstract = {Inertial measurement unit (IMU)-based gait segmentation is widely employed in medical applications and plays a crucial role in recognizing gait phases. However, existing methods primarily focus on normal gait patterns, limiting their applicability to pathological cases such as small-stepped, dragging, and circumduction gaits. In this study, we propose a novel gait segmentation framework that can effectively handle both normal and abnormal gait patterns, thereby enhancing generalization of medical applications. Our main contribution is to expand gait segmentation to abnormal gait patterns in two ways: (1) We propose a new definition of gait segmentation to ensure equal treatment of normal and abnormal gaits, facilitating a more inclusive approach. (2) We also propose a novel network called gait segmentation neural network (GaitSeg Net), a deep learning model that integrates a convolutional neural network, bidirectional long short-term memory and transformer for robust feature extraction. This architecture employs wide-kernel CNNs to mitigate noise-related issues and a convolutional feedforward layer to filter out irrelevant information, significantly improving segmentation accuracy. We recorded a new dataset encompassing various normal and abnormal gaits for training and validation. Experimental results demonstrate that GaitSeg Net outperforms existing methods, achieving an F1-score of 98.16 %. Compared to a previous study, our method improves accuracy from 96.88 % to 97.50 % in walking and running tasks. Furthermore, our model maintains high accuracy for abnormal gaits (small-stepped gait: 96.1 %, dragging gait: 96.6 %, circumduction gait: 97.6 %), confirming its robustness. These results highlight the potential of our approach in extending gait segmentation to pathological movement patterns, marking a significant advancement in both artificial intelligence applications and biomedical engineering.}, archive = {J_EAAI}, author = {Changyu Zhao and Yuanjian Jin and Ruoding An and Hirotaka Uchitomi and Yoshihiro Miyake}, doi = {10.1016/j.engappai.2025.111561}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111561}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning-based extension of gait segmentation to abnormal patterns using inertial measurement units}, volume = {158}, year = {2025}, }
• (2025). A robust framework for interference-resistant surface crack detection and measurement in engineering structures. EAAI, 158, 111557. (www)

  @article{ , abstract = {Modern engineering structures demand efficient and precise structural health monitoring to ensure safety and longevity, particularly in civil engineering where the intelligent detection and analysis of apparent cracks is critical. This study presents an integrated framework that synergizes advanced artificial intelligence with innovative imaging techniques to achieve robust, interference-resistant detection and measurement of surface cracks in civil engineering. We refine a mobile imaging system, with industrial cameras, precision zoom lenses, and a laser rangefinder, to secure stable data acquisition and accurate pixel-to-size conversion under challenging conditions, thereby providing a foundation for artificial intelligence driven crack detection. On the algorithm side, two optimized deep learning architectures are developed, leveraging depthwise separable convolutions and compound scaling to enhance precision and robustness. Rigorous evaluations confirm that tailored loss functions significantly improve convergence and accuracy. Across various lighting and background conditions, the proposed framework achieves an average crack pixel extraction accuracy of over 90 %, showing strong resistance to visual interference. Comparative experiments in multiple real-world scenarios demonstrate that it outperforms seven existing approaches in both accuracy and efficiency. In addition, we introduce a morphological transformation, which based algorithm links pixel-level detection with measurement of crack geometry and branching. Unlike previous studies, we consider the impact of imaging distance and camera shaking during field operation. The results remain accurate and stable even when the camera swings within 26°at different distances, highlighting practical robustness. Overall, this work advances artificial intelligence applications in fault detection and analysis, offering a practical solution for engineering maintenance.}, archive = {J_EAAI}, author = {Jiangfeng Liu and Shijia Ma and Zhipeng Wang and Zhijie Jian and Tengyuan Zhang and Ruinian Sun}, doi = {10.1016/j.engappai.2025.111557}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111557}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust framework for interference-resistant surface crack detection and measurement in engineering structures}, volume = {158}, year = {2025}, }
• (2025). An unsupervised anatomy-aware dual-constraint cascade network for lung computed tomography deformable image registration. EAAI, 158, 111548. (www)

  @article{ , abstract = {Deformable image registration (DIR) of lung computed tomography (CT) facilitates precise radiotherapy for lung cancer by aligning images from different respiratory states or time points. Due to the high resolution of lung CT image data, the large amplitude of lung motion, and significant differences in image intensity, achieving fast and accurate DIR of lung CT is challenging. In this study, we propose an unsupervised Anatomy-aware Dual-constraint Cascade network, called ADC-Net. We construct an efficient single-level network based on fully convolutional networks, into which a tri-directional deformation capture (TDC) block is designed and integrated. The TDC block calculates voxel correlations along three spatial dimensions separately, comprehensively capturing deformations in different directions, particularly large deformations along the lung's vertical axis. To further enhance the awareness of large deformations as well as internal anatomical details, we propose a dual-constraint cascade mode. In this mode, contour-enhanced and vessel-enhanced images used as constraints, along with the original images, are fed into three single-level networks, which enables accurate registration of large deformations while maintaining precise alignment of internal details. Compared to the state-of-the-art cascade method, MANet, ADC-Net achieves a registration accuracy improvement of 3 % and 4 % on the Dir-lab and ThoraxCBCT datasets, respectively, with a notable accuracy increase of 16 % for images with large deformations. Additionally, the low-computational-cost upsampling structure adopted by the single-level network ensures efficiency and makes ADC-Net 21 % faster than MANet. ADC-Net's high accuracy, rapid processing speed, and robust applicability in DIR of lung CT suggest its potential value for clinical engineering applications.}, archive = {J_EAAI}, author = {Wenbin Wu and Yifan Gao and Xin Jin and Rui Zhang and Yuemei Pan and Xin Gao}, doi = {10.1016/j.engappai.2025.111548}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111548}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An unsupervised anatomy-aware dual-constraint cascade network for lung computed tomography deformable image registration}, volume = {158}, year = {2025}, }
• (2025). Generalized deep neural network for seismic site response prediction with transfer learning. EAAI, 158, 111546. (www)

  @article{ , abstract = {Accurate prediction of site-specific seismic responses plays a pivotal role in evaluating earthquake effects on infrastructure. Traditional physics-based methods suffer from inherent model assumptions, significant parameter uncertainty, and high computational costs. This study proposes a generalized deep neural network that integrates seismic motion data and site information to predict three-directional seismic responses across various site types. Trained on an extensive dataset of recorded data from Kiban Kyoshin Network in Japan, the model demonstrated excellent performance on the test set, with correlation coefficients reaching 97 % between the predicted and target results. Utilizing transfer learning techniques, it was adapted to seismic response prediction at new sites not included in the training set. Compared to the state-of-the-art finite element method, the retrained model significantly improved prediction accuracy, with an overall average error reduction of approximately 50 %. Additionally, the model effectively captured the nonlinear response characteristics of a site during strong seismic events without any strong motion data to retrain. The proposed model demonstrated superior prediction accuracy, higher computational efficiency, and stronger generalization capabilities compared to traditional physics-based models.}, archive = {J_EAAI}, author = {Lin Li and Feng Jin and Duruo Huang and Chunhui He and Fulong Ma}, doi = {10.1016/j.engappai.2025.111546}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111546}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generalized deep neural network for seismic site response prediction with transfer learning}, volume = {158}, year = {2025}, }
• (2025). Retrieval-augmented visual parcel invoice understanding transformer for address correction. EAAI, 158, 111542. (www)

  @article{ , abstract = {In automated postal and logistics operations, Visual Parcel Invoice Understanding technology (VPIU) addresses the challenging task of recognizing named entities such as addresses, names, and phone numbers. In particular, the VPIU performance is significantly reduced when the addresses are extracted. Therefore, we propose a novel Retrieval-augmented-VPIU model (Re-VPIU). The proposed Re-VPIU consists of a Document Understanding Transformer (DONUT) for information extraction and a Large Language Model (LLM) with Retrieval-Augmented Generation (RAG) for information correction. First, using an end-to-end approach, the DONUT extracts information regarding addresses, names, and phone numbers from the invoice images. Next, the RAG detects and corrects errors in the extracted address information according to the Korean address database and prompts. Proposed Re-VPIU model is the first application of RAG in the context of visual document understanding. Unlike traditional models, Re-VPIU does not require retraining to incorporate new address data, as it dynamically integrates external knowledge through retrieval. Furthermore, it enables both character- and word-level corrections simultaneously, leveraging the linguistic capabilities of LLMs. To validate the proposed method's effectiveness, several comparative experiments were conducted using a Korean invoice dataset. The experiments showed that the proposed method achieved an F1 Score of 83.42 % when evaluating the accuracy of the complete extraction, which was 12.17 % higher than that of the current state-of-the-art model.}, archive = {J_EAAI}, author = {Yeon-Bi Jeong and Hong-il Seo and Yoo-Hwan Kim and Won-Yeol Kim}, doi = {10.1016/j.engappai.2025.111542}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111542}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Retrieval-augmented visual parcel invoice understanding transformer for address correction}, volume = {158}, year = {2025}, }
• (2025). Complex system modeling using deviation-smoothing belief rule base with training and optimization. EAAI, 158, 111539. (www)

  @article{ , abstract = {In the operation of complex systems, damage caused by aging and other factors can significantly impact their safety. Consequently, rational modeling approaches are essential for ensuring the stable and reliable operation of such systems. Due to their intricate internal structures and the influence of external environments, models of complex systems confront numerous uncertainties. The belief rule base (BRB), as an effective artificial intelligence (AI) tool for knowledge representation and reasoning, is capable of handling uncertain information through data-driven and knowledge-driven approaches, making it suitable for complex system modeling. However, existing BRB modeling approaches consider only a single interval when calculating rule matching degrees, which can lead to the issue of zero activation of rules; additionally, some rules may lack relevance and rationality in actual systems, thereby increasing model complexity and affecting performance due to redundancy. Therefore, a complex system modeling system based on deviation-smoothing belief rule base (DS-BRB) has been proposed. Firstly, this approach employs a matching degree calculation approach that integrates bias computation with smoothing techniques, enabling effective rule activation over a broader range of intervals. Subsequently, a specialized rule reduction approach designed for rule-level analysis is utilized, iteratively reducing rules based on their weights. Finally, the proposed methodology is evaluated using pipeline leakage detection and various public benchmark datasets. Finally, experiments on pipeline leakage detection and benchmark datasets demonstrate that DS-BRB resolves zero activation, achieves higher modeling accuracy with a reduced rule base, and highlights BRB's potential as an interpretable AI tool for complex system modeling.}, archive = {J_EAAI}, author = {Chao Sun and Jiahao Mai and Wei He and Hailong Zhu and Qi Liu}, doi = {10.1016/j.engappai.2025.111539}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111539}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Complex system modeling using deviation-smoothing belief rule base with training and optimization}, volume = {158}, year = {2025}, }
• (2025). A temporal cross-contrastive self-supervised learning framework for high-speed train bearing fault diagnosis: Addressing limited labeling and speed variability. EAAI, 158, 111537. (www)

  @article{ , abstract = {High-speed train (HST) bogie bearings operating under variable speed conditions exhibit significant domain distribution shifts. Moreover, the scarcity of labeled data in HST systems significantly compromises the accuracy and generalization capability of intelligent fault diagnosis methods. To address these challenges, we propose a novel self-supervised learning framework based on temporal cross-contrastive prediction (TCCP). In the upstream pretext task, an autoregressive prediction module is designed to learn domain-invariant representations from temporal data. Furthermore, we introduce a cross-prediction module to enhance the robustness of these domain-invariant representations. The temporal contrastive module is jointly optimized with the cross-prediction and autoregressive prediction modules to learn robust discriminative representations. In the downstream task, limited labeled data is utilized to fine-tune the pre-trained model for accurate fault diagnosis. The effectiveness of the proposed TCCP framework was extensively validated using both a public variable-speed bearing dataset and a self-built variable-speed HST bogie bearing dataset under limited labeled sample conditions. Experimental results demonstrate that our proposed method achieves superior fault diagnosis performance, yielding improvements in accuracy ranging from 0.86 % to 7.36 % on the HST bogie bearing dataset.}, archive = {J_EAAI}, author = {Yuyan Li and Tiantian Wang and Jingsong Xie and Jinsong Yang and Tongyang Pan and Buzhao Niu and Xiang Yu}, doi = {10.1016/j.engappai.2025.111537}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111537}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A temporal cross-contrastive self-supervised learning framework for high-speed train bearing fault diagnosis: Addressing limited labeling and speed variability}, volume = {158}, year = {2025}, }
• (2025). Transfer learning on stacked machine-learning model for predicting pull-out behavior of steel fibers from concrete. EAAI, 158, 111533. (www)

  @article{ , abstract = {Fiber-reinforced concrete (FRC) can enhance the concrete's resistance to cracking, impact, and fatigue, making it suitable for high-stress applications and infrastructures. Application of artificial intelligence in engineering problems can boost the design procedure while reducing the experimental cost and time. The aim of this study is to provide a transfer learning-based stacked machine learning (ML) model for predicting average and equivalent bond strengths, pull-out energy, maximum load, and corresponding slip capacity of steel fiber from concrete. A data-rich framework, including 472 data points, was constructed based on a comprehensive literature review. The overall results show that the fine-tuned bagging regression model provided the most accurate predictions, whereas artificial neural networks (ANNs) and gradient boosting ML models were not sufficiently reliable. To overcome the prediction shortcoming, experimental test was done to prepare unseen dataset for validation, and the transfer learning-based stacked ML model proposed to estimate all of the output targets in one model, with the highest accuracy (i.e., R 2 of 99.4 %) and the lowest error values (i.e., average error of 0.25 %) amongst those 18 ML models investigated in this research. Then, a graphical user interface (GUI) was provided to ease similar prospective analyses and obtain predictions, while proposing the update in the database for future studies. The proposed model paves the way for the prediction and evaluation of the fiber pull-out performance.}, archive = {J_EAAI}, author = {Torkan Shafighfard and Neda Asgarkhani and Farzin Kazemi and Doo-Yeol Yoo}, doi = {10.1016/j.engappai.2025.111533}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111533}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transfer learning on stacked machine-learning model for predicting pull-out behavior of steel fibers from concrete}, volume = {158}, year = {2025}, }
• (2025). State of charge estimation of lithium-ion batteries using improved multi-attention long short-term memory extended kalman filtering model. EAAI, 158, 111526. (www)

  @article{ , abstract = {An improved multi-attention long short-term memory extended Kalman filtering (MALSTM-EKF) model is proposed for high-precision state of charge (SOC) estimation of lithium-ion batteries. The model integrates a multi-head attention mechanism into the long short-term memory (LSTM) to enhance temporal feature extraction and accelerate learning by focusing on relevant information at each time step. The MALSTM model first provides an initial SOC estimation, which is then refined through an EKF for noise filtering and state correction, ensuring enhanced accuracy and robustness under dynamic conditions. This combination provides higher accuracy and stability, especially in complex battery conditions. The University of Maryland's public dataset for the A123 18650 LiFePO4 battery validates the model, covering diverse temperatures and complex working conditions. Performance comparisons with the adaptive EKF (AEKF), conventional LSTM, and MALSTM models show that the proposed MALSTM-EKF achieves faster convergence and higher accuracy. Key evaluation metrics indicate a root mean square error (RMSE) of 0.343 %, a mean absolute error (MAE) of 0.236 %, and a mean absolute percentage error (MAPE) of 1.135 %. Additionally, a statistical analysis of variance (ANOVA) confirms that the performance differences between models are statistically significant, validating the superiority of the MALSTM-EKF. The results highlight its effectiveness for real-world applications, such as electric vehicles (EVs), which ensure optimal energy management and driving range, and renewable energy systems, improving power distribution and grid stability. The findings establish the improved MALSTM-EKF as a robust and reliable SOC estimation method, surpassing existing techniques in precision and adaptability across varying conditions.}, archive = {J_EAAI}, author = {Etse Dablu Bobobee and Shunli Wang and Paul Takyi-Aninakwa and Guangchen Liu and Ebenezer Koukoyi}, doi = {10.1016/j.engappai.2025.111526}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111526}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {State of charge estimation of lithium-ion batteries using improved multi-attention long short-term memory extended kalman filtering model}, volume = {158}, year = {2025}, }
• (2025). Dynamic-aware hypergraph learning for stock recommendation. EAAI, 158, 111525. (www)

  @article{ , abstract = {The stock market is influenced by multiple factors, making stock recommendation a challenging task. In recent years, several studies have improved recommendation performance by modeling higher-order relationships between stocks and industries through hypergraphs. However, the existing models still face numerous challenges: such as relying solely on a single stock feature, failing to account for the dynamic influence of the industry on individual stocks, and overlooking the heterogeneity in stock performance within the same industry. To address these challenges, a Dynamic-Aware Hypergraph Learning (DAHL) is proposed for stock recommendation. Firstly, the model designs a dynamic-aware module that captures the diverse and dynamic relationships between industries and individual stocks through an attention mechanism, emphasizing the heterogeneity of stock performance. Secondly, the model introduces industry average returns adjusted for annual volatility as hyperedge weights. This optimization improves the industry hypergraph structure and effectively reflects the risk level of each industry. Finally, a gating mechanism is integrated into the hypergraph convolution to filter and retain important features, further improving the precision and effectiveness of information propagation. Extensive experiments demonstrate that DAHL outperforms state-of-the-art models, achieving average return ratio of 69 %, 59 % and 63 % on China's A-share, National Association of Securities Dealers Automated Quotations (NASDAQ), and New York Stock Exchange (NYSE), respectively.}, archive = {J_EAAI}, author = {Shaoqi Ma and Jiacheng Han and Chao Luo}, doi = {10.1016/j.engappai.2025.111525}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111525}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic-aware hypergraph learning for stock recommendation}, volume = {158}, year = {2025}, }
• (2025). A brain-inspired projection contrastive learning network for instantaneous learning. EAAI, 158, 111524. (www)

  @article{ , abstract = {The biological brain can learn quickly and efficiently, while the learning of artificial neural networks is astonishing time-consuming and energy-consuming. Biosensory information is quickly projected to the memory areas to be identified or to be signed with a label through biological neural networks. Inspired by the fast learning of biological brains, a projection contrastive learning model is designed for the instantaneous learning of samples. This model is composed of an information projection module for rapid information representation and a contrastive learning module for neural manifold disentanglement. An algorithm instance of projection contrastive learning is designed to process some machinery vibration signals and is tested on several public datasets. The test on a mixed dataset containing 1426 training samples and 14,260 testing samples shows that the running time of our algorithm is approximately 37 s and that the average processing time is approximately 2.31 ms per sample, which is comparable to the processing speed of a human vision system. A prominent feature of this algorithm is that it can track the decision-making process to provide an explanation of outputs in addition to its fast running speed.}, archive = {J_EAAI}, author = {Yanli Yang}, doi = {10.1016/j.engappai.2025.111524}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111524}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A brain-inspired projection contrastive learning network for instantaneous learning}, volume = {158}, year = {2025}, }
• (2025). A cascaded strategy-based hierarchical reinforcement learning algorithm for lane change decision-making. EAAI, 158, 111522. (www)

  @article{ , abstract = {Hierarchical reinforcement learning (HRL) has demonstrated considerable promise in addressing complex driving tasks. However, existing HRL-based autonomous driving decision systems face challenges such as inefficient convergence, lack of interdependence among driving maneuver strategies (including throttle/brake control and steering adjustments), and inadequate risk assessment mechanisms, all of which impede the safety and stability of lane-changing decisions. This study proposes a novel HRL framework for continuous lane-changing decision planning. This framework establishes cascaded relationships between driving maneuvers strategies and integrates a comprehensive risk assessment mechanism to address these challenges. Initially, a hierarchical decision model is developed, where the high-level determines the lane-changing intent, while the low-level manages continuous and precise maneuvers. Subsequently, by integrating a Bayesian network, the cascading between throttle/brake openings and steering angles is achieved, optimizing the system's joint strategy distribution. Furthermore, a comprehensive risk assessment mechanism that evaluates the cooperation level of drivers and the severity of potential collisions is designed to encourage agents to adopt strategies that minimize risk. The effectiveness of the proposed decision-making framework has been validated through comparative experiments in mixed traffic scenarios simulated within the Car Learning to Act (CARLA) environment and corroborated with human driving data from the Next Generation Simulation (NGSIM) database.}, archive = {J_EAAI}, author = {Heng Du and Dingfa Lin and Xiaolong Zhang and Lingtao Wei and Shizhao Zhou and Xuanhao Cheng and Luxin Zhang and Jin Jiang}, doi = {10.1016/j.engappai.2025.111522}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111522}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cascaded strategy-based hierarchical reinforcement learning algorithm for lane change decision-making}, volume = {158}, year = {2025}, }
• (2025). Optimizing deep learning-driven computer vision for civil infrastructure defect identification: Challenges and strategies. EAAI, 158, 111521. (www)

  @article{ , abstract = {With advancements in Internet of Things (IoT) technologies and deep learning, Structural Health Monitoring (SHM) is progressing towards long-distance and intelligent applications. To promote the widespread adoption of deep learning in vision-based SHM, this survey compiles optimization strategies derived from deep learning models specifically designed for defect identification in civil infrastructure—an area that has not been comprehensively explored. First, a concise overview of fundamental deep learning models for vision-based defect identification is provided. Next, optimization methods are categorized into three main groups, each addressing a distinct challenge encountered in practical vision-based SHM: optimizing considering defect-specific characteristics, lightweight design for real-time identification, and enhance robustness under complex environmental conditions. These methods are further classified systematically based on their similar features. This survey offers researchers a deeper understanding of the challenges in vision-based defect identification and assists them in selecting appropriate optimization techniques to address these challenges. Ultimately, it aims to enhance the effective deployment of deep learning models in vision-based SHM by improving accuracy, enabling real-time operation, and facilitating automated defect identification.}, archive = {J_EAAI}, author = {Ronghua Fu and Zhimeng Huang and Maosen Cao and Drahomír Novák and Chunhui Xie and Jinwen Huang}, doi = {10.1016/j.engappai.2025.111521}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111521}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing deep learning-driven computer vision for civil infrastructure defect identification: Challenges and strategies}, volume = {158}, year = {2025}, }
• (2025). Multi-scale feature fusion with knowledge distillation for object detection in aerial imagery. EAAI, 158, 111518. (www)

  @article{ , abstract = {Object detection in aerial images faces problems such as small and dense objects as well as occlusion. Existing methods usually adopt a fusion structure similar to a feature pyramid network (FPN-like), which introduces a small number of parameters but has a low detection accuracy. Therefore, to effectively balance the parameters of the model and the detection accuracy, this paper proposes a multi-scale feature fusion with knowledge distillation for object detection in aerial imagery (MFF-KD), which comprises a multi-scale feature fusion (MFF) network and channel and spatial attention knowledge distillation (CSAKD). First, we design a scale-aware feature fusion module (SAFFNet) in the MFF network, which is capable of reducing the model complexity while preserving more features of small objects. Secondly, we design an efficient multi-scale self-attention module (EMSA) that is integrated into the deep feature extraction process of the MFF network to capture continuous features of occluded objects. Finally, we propose a CSAKD knowledge distillation method, which enhances both foreground and effective background information in the student model, thereby improving the detection accuracy of the MFF network while controlling the growth of model parameters. We conducted extensive experiments on three publicly available aerial image datasets to validate the effectiveness of our method. The experimental results show that our method achieves good detection results on all three aerial image datasets. In addition, the parameter count of our method is 10.5M, and the real-time detection speed is 111 frames per second (FPS).}, archive = {J_EAAI}, author = {Mengyuan Li and Xingzhu Liang and Qicheng Hu and Yu-e Lin and Chenxing Xia}, doi = {10.1016/j.engappai.2025.111518}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111518}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale feature fusion with knowledge distillation for object detection in aerial imagery}, volume = {158}, year = {2025}, }
• (2025). Machine learning-based multiphase pipeline network modeling. EAAI, 158, 111517. (www)

  @article{ , abstract = {Surface network models are fundamental in the process of optimizing oil and gas field development plans. Evaluating pressure drop is key to designing optimal surface networks. Traditionally, pressure drop is estimated using surface network models that rely on correlations. Recent advancements in machine learning (ML) have introduced pressure drop prediction models, but these models are usually trained on limited datasets and do not operate within network solvers, limiting their applicability. A solution is needed that leverages ML within network modeling, while maintaining high accuracy and computational efficiency. This paper introduces a novel ML-based surface network modeling solution. The proposed approach integrates ML models within a network solver and introduces upscaling, that reduces the number of pipeline segments while preserving accuracy. The ML models are trained using synthetic data generated from a commercial simulator, allowing it to cover a wide range of operating conditions. Among the tested ML algorithms, Artificial Neural Networks showed superior performance. The ML-based network solver demonstrated high accuracy, achieving absolute percentage errors below 5 % for pressure drop, below 3 % for inlet pressure, and zero for flow rates. For the processing time, the ML solver using upscaling outperformed the commercial simulator across all tested cases. To the best of our knowledge, pressure drop evaluation using ML has not been applied in a modular approach within network solvers. The efficiency gains provided by the proposed solution enable its effective use in surface facility optimization loops, where network models must be solved numerous times to optimize field development.}, archive = {J_EAAI}, author = {Deniz Alakoum and Kassem Ghorayeb}, doi = {10.1016/j.engappai.2025.111517}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111517}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-based multiphase pipeline network modeling}, volume = {158}, year = {2025}, }
• (2025). Pseudo-fault data enhanced relation network for fault detection and localization in train transmission systems. EAAI, 158, 111515. (www)

  @article{ , abstract = {Fault detection and localization is a critical technology for ensuring the safe and cefficient operation of train transmission systems (TTS). However, due to the frequent scarcity or absence of fault data, achieving precise detection and localization faces many challenges. To address this challenge, this paper proposes a novel pseudo-fault data enhanced relation network (PE-ERN) for fault detection and localization in TTS. The PE-ERN method generates pseudo-fault data by combining fault information from similar external equipment and out-of-distribution data. This strategy enriches the training dataset, enabling model to extract intrinsic heath state information from available health data. Additionally, a feature concatenation mechanism is developed to generate feature pairs by combining health state data with pseudo-fault data. This mechanism uncovers both health-unique and health-inherent attributes, which enhances the PE-ERN's ability to distinguish between fault and health states. Experimental results from subway TTS cases demonstrate that the proposed PE-ERN method outperforms existing fault detection techniques, achieving superior fault localization and detection accuracy across various fault modes, including single-fault, component-level compound-fault, and system-level compound-fault scenarios.}, archive = {J_EAAI}, author = {Zhixu Duan and Ruoxin Liu and Zuoyi Chen and Hong-Zhong Huang}, doi = {10.1016/j.engappai.2025.111515}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111515}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pseudo-fault data enhanced relation network for fault detection and localization in train transmission systems}, volume = {158}, year = {2025}, }
• (2025). Hybrid T-spherical fuzzy decision analytics methodology: Leveraging lance distance induction. EAAI, 158, 111513. (www)

  @article{ , abstract = {T-Spherical Fuzzy (T-SF) set theory provides an advanced approach for modeling uncertain and imprecise information by incorporating degrees of positivity, impartiality, negativity, and refusal. The present research devises new T-SF Lance distances designed to accommodate biased data and quantify dissimilarities in complex decision scenarios. Integrated into the MEthod based on the Removal Effects of Criteria (MEREC) and the Multi-Attribute Ideal-Real Comparative Analysis (MAIRCA), the Lance distance supports the development of a hybrid T-SF MEREC-MAIRCA framework. The proposed methodology enhances objectivity in weight determination and consistency in evaluating alternative solutions under uncertainty. The T-SF MEREC component employs the Lance distance to construct a synthesized index that quantifies performance reductions, while the T-SF MAIRCA component integrates theoretical and actual appraisal outcomes, score functions, and Lance distance-based assessments to effectively prioritize alternatives. A case study on smart farming in Taoyuan City, Taiwan, illustrates the framework's practical value. Seven smart farm operation models were evaluated, with the rooftop urban agriculture model ( s 3 ) identified as optimal, achieving the lowest adverse gap measure ( Q 3 = 0.4395). These findings validate the proposed framework's robustness and suitability for complex real-world decision problems, offering policymakers a reliable, data-driven tool for sustainable urban agriculture planning.}, archive = {J_EAAI}, author = {Ting-Yu Chen}, doi = {10.1016/j.engappai.2025.111513}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111513}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid T-spherical fuzzy decision analytics methodology: Leveraging lance distance induction}, volume = {158}, year = {2025}, }
• (2025). A novel combination of sigmoid and radial basis neural networks for the monkeypox transmission system. EAAI, 158, 111512. (www)

  @article{ , abstract = {Objectives The aim of this research is to present the numerical solutions of the monkeypox transmission system (MTS) by applying a novel stochastic computing framework. The dynamical MTS is categorized into a human and rodent, which is divided into exposed, susceptible, infected, recovered, and clinically diseased human, while the rodent is categorized into infected, susceptible, and exposed. Method A design of novel deep neural network (NDNN) process using the combination of sigmoid and radial basis functions with fourteen and twenty-four numbers of neuron in the hidden layers (HLs) is presented for solving the MTS. The process based on the NDNN works by taking two HLs by using log-sigmoid activation function and fourteen neurons in the first HL, while radial basis activation function and twenty-four neurons in the second HL. An efficient scale conjugate gradient scheme will be used for the optimization to solve both human and rodent classes of the nonlinear MTS. Finding For the correctness of this novel two HLs procedure, the overlapping of the results based on the obtained and literature results is performed. The negligible values of the absolute error around 10 −04 to 10 −05 are also obtained. Furthermore, the reduced values of the mean square error based on the validation, training, and testing validate the authenticity of the designed scheme. Novelty The novel designed NDNN process with two layers, different activation functions and optimization based scale conjugate performances are exploited first time to solve the nonlinear dynamics of the MTS.}, archive = {J_EAAI}, author = {Zulqurnain Sabir and M.A. Abdelkawy}, doi = {10.1016/j.engappai.2025.111512}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111512}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel combination of sigmoid and radial basis neural networks for the monkeypox transmission system}, volume = {158}, year = {2025}, }
• (2025). Integration of hybrid machine learning model with differential evolution algorithm for underground pipeline corrosion prediction. EAAI, 158, 111511. (www)

  @article{ , abstract = {Pipeline corrosion compromises pipelines' structural integrity and safety, necessitating reduction to prevent leaks, environmental damage, and potentially catastrophic failures. Although several field and theoretical strategies have been adopted in the literature to determine pipeline corrosion, some limitations remain, such as providing cost-effective methods, less complex procedures, and reliable predictive models. This research seeks to evaluate the ability of a newly developed machine learning (ML) model based on Extreme Learning Machine (ELM), Support Vector Machine Regressor (SVR), Extreme Gradient Boosting (XGB), and Multivariate Adaptive Regression Spline (MARS) and nature-inspired Differential Evolution (DE) optimization for pipelines corrosion prediction. As a secondary research objective, the proposed ML model was integrated with a feature selection (FS) algorithm to identify the essential parameters of corrosion determination. The research used a dataset from an open-source literature review with 259 observations. There are six pipeline parameters, i.e., redox potential (rp), potential of hydrogen (ph), pipe-to-soil potential (pp), soil resistivity (re), pipeline age (t), and pipeline coating type (ct) and six soil parameters, the soil in which the pipeline is buried, i.e., soil textural class (class), water content (wc), bulk density (bd) and dissolved chloride (cc), bicarbonate (bc), and sulfate (sc) ion concentrations. The predicted results indicated that XGB_FS model performed the best with maximum determination coefficient (R 2 = 0.8407) and minimum root mean square error (RMSE = 0.395). The research evaluated the efficiency of the developed model, the reliability of using such an algorithm in corrosion predictions, and whether it enhances the average performance when combined with other models, such as hybrid ML models.}, archive = {J_EAAI}, author = {Ahmad Fares Al-Nouti and Zaher Mundher Yaseen}, doi = {10.1016/j.engappai.2025.111511}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111511}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integration of hybrid machine learning model with differential evolution algorithm for underground pipeline corrosion prediction}, volume = {158}, year = {2025}, }
• (2025). Low-light underwater image enhancement with retinex-guided mamba network. EAAI, 158, 111509. (www)

  @article{ , abstract = {Low-light underwater image Enhancement remains a challenging problem, even with advanced deep-learning methods, due to serious illumination degeneration. To address this issue, we present a Retinex-guided Mamba network for low-light Underwater Image Enhancement (RM-UIE), featuring dual pathways dedicated to correction reflectance and illumination maps. Concretely, we first design a Multi-scale Retinex Estimator (MRE) to transform the input image into two intermediary spaces, each approximating the desired reflectance and illumination maps. Then, we develop a 8-Direction Mamba Block (8D-MB) to refine reflectance and illumination maps. The core operator in the 8D-MB is a novel eight-direction Mamba scanning mechanism that can more efficiently extract the spatial relationship of images. Finally, extensive qualitative and quantitative experiments on mainstream datasets demonstrate that the proposed method significantly outperforms existing methods in both illumination and detail restoration. Additionally, the RM-UIE outperforms the mainstream methods, achieving a Peak Signal-to-Noise Ratio (PSNR) of 27.4237 and a Structural Similarity Index Measure (SSIM) of 0.8915 on the Underwater Image Enhancement Benchmark (UIEB) dataset, which further validates its effectiveness and generalizability. The code is available at: https://github.com/WYJGR/RMUIE/ .}, archive = {J_EAAI}, author = {Jiaquan Yan and Yijian Wang and Hao Hu and Long Yao}, doi = {10.1016/j.engappai.2025.111509}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111509}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Low-light underwater image enhancement with retinex-guided mamba network}, volume = {158}, year = {2025}, }
• (2025). A machine learning-based interpretable fault diagnosis method for rolling bearings under unbalanced samples. EAAI, 158, 111505. (www)

  @article{ , abstract = {Bearing fault diagnosis methods based on deep learning face the problems of unbalanced datasets and insufficient interpretability of the decision-making process. To address the above problems, an interpretable imbalance fault diagnosis framework is proposed. Firstly, clustering and density-based hybrid method (CDBH) is applied to solve the multi-classification problem of unbalanced datasets, and the bearing data distribution is balanced by reducing normal type samples and increasing fault category samples. Secondly, multi-domain feature sets such as time domain, frequency domain and entropy features of bearing vibration signals are extracted, and these features are incorporated into the model input to enhance the physical interpretability of fault diagnosis. Then, a scaled dot product attention adaptive temporal convolutional network (STCN) based on scaled dot product attention mechanism (SA) and triangular topology aggregation optimizer (TTAO) is designed to improve the diagnosis performance. Finally, a deep learning interpretation technology combining signal preprocessing and gradient-weighted class activation mapping (Grad-CAM) is proposed to visualize the contribution of multi-domain features in the fault diagnosis process, thereby realizing the physical interpretability analysis of the decision process. The proposed fault diagnosis framework is verified using real experiments and public datasets. Experimental results show that the proposed framework has significant performance advantages and interpretability.}, archive = {J_EAAI}, author = {Yingxue Chen and Guanxiang Feng and Huatao Chen and Linfeng Gou and Siji Wang}, doi = {10.1016/j.engappai.2025.111505}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111505}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A machine learning-based interpretable fault diagnosis method for rolling bearings under unbalanced samples}, volume = {158}, year = {2025}, }
• (2025). A lightweight defect detection network for titanium strip via efficient convolution and feature-driven mechanism. EAAI, 158, 111503. (www)

  @article{ , abstract = {This study proposes a lightweight defect detection network for titanium strip via efficient convolution and feature-driven mechanism (ECFD-Net) to balance model complexity and detection performance. A shuffle grouped convolution (SGConv) is proposed to enhance accuracy while reducing computational complexity. By employing a progressive feature-driven mechanism, a feature extraction module is developed to improve the detection of weak features. By refining neck stage and reconstructing the feature extraction module within the backbone, model parameters are further compressed. A performance-efficiency evaluation metric is formulated by integrating nonlinear functions and composite weight allocation strategies, and the overall efficiency of the model is quantified using nonlinear mapping functions. ECFD-Net surpasses advanced object detection models by attaining a mean average precision at an intersection over union threshold of 0.5 ( mAP 0.5) of 0.799 and a recall of 0.782 with only 1.04 million parameters, increasing mAP 0.5 by 8.2 % and reducing parameters by 98.2 % compare to the you only look once version 12x (YOLOv12x).}, archive = {J_EAAI}, author = {Yang Zhu and Yong-Cheng Lin and Xian-Hua Tan and Shu-Xin Li}, doi = {10.1016/j.engappai.2025.111503}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111503}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight defect detection network for titanium strip via efficient convolution and feature-driven mechanism}, volume = {158}, year = {2025}, }
• (2025). Feature selection-based rough graph classification via graph convolutional neural networks. EAAI, 158, 111502. (www)

  @article{ , abstract = {In various applications, graph data is usually omnipresent. Many existing methods integrate graph classification and feature selection to select informative features; however, uncertainty in graph data is easily ignored and then the relationship between nodes has been not described fully. Additionally, when processing graph data, graph convolutional neural networks (GCNs) fall short in illustrating the structural features and inherent semantic of whole graphs, and relying on node features will lead to over-smoothing, which can affect the efficacy of graph classification. To address these flaws, this article presents a novel feature selection-based rough graph classification methodology via GCNs. Firstly, the average value of each node feature values is calculated to be sorted. A variance filtering-based feature selection scheme is constructed to remove irrelevant node features. Then, the upper approximate and lower approximate feature sets will be divided by a variance threshold. Secondly, according to the relationship between the upper and lower approximates, the upper approximate weight and lower approximate weight of nodes are provided, and the upper approximate weight and lower approximate weight of edges between nodes can be provided to remove the influence of redundant information on nodes. The weight coefficient between nodes is learned by calculating the similarity between nodes to create the upper and lower approximation graphs, and then rough graphs will be constructed to fully display the uncertainty between nodes. Thirdly, an updated self-attention pooling layer scheme is presented to evaluate the importance of nodes, GraphConv is used to extract features for an output feature matrix and obtain a sequence of important nodes. Feature matrix and adjacency matrix can be updated, the hierarchical pooling structure is studied to retain important nodes in network training, and then a simplified rough graph can be obtained. Thus, an end-to-end GCN model is designed by simultaneously considering the structure and node features of rough graphs. Finally, a novel rough graph classification algorithm via GCNs is designed. Experiments on 15 graph datasets illustrate the greater effectiveness of the designed methodology relative to other competitive algorithms.}, archive = {J_EAAI}, author = {Lin Sun and Yiwen Cai and Weiping Ding and Jiucheng Xu}, doi = {10.1016/j.engappai.2025.111502}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111502}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature selection-based rough graph classification via graph convolutional neural networks}, volume = {158}, year = {2025}, }
• (2025). An initiative in medical diagnosis for detecting the factors of hypoglycemia disease with a new approach of interval-valued pythagorean fuzzy linear diophantine aczel alsina aggregation operators. EAAI, 158, 111499. (www)

  @article{ , abstract = {Hypoglycemia is a situation when blood sugar levels drop below normal. It is most commonly associated with Diabetes, particularly in individuals who are taking insulin or other medications that increase insulin secretion. However, it can also occur in people without Diabetes due to various factors like fasting, alcohol consumption, or certain medical conditions. Our main goal is to create a path to calculate the ranking of the most risky factor that can cause Hypoglycemia in the human body. By adopting interval-based fuzzy logic and preserving the Pythagorean constraints, interval-valued Pythagorean fuzzy sets (IVPyFS) provide an effective method for modeling and resolving the issues that involve ambiguity, uncertainty, and incomplete information. The IVPyFS allows the experts to describe their opinions independently using the degree of membership (MD) and non-membership (NMD). We use the Aczel-Alsina operations to enhance the flexibility when the information is obtained to detect hypoglycemia. Consequently, we created a new idea of interval-valued Pythagorean fuzzy (IVPyF) linear Diophantine set (IVPyFLDS). We have formed the weighted average and geometric operators using Aczel-Alsina triangular norms. We investigate some fundamental properties of the developed operators. Furthermore, we observe the sensitivity of the results and compare the results for the justification.}, archive = {J_EAAI}, author = {Sarah Asghar and Syed Tauseef Saeed and Zeeshan Ali and Amir Hussain and Abdulrahman A. Almehizia and Salman Saleem}, doi = {10.1016/j.engappai.2025.111499}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111499}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An initiative in medical diagnosis for detecting the factors of hypoglycemia disease with a new approach of interval-valued pythagorean fuzzy linear diophantine aczel alsina aggregation operators}, volume = {158}, year = {2025}, }
• (2025). An integrated p,q-quasirung orthopair fuzzy decision-making approach for strategic selection of competitive intelligence platforms. EAAI, 158, 111498. (www)

  @article{ , abstract = {In an increasingly globalized world, Competitive Intelligence (CI) plays a vital role for export-oriented businesses aiming to maintain a competitive advantage and identify opportunities in international markets. Accurate, timely, and comprehensive information is essential for understanding market dynamics, evaluating competitors, and analysing customer behaviour. However, selecting reliable commercial intelligence websites presents challenges, such as issues of data quality, pricing, usability, and coverage. This study addresses these challenges by introducing a scientific decision-making framework using a fuzzy Multi-Criteria Decision-Making (MCDM) approach to handle the uncertainty in the selection process. The research proposes a novel decision-making model based on p,q-Quasirung Orthopair Fuzzy Numbers (p,q-QOFNs), applying p,q-quasirung operators to calculate expert weights. It integrates the Simple Weight Calculation (SIWEC) and Multi-Attributive Border Approximation Area Comparison (MABAC) methods, called “p,q-quasirung-SIWEC-MABAC”, to determine criteria weights and rank website alternatives. This model enhances the integration of subjective evaluations, improving both robustness and efficiency in decision-making. A case study validates the model's practical application in evaluating CI websites, supported by sensitivity and comparative analyses confirming the model's reliability across diverse scenarios. Findings highlight that data security and reliability are the most critical factors in CI website selection. Among the evaluated platforms, A3 emerges as the top choice due to its detailed insights into textile import trends and supplier analysis. This research contributes a unique methodology to CI literature by enhancing export decision-making processes through advanced fuzzy logic techniques, ultimately helping businesses navigate the complexities of global trade more effectively.}, archive = {J_EAAI}, author = {Sinan Çizmecioğlu and Ahmet Çalık and Erfan Babaee Tirkolaee}, doi = {10.1016/j.engappai.2025.111498}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111498}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated p,q-quasirung orthopair fuzzy decision-making approach for strategic selection of competitive intelligence platforms}, volume = {158}, year = {2025}, }
• (2025). A design concept evaluation model with q-rung orthopair fuzzy rough einstein interaction aggregation operator and combined border approximation area compromise for ideal solution method. EAAI, 158, 111497. (www)

  @article{ , abstract = {Design concept evaluation (DCE) is an important task in the initial stage of product design, directly impacting the subsequent development and production of the new product. The process of DCE involves the expression of uncertain information, the determination of evaluation criteria weight, and the fusion of information, which makes it a challenging task. However, most of the current assessment models have a single expression of uncertainty information, the determination of evaluation criteria weights is limited to subjective or objective information, and the fusion of information is not accurate enough. Therefore, the design concept evaluation model with q- rung orthopair fuzzy rough ( q -ROFR) sets and a combination weight method is proposed in this paper. The model adopts the q -ROFR sets to express the uncertainty information, which has the advantages of parameterization flexibility and compatibility for complex uncertainties. Moreover, a novel entropy that can effectively measure the fuzziness of the q -ROFR sets is introduced to calculate the objective criteria weights, based on which a novel combination weight method is introduced, which solves the subjectivity and the data dependence of the single-weight method to determine the criteria weights more reasonably. Then, the combined border approximation area compromise for ideal solution (CoBACoFISo) method with strong applicability is developed to rank design concept alternatives. Furthermore, relevant algorithms are developed to make the model intelligent. Finally, the superiority of the proposed method is verified by a practical case study, sensitivity analysis, and comparisons. The results show that the proposed model is effective.}, archive = {J_EAAI}, author = {Liming Xiao and Tao Fang and Guangquan Huang and Muhammet Deveci}, doi = {10.1016/j.engappai.2025.111497}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111497}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A design concept evaluation model with q-rung orthopair fuzzy rough einstein interaction aggregation operator and combined border approximation area compromise for ideal solution method}, volume = {158}, year = {2025}, }
• (2025). A culturally-aware benchmark for person re-identification in modest attire. EAAI, 158, 111494. (www)

  @article{ , abstract = {Person Re-Identification (ReID) is a fundamental task in computer vision with critical applications in surveillance and security. Despite progress in recent years, most existing ReID models often struggle to generalize across diverse cultural contexts, particularly in Islamic regions like Iran, where modest clothing styles are prevalent. Existing datasets predominantly feature Western and East Asian fashion, limiting their applicability in these settings. To address this gap, we introduce Iran University of Science and Technology Person Re-Identification (IUST_PersonReId), a dataset designed to reflect the unique challenges of ReID in new cultural environments, emphasizing modest attire and diverse scenarios from Iran, including markets, campuses, and mosques. Experiments on IUST_PersonReId with state-of-the-art models, such as Semantic Controllable Self-supervised Learning (SOLIDER) and Contrastive Language–Image Pretraining Re-Identification (CLIP-ReID), reveal significant performance drops compared to benchmarks like Market1501 and Multi-Scene MultiTime (MSMT17), specifically, SOLIDER shows a drop of 50.75% and 23.01% Mean Average Precision (mAP) compared to Market1501 and MSMT17 respectively, while CLIP-ReID exhibits a drop of 38.09% and 21.74% mAP, highlighting the challenges posed by occlusion and limited distinctive features. Sequence-based evaluations show improvements by leveraging temporal context, emphasizing the dataset’s potential for advancing culturally sensitive and robust ReID systems. IUST_PersonReId offers a critical resource for addressing fairness and bias in ReID research globally.}, archive = {J_EAAI}, author = {Alireza Sedighi Moghaddam and Fatemeh Anvari and Mohammadjavad Mirshekari Haghighi and Mohammadali Fakhari and Mohammad Reza Mohammadi}, doi = {10.1016/j.engappai.2025.111494}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111494}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A culturally-aware benchmark for person re-identification in modest attire}, volume = {158}, year = {2025}, }
• (2025). A semi-supervised dual-path model for underground defect detection. EAAI, 158, 111493. (www)

  @article{ , abstract = {Ground penetrating radar (GPR) is widely used for detecting underground cavities due to its continuous, non-destructive, and high-precision capabilities. However, the interpretation of GPR images largely relies on manual efforts, which are inefficient and subjective. This study proposes a semi-supervised deep learning model based on You Only Look Once version 8 (YOLOv8) for detecting underground targets, particularly voids and cavities. The proposed YOLOv8 model features a dual-path architecture. Specifically, the attention-enhanced path integrates a convolutional block attention module (CBAM) to focus on key features for precision, while the lightweight path utilizes the MobileNet to reduce model parameters for efficiency. A two-stage data augmentation method is proposed. First, the dual-path YOLOv8 model is pre-trained using real radar images and simulated images generated from physics-based numerical simulations. Then, the pre-trained model assigns pseudo-labels to unlabeled data produced by a cycle-consistent generative adversarial network (CycleGAN). These pseudo-labeled data are then incorporated into the training dataset to further enhance the proposed YOLOv8 model. An engineering example from Hangzhou, China, validates the effectiveness of the proposed model. The results show that the proposed YOLOv8 model outperforms alternative models in terms of accuracy and efficiency. The detection accuracy of the proposed model is improved by incorporating numerical simulation data and is further enhanced when pseudo-labeled data are added. In addition, the high-quality dataset established in this study provides valuable resources for future research on underground target detection.}, archive = {J_EAAI}, author = {Shaoxiang Zeng and Gengxin Wang and Honglei Sun and Yuanqin Tao and Xiaodong Pan}, doi = {10.1016/j.engappai.2025.111493}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111493}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A semi-supervised dual-path model for underground defect detection}, volume = {158}, year = {2025}, }
• (2025). Low-light image enhancement network based on central difference convolution. EAAI, 158, 111492. (www)

  @article{ , abstract = {Since the convolutional neural networks and transformers used in existing low-light image enhancement methods were prone to ignore high-frequency information, resulting in blurred details of the enhanced image, this affected the performance of computer vision tasks at night. Therefore, we propose a novel low-light image enhancement network based on central difference convolution (CDCLNet). This method uses traditional image processing methods to help the network extract high-frequency information. Specifically, firstly, in order to fully expose the hidden high-frequency details, the proposed method uses the multi-exposure strategy based on bright and dark masks to expose the image to different levels. Secondly, the complementary information between multi-exposure images is fused through the first-stage network. Finally, the second-stage network suppresses the amplified noise and enhances the details. In addition, We design a central difference convolution module (CDCM) with channel attention to adaptively extract gradient-level detailed features according to the need of the two-stage network. In order to make the network notice illumination non-uniformity, we propose a multi-scale feature attention module (MFAM), which extracts multi-scale features in each channel and generates channel-specific attention maps. Experiments on four public datasets show that the proposed method can enhance the details more effectively than mainstream methods, and achieves the highest structural similarity index on two paired datasets, with an average value of 0.899.}, archive = {J_EAAI}, author = {Yong Chen and Shangming Chen and Huanlin Liu and Hangying Xiong and Yourui Zhang}, doi = {10.1016/j.engappai.2025.111492}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111492}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Low-light image enhancement network based on central difference convolution}, volume = {158}, year = {2025}, }
• (2025). Temporal consistency ensemble empirical mode decomposition for forecasting practical metal price. EAAI, 158, 111490. (www)

  @article{ , abstract = {Accurately forecasting metal prices is critical to economic, industrial, and energy markets. However, traditional time series models often rely on future data, limiting their real-world applicability. This study found that decomposition methods utilizing future data inflate model performance, and the high accuracy of forecasting models is mainly due to these unrealistic assumptions. In this paper, we propose a novel Temporal Consistency Ensemble Empirical Mode Decomposition (TC-EEMD) method designed for rolling scenarios for practical forecasting models. The performance of the method was evaluated using Support Vector Regression (SVR), Long Short-Term Memory (LSTM) networks, and Transformer on real precious metal price data. The results show that TC-EEMD and hybrid approaches improve forecast accuracy and robustness, mitigating noise dependence and stabilizing forecasts across different industrial applications.}, archive = {J_EAAI}, author = {Yujin Choi and Dongbin Kim and Jaewook Lee}, doi = {10.1016/j.engappai.2025.111490}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111490}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Temporal consistency ensemble empirical mode decomposition for forecasting practical metal price}, volume = {158}, year = {2025}, }
• (2025). An ensemble machine learning model for predicting the need for icebreaker assistance in ice-covered waters. EAAI, 158, 111489. (www)

  @article{ , abstract = {Winter navigation presents challenges due to ice conditions, necessitating typical navigation modes: independent navigation and icebreaker assistance. Current navigation mode estimations rely on navigators' expertise, which is subjective and difficult to standardize. Motivated by the complexities of current estimations and the need for icebreaker resource optimization, this study proposes neural oblivious decision ensembles, a deep learning model, to estimate navigation modes based on ship characteristics and operational conditions. Given the inherently imbalanced data, where icebreaker assistance cases are fewer compared to independent navigations, the focal loss function is employed to emphasize the minority class. The results show that the proposed model outperforms benchmarks like random forest and gradient boosting, achieving 97 % accuracy, 95 % precision, 93 % recall, and 94 % F1 score, with up to a 10 % recall and 6 % F1 score improvement. By quantifying prediction probabilities and uncertainties, the model enables informed decision-making, where high-probability, low-uncertainty predictions can reliably guide estimations. The findings demonstrate that the proposed model can generate spatially scalable maps to highlight areas requiring assistance and provide granular estimates along ship routes. Predictions with understandable visual representations can support proactive icebreaker allocation. These insights lay the groundwork for developing an intelligent decision-support system and future resource optimization.}, archive = {J_EAAI}, author = {Cong Liu and Mikko Suominen and Mashrura Musharraf}, doi = {10.1016/j.engappai.2025.111489}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111489}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An ensemble machine learning model for predicting the need for icebreaker assistance in ice-covered waters}, volume = {158}, year = {2025}, }
• (2025). Classification of animal species using efficient neuron attention stage-by-stage network. EAAI, 158, 111488. (www)

  @article{ , abstract = {Classifying animal species is crucial for understanding their evolutionary relationships and characteristics. This article introduces the Efficient Neuron Attention Stage-by-Stage Network (ENAS-Net) for animal species classification. The process involves obtaining animal images from a database, applying anisotropic diffusion for noise reduction, and using O-SegNet for segmentation. Image augmentation techniques like colour augmentation, flipping, and rotation are used to enhance performance. Features are extracted using Entropy-based Opponent Color Local Binary Pattern (E-OCLBP) and Convolutional Neural Network (CNN). The ENAS-Net is a combination of Neural Architecture Search Network (NASNet) and EfficientNet. Moreover, for dataset 1, the ENAS-Net recorded an accuracy of 91.405 %, a True Positive Rate (TPR) of 92.045 %, and a True Negative Rate (TNR) of 91.549 %. For dataset 2, the proposed approach demonstrated strong performance, achieving an accuracy rate of 92.172 %. Additionally, it maintained a TPR of 92.994 %, effectively identifying relevant instances, while sustaining a TNR of 92.452 %, minimizing false detections.}, archive = {J_EAAI}, author = {Lingaraj A. Hadimani and Manjunath R. Hudagi and Sachin Urabinahatti and Sanjeevkumar Angadi and Basavaraj A. Patil}, doi = {10.1016/j.engappai.2025.111488}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111488}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Classification of animal species using efficient neuron attention stage-by-stage network}, volume = {158}, year = {2025}, }
• (2025). Longest & uniform subsequence representation: Time interval dependent data augmentation for sequential recommendation. EAAI, 158, 111487. (www)

  @article{ , abstract = {Sequential Recommendation (SR) aims to effectively infer users’ interests or behaviors in future, widely using in e-commerce and social media fields. Existing studies show different time intervals in a sequence may lead to preference drift, and especially that uniform time intervals in a sequence can improve the prediction performance compared with non-uniform has been proved. However, this kind of methods only focus on global time interval changes usually, but often neglect the role of local time intervals, which may lead to not capturing short-term user interest fluctuations. Thus, we propose the longest uniform subsequence insertion (LUI) in single-user (intra-sequence), and the longest uniform subsequence substitution (LUS) of multi-user (inter-sequence) methods, which consider not only local but global sequence-related view information. Additionally, considering the sensitivity of sequence length, we use different combinations of augmentations to address the problem of skewed sequence length distribution in the dataset. Next, Fourier transform is employed to integrate low-frequency with high-frequency information by injecting fine-grained sequential patterns to inductive biases. Self-supervised learning method is also used to alleviate the problems of data sparsity and noise. At last, the experimental results on three benchmark datasets show that the proposed LURec model outperforms several typical recommendation models.}, archive = {J_EAAI}, author = {Xuegang Wu and Ting Wang}, doi = {10.1016/j.engappai.2025.111487}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111487}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Longest & uniform subsequence representation: Time interval dependent data augmentation for sequential recommendation}, volume = {158}, year = {2025}, }
• (2025). Dynamic visual attention-based neuron awakening and shifting in deep reinforcement learning. EAAI, 158, 111486. (www)

  @article{ , abstract = {Deep reinforcement learning (DRL) has made significant strides in efficient decision-making for complex tasks, yet exploration efficiency remains a primary limitation that affects overall performance. Effective exploration is essential in dynamic environments common to both simulated and real-world scenarios. Inspired by human use of visual information, visual attention mechanisms can help filter irrelevant details in image-based tasks. However, their application in DRL has not achieved expected results. This paper analyzes the limitations of visual attention in DRL and identifies key factors that contribute to performance decline. We propose a new method, “Focus on Important Content” (FIC), to overcome these challenges. First, we address early zero activation of the dynamic visual attention map by resetting the weights of layers reaching a dormancy threshold. Second, we introduce context-aware information to reduce dormant neurons, improving learning efficiency and performance. Our method is tested in the Atari 100K environment, an abstraction of various dynamic scenarios, achieving an Inter Quartile Mean (IQM) of 1.21. Notably, with context-aware mechanisms, FIC reduces dormant neurons by an average of 6% after 100K iterations, showing a significant improvement in exploration efficiency. These findings demonstrate that FIC not only enhances DRL performance but also offers a scalable solution for integrating visual attention mechanisms into DRL, with potential applications in dynamic real-world tasks such as automated inspection, robotic navigation, and object tracking.}, archive = {J_EAAI}, author = {Jialin Ma and Ce Li and Zhiqiang Feng and Limei Xiao}, doi = {10.1016/j.engappai.2025.111486}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111486}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic visual attention-based neuron awakening and shifting in deep reinforcement learning}, volume = {158}, year = {2025}, }
• (2025). Improved active disturbance rejection control for electro-hydrostatic actuators via actor-critic reinforcement learning. EAAI, 158, 111485. (www)

  @article{ , abstract = {An improved active disturbance rejection control method based on Actor-Critic reinforcement learning (A-C ADRC) is proposed for disturbance compensation and parameter tuning of electro-hydrostatic actuators (EHA). This artificial intelligence-enhanced approach specifically addresses the adverse effects of system parameter uncertainties and time-varying disturbances encountered in EHA operation. Structurally, the cost function and value function are approximated by the radial basis function (RBF) neural network. The critic neural network uses the Lagrangian performance index to construct the value function, evaluate the control effect and feedback to the input end of the actor neural network. The actor neural network uses gradient descent to adaptively update the weights so that realizes the mapping from the system state to the control parameters. The stability and convergence of the closed-loop system are proved by the Lyapunov function. The results show that the proposed control method has made the error of the system converge to the neighborhood of zero without prior knowledge, and has parameter self-tuning ability. Under different target signals, the control parameters respond quickly and converge. Under load disturbance or white noise disturbance, the weights of actor and critic neural networks’ output are always bounded, and the control performance remains unchanged under similar disturbances of different sizes. The control method has good anti-disturbance performance.}, archive = {J_EAAI}, author = {Yingrui Li and Yixuan Jiang and Jiayu Lu and Cao Tan}, doi = {10.1016/j.engappai.2025.111485}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111485}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improved active disturbance rejection control for electro-hydrostatic actuators via actor-critic reinforcement learning}, volume = {158}, year = {2025}, }
• (2025). Enhancing autonomous vehicle control in complex scenarios with deep transformer reinforcement learning. EAAI, 158, 111483. (www)

  @article{ , abstract = {Autonomous driving (AD) aims to significantly enhance road safety and driving comfort, contributing to a substantial reduction in traffic-related fatalities. With the rapid advancement of Artificial Intelligence (AI), particularly Deep Reinforcement Learning (DRL), the control of Autonomous Vehicles (AVs) has seen notable progress. While traditional approaches often rely on Convolutional Neural Networks (CNNs) within Deep Q Networks (DQNs) to process AV state information, they typically struggle with capturing long-range dependencies and often overlook the dynamic presence of diverse road users, such as pedestrians, cyclists, and motorcyclists. In this study, we propose a novel DRL model based on Deep Transformer Q Networks (DTQNs) to address these limitations. Our model integrates Transformer-based architectures to better model temporal and spatial dependencies in complex driving environments. Moreover, it explicitly accounts for the behavior of surrounding road participants to ensure safer and more efficient navigation. We evaluate our approach using the CAR Learning to Act (CARLA) simulator in scenarios involving heavy traffic and mixed road users. Experimental results demonstrate that our DTQN model significantly outperforms traditional DQN methods by reducing both travel delays and collision rates, thereby highlighting its potential for real-world autonomous driving applications.}, archive = {J_EAAI}, author = {Badr Ben ELallid and Nabil Benamar and Miloud Bagaa and Nabil Mrani}, doi = {10.1016/j.engappai.2025.111483}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111483}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing autonomous vehicle control in complex scenarios with deep transformer reinforcement learning}, volume = {158}, year = {2025}, }
• (2025). Long short-term memory neural network- decline curve analysis production forecast method for horizontal wells in tight reservoir based on sequence decomposition and reconstruction. EAAI, 158, 111482. (www)

  @article{ , abstract = {Oil production is a key parameter for evaluating geological potential. Conventional methods struggle to forecast nonlinear and non-stationary production sequences with a strong temporal trend due to the poor physical properties of tight reservoirs. This research proposes a forecast method that integrates a long short-term memory (LSTM) neural network with decline curve analysis (DCA). First, empirical mode decomposition (EMD) is applied to the production sequence, extracting multiple fluctuating intrinsic mode functions (IMF) related to manual construction and a residual (RES) representing reservoir energy depletion. Approximate entropy (ApEn) is then used to categorize each IMF into three groups based on sequence decomposition results, facilitating the reconstruction of the production sequence. LSTM forecasts the recombined IMF sequence, while DCA predicts the residual component. Results indicate that EMD effectively separates time trends, and both reconstructed components and residuals can be accurately predicted. Compared with stand-alone LSTM, back-propagation (BP) neural network, random-forest (RF) and convolutional-neural-network (CNN) models, the proposed method reduces production forecast errors by at least 25 %. This research incorporates signal-processing techniques and physical constraints into production forecasting. These enhancements provide a more accurate and reliable method for forecasting production in hydraulically fractured horizontal wells within tight-oil reservoirs.}, archive = {J_EAAI}, author = {Jinxin Cao and Yiqiang Li and Yaqian Zhang and Xuechen Tang and Qihang Li and Yuling Zhang and Zheyu Liu}, doi = {10.1016/j.engappai.2025.111482}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111482}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Long short-term memory neural network- decline curve analysis production forecast method for horizontal wells in tight reservoir based on sequence decomposition and reconstruction}, volume = {158}, year = {2025}, }
• (2025). Enhancing identification accuracy: Novel fractional activation function as a versatile tool for improved model performance. EAAI, 158, 111481. (www)

  @article{ , abstract = {This study introduces and evaluates the novel Riemann–Liouville (RL) conformable fractional derivative-based Adaptable-Shifted-Fractional-Rectified-Linear-Unit ( R L ASFReLU ) activation function within multi-layer perceptron models, showcasing its superior performance across three distinct nonlinear benchmark systems. R L ASFReLU enhances identification accuracy by integrating a robust bias shift via the input standard deviation and extending the fractional order to α ∈ [ 0 , 2 ] , offering greater flexibility than existing fractional ReLU variants. It demonstrates faster convergence and improved ability to capture complex nonlinear patterns. The proposed function is adaptable across multiple machine learning domains, providing a versatile tool for advancing neural network design.}, archive = {J_EAAI}, author = {Meshach Kumar and Utkal Mehta and Giansalvo Cirrincione}, doi = {10.1016/j.engappai.2025.111481}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111481}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing identification accuracy: Novel fractional activation function as a versatile tool for improved model performance}, volume = {158}, year = {2025}, }
• (2025). Coordinated deep policy learning for frequency-constrained energy management for second-to-second energy balancing of microgrids. EAAI, 158, 111479. (www)

  @article{ , abstract = {Hybrid microgrids (HMGs) are susceptible to frequency distractions due to their low-inertia and randomness of renewable resources. The traditional energy management systems (EMSs) and linear controllers have difficulty dealing with the uncertainty prevalent within the system due to nonlinearity and time-based conditions. This necessitates considering short-term imbalances in HMGs while the intermittency of renewable resources can highly affect the second-to-second time frame of the power system. To address this issue, this work proposes two stage frameworks for an HMG with second-to-second power imbalances: i ) an efficient energy management system is developed to reduce costs and to improve reliability of microgrids. The proximal policy optimization (PPO) with actor and critic neural networks is utilized to solve EMS problem, ii ) a secondary controller based on the non-linear backstepping controller (NBC) is developed to mitigate the dynamic fluctuations of frequency deviation. In this application, the IEEE 39-bus is considered as the benchmark system to study second-to-second power imbalances in the HMGs. The risk of bottlenecks for the test-system with various risk indices is calculated. Transient simulations of the HMG reveal the improvement of operation of the power system from security and stability point of view. The comparison analysis with the prevalent scheme demonstrates the suggested NBC scheme can provide a higher level of stability than prevalent state-of-the-art controllers.}, archive = {J_EAAI}, author = {Kiavash Parhizkar and Borzou Yousefi and Mohammad Rezvani and Abdolreza Noori Shirazi}, doi = {10.1016/j.engappai.2025.111479}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111479}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Coordinated deep policy learning for frequency-constrained energy management for second-to-second energy balancing of microgrids}, volume = {158}, year = {2025}, }
• (2025). Generative artificial intelligence-based modified abstractive cross attention enabled sequence to sequence model for abstractive hindi text summarization. EAAI, 158, 111478. (www)

  @article{ , abstract = {Abstractive Text summarization is the process of providing a concise as well as cohesive summary, which encapsulates the vital information from the original text. Although the supervised models now in use are competent, they frequently rely on annotated datasets and create challenges regarding uninterpretability and limited generalization ability. To overcome the limitations, this research proposes a Generative Artificial Intelligence Sequence to sequence the Bidirectional Encoder Representations from the Transformers model to generate concise summaries. During the generation of the output sequence, the model takes into account data from various segments of the input sequence by utilizing a modified abstractive cross-attention technique. Specifically, the Generative Artificial Intelligence assists in removing grammatical mistakes in summaries via the application of the Global Surrogate method, which ensures the clarity and fluency of the output summary. In addition, the encoder-decoder model enables the accurate summary generation process, vastly improving fluency and accuracy. Furthermore, the experimental outcomes show that the Generative Artificial Intelligence Sequence to sequence the Bidirectional Encoder Representations from the Transformers model surpasses the conventional text summarization techniques concerning Bilingual Evaluation Understudy of 0.71 and Metric for Evaluation of Translation with Explicit Ordering of 0.73, which shows that the proposed model generates a meaningful summary from the given text.}, archive = {J_EAAI}, author = {Babita Verma and Ani Thomas and Rohit Kumar Verma}, doi = {10.1016/j.engappai.2025.111478}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111478}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generative artificial intelligence-based modified abstractive cross attention enabled sequence to sequence model for abstractive hindi text summarization}, volume = {158}, year = {2025}, }
• (2025). Spatial-temporal modeling for multi-scale wind speed predictions in rail transit networks: A graph neural network-based methodology. EAAI, 158, 111477. (www)

  @article{ , abstract = {The operation safety and efficiency of rail transit network are vulnerable to strong winds. An essential requirement exists for employing field monitoring data to estimate wind speed trends and make subsequent emergency decisions in time. However, the current warning systems primarily focus on wind speeds at individual monitoring spots or single lines, while the spatial dependency of wind patterns within a larger space context is ignored. To bridge this research gap, this study proposes a pioneering network-scale wind speed prediction approach based on spatial-temporal graph neural networks. First, a simulated wind stress vector field is devised, leveraging time-lagged cross correlation, to enhance the input wind sensors graph. This novel approach improves upon the traditional predefined graph structure, which relies solely on geographical distances. Second, diffusion convolution recurrent neural network is applied to model the spatial-temporal wind diffusion patterns and predict future wind speed. By carrying out comparison analysis on two real-world railroad datasets with different spatial and temporal granularity, the proposed model demonstrates its applicability in multi-scale wind predictions, significantly outperforming widely used baselines, particularly in multi-step ahead predictions. In addition, ablation study is conducted to assess and interpret the contributions of each component within the model. The proposed method in this study, along with obtained prediction results, is expected to support the network dispatching and delay assessment during extreme weather conditions, thereby enhancing the effectiveness of wind warning systems and strengthening rail safety management.}, archive = {J_EAAI}, author = {Yujie Huang and Zhipeng Zhang and Wenqiang Zhao and Hao Hu}, doi = {10.1016/j.engappai.2025.111477}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111477}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatial-temporal modeling for multi-scale wind speed predictions in rail transit networks: A graph neural network-based methodology}, volume = {158}, year = {2025}, }
• (2025). Exploiting long-term markovian feature importance via dual attention for partially-connected differential architecture search. EAAI, 158, 111476. (www)

  @article{ , abstract = {—Differentiable architecture search (DARTS) is implemented as a gradient-based search method for neural architecture generation. However, DARTS suffers from unbalanced competition between unweighted and weighted operations in the search phase of the supernetwork, resulting in a collapse of the search architecture. In this paper, exploiting long-term markovian feature importance via dual attention for partially-connected differential architecture search (MA-DARTS) is proposed, to overcome the excessive accumulation of unweighted operation dominance by reducing redundant features in the supernetwork. First, spatial location attention factors for different semantic groups are learned through spatial attention. The grouped attention approach contributes to capture changes in the spatial semantic importance of search features. Secondly, the channel feature importance is obtained by learning channel attention weights without dimensionality reduction through a one-dimensional convolution factor. Finally, a Markov chain-based long-term importance feature channel selection strategy is designed. This strategy dynamically transmits key features to the search space, which improves the utilization of effective feature information in search. The experimental results demonstrate that MA-DARTS effectively suppresses the problem of excessive proportion of unweighted operations during the search process, achieving better network performance while ensuring the stability of the architecture search. Meanwhile, the proposed method achieves 0.43 %, 0.68 % and 2.2 % accuracy improvement compared to DARTS on Canadian institute for advanced research CIFAR-10, CIFAR-100 and ImageNet datasets.}, archive = {J_EAAI}, author = {Honggui Han and Qiyu Zhang and Fangyu Li and Yongping Du}, doi = {10.1016/j.engappai.2025.111476}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111476}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploiting long-term markovian feature importance via dual attention for partially-connected differential architecture search}, volume = {158}, year = {2025}, }
• (2025). Explainable attention-based fuzzy residual convolution network for solar cell defect identification and classification. EAAI, 158, 111475. (www)

  @article{ , abstract = {The demand for renewable energy is increasing globally, which has led to significant progress in solar cell technology. Solar cell plates convert sunlight into electrical energy used in several utilities. Various faults, including cracks, delamination, and cell breakdowns, can significantly affect the performance of a photovoltaic module. Conventional manual inspection techniques are costly and prone to human mistakes, rendering them unfeasible on a broad scale. Advancements in AI (artificial intelligence) techniques, such as ML (machine learning) and DL (deep learning), have shown effectiveness in automatic solar cell defect detection. DL automatically extracts spatial features and improves solar cell defect detection. However, ML-based methods need hand-crafted solar cell analysis features. Indeed, DL performance correlates to the quality of the annotated dataset in which the expert is unsure about classifying damaged and good solar panels. To approach such a problem, we have designed an attention-based fuzzy residual convolution network, which generates crisp and fuzzy values to enhance solar cell defect identification. In the proposed model, five convolution layers extract high-dimensional spatial features. In addition, we designed a residual attention module, which focuses on the relevant region of the solar cell image. The improved feature map is passed to the fuzzy inference module to capture more fine-grained spatial features using the Gaussian fuzzy membership rule. Furthermore, a gradient-weighted module is implemented on the model to make it more explainable in defect analysis. We assessed the effectiveness of our model on three datasets and compared the results with six other methods, which achieved significant improvement according to the benchmarks. —Source code of the method is available via GitHub URL: https://github.com/dhirujis02/Solarcell .}, archive = {J_EAAI}, author = {Dhirendra Prasad Yadav and Bhisham Sharma and Gustavo Olague}, doi = {10.1016/j.engappai.2025.111475}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111475}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable attention-based fuzzy residual convolution network for solar cell defect identification and classification}, volume = {158}, year = {2025}, }
• (2025). ATCN-BiGRU: A hybrid deep learning framework based temporal convolutional network and stacked bidirectional gate recurrent units for traffic flow prediction in urban scenarios. EAAI, 158, 111473. (www)

  @article{ , abstract = {Traffic flow forecasting research plays an important role in promoting the development of the intelligent traffic system (ITS). However, the existing methods still deal with the road network statically, ignoring the spatiotemporal coupling relationship between traffic flow data, and lacking a unified prediction framework to integrate the advantages of different models. In this research, a deep hybrid learning framework for urban scenarios is proposed, which is coupled by the temporal convolutional network (TCN), stacked bidirectional gate recurrent units (Bi-GRU), and the self-attention mechanism. The proposed framework can fully learn the potential relationship between nodes in the transportation network at different time steps. We validate the framework using multiple datasets, demonstrating that the ATCN-BiGRU framework maintains fast computational efficiency. Real-world online testing experiments further show that the ATCN-BiGRU model achieves significant improvements in prediction accuracy across various evaluation criteria.}, archive = {J_EAAI}, author = {Liyue Fu and Tong Wang and Min Ouyang and Ling Zhao and Xiaofeng Yin}, doi = {10.1016/j.engappai.2025.111473}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111473}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ATCN-BiGRU: A hybrid deep learning framework based temporal convolutional network and stacked bidirectional gate recurrent units for traffic flow prediction in urban scenarios}, volume = {158}, year = {2025}, }
• (2025). Optimized hyperparameters for retail sales forecasting using grid search. EAAI, 158, 111472. (www)

  @article{ , abstract = {The growth in the retail industry is driven by technological advancement, further accelerated by the pandemic and evolving customer behavior. Due to this, retail sales are exhibiting abnormal fluctuations; therefore, accurate sales forecasting for various products is necessary. This paper focuses on applying artificial intelligence tools and their improvement through an optimized hyperparameter grid search methodology. The results show that Gradient Boosting Regression outperforms Recurrent Neural Networks, Long-Short Term Memory, and Gated Recurrent Unit models with hyperparameter learning rates of 0.2 and 0.3, and estimators of 2000 and 2500, for products categories including technology, shoes, food and beverages, clothing cosmetics, toys, and souvenirs. However, for books, the recurrent neural network outperforms the others. This study emphasizes the critical role of hyperparameter tuning in achieving optimal model performance. Improving retail time series sales forecasting, comparing various sales forecasting techniques, and projecting the contributions of various products are among the potential theoretical contributions of this work. Our research offers retail stakeholders with valuable insights that support the selection of the most suitable forecasting models based on specific retailer needs and product offerings.}, archive = {J_EAAI}, author = {Manjunath S. Vhatkar and Pramod Sanjay Mahajan and Rakesh D. Raut and Naoufel Cheikhrouhou and Sudishna Ghoshal}, doi = {10.1016/j.engappai.2025.111472}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111472}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimized hyperparameters for retail sales forecasting using grid search}, volume = {158}, year = {2025}, }
• (2025). Federated learning in network traffic classification: Taxonomy of implementation, application, and impact on sixth-generation wireless networks. EAAI, 158, 111471. (www)

  @article{ , abstract = {Objectives Network traffic classification (NTC) is crucial for network management. However, the surge in Internet traffic poses scalability and privacy issues. Researchers are turning to federated learning (FL) to tackle these issues. Despite the importance of FL, there is lack of literature that comprehensively reviews its implementation and application within NTC and its impact on Sixth-Generation Wireless (6G) networks. Current surveys cover the technical aspects of NTC without fully addressing the integration of FL and its broader implications. This study aims to review the technical implementation of FL in NTC and its application to various network-related areas, including 6G. Methods This study presents a taxonomy for FL implementation in NTC, considering aspects such as learning and aggregation approaches, topology, and client operations. It examines the limitations of these elements and their effects on performance, efficiency, scalability, and their impact on 6G. This study outlines a taxonomy for FL applications, focusing on privacy preservation, scalable classification, and shared security intelligence. Novelty The proposed taxonomy provides insights into research landscape and highlights its limitations. The analysis of the impact of FL-based NTC on 6G provides insight into its integration and implementation challenges. This study discusses open issues and advocates for future research directions in FL for NTC, including 6G. Findings The study identifies areas needing improvement such as privacy, addressing security, single-point-of-failure, hardware limitations, delays and heterogeneity concerns. The findings of this paper show that an optimal implementation approach is essential to cater for heterogeneity and real-time requirements of the network environment.}, archive = {J_EAAI}, author = {Azizi Ariffin and Firdaus Afifi and Faiz Zaki and Hazim Hanif and Nor Badrul Anuar}, doi = {10.1016/j.engappai.2025.111471}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111471}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Federated learning in network traffic classification: Taxonomy of implementation, application, and impact on sixth-generation wireless networks}, volume = {158}, year = {2025}, }
• (2025). Knowledge and data dual-driven cyber-physics system for intelligent monitoring and compensation of machine tool dynamic coupling error. EAAI, 158, 111470. (www)

  @article{ , abstract = {The interpretability of data-driven cyber-physics system (CPS) is limited, while traditional model-driven methods struggle with the time-varying requirements of dynamic performance. CPS combined with the artificial intelligence (AI) is used to monitor and compensate the states of machine tools, which is becoming increasingly important for machining under dynamic working conditions. This paper proposes a knowledge and data dual-driven cyber-physics system to control the dynamic coupling error of the machine tool and improve the contour accuracy. The proposed hybrid deep learning model is combined with the temporal attention mechanism to extract dynamic information of important features from historical data. Theoretical models are constructed to study the coupling effect among thrust harmonics, non-uniform rational B-Splines interpolation errors and kinematic constraints, and they are embedded as the prediction boundary of the hybrid deep learning to improve the interpretability of the model. The data interaction between the cyber module and physical module creates a closed loop to realize the deep integration. Multiple case studies have verified the feasibility of convolutional neural network and gate recurrent unit in terms of compensation accuracy and their robustness under complex working conditions. The minimum root mean square error predicted on the self-constructed dataset is 7.3649 × 10 −5 mm. For machining under different position-velocity-time interpolation periods, the minimum average contour error is 5.6424 × 10 −4 mm, which is 44.19 % lower than other traditional methods. This study proposes a unified framework for the deep integration of CPS and AI algorithms, providing significant insights into the contour accuracy compensation of ultra-precision machining.}, archive = {J_EAAI}, author = {Chengyi Wu and Shijun Ji and Ji Zhao}, doi = {10.1016/j.engappai.2025.111470}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111470}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Knowledge and data dual-driven cyber-physics system for intelligent monitoring and compensation of machine tool dynamic coupling error}, volume = {158}, year = {2025}, }
• (2025). Deep reinforcement learning for optimization of spiral shaft design in shield machine. EAAI, 158, 111469. (www)

  @article{ , abstract = {The increasing demand for highly customized shield machines imposes greater efficiency and accuracy requirements on the optimization. To address the inefficiencies and susceptibility to local optima in existing structural design methods, a novel approach to construct a simulation analysis surrogate model for creating an environment for the optimization of spiral shaft is introduced. It also improves the Deep Deterministic Policy Gradient (DDPG) algorithm to optimize the structural parameters. In the surrogate model, a Goal-oriented autoencoder (GOAE)-classifier model is employed to discriminate the feasibility of samples, and a hybrid surrogate model, utilizing a Self-attention Artificial Neural Network (Self-attention ANN) weight allocation mechanism, makes precise predictions for feasible samples. This model automatically assigns adaptive weights to sub-surrogate models. Within the DDPG framework, a novel serial-parallel hybrid structure for the Actor network is proposed, harnessing the specialized feature representation capabilities of multiple networks to enhance optimization policy accuracy. Additionally, an experience replay filtering mechanism based on sample similarity is introduced to ensure sample diversity and boost the performance of the optimization policy. A simulation analysis surrogate model is constructed on a generated dataset, and the improved DDPG algorithm is leveraged for the optimization of spiral shafts based on this surrogate model. Experimental results demonstrate that the constructed simulation analysis surrogate model facilitates rapid and precise analysis of spiral shaft, while the improved DDPG algorithm successfully optimizes spiral shaft structural parameters, ultimately improving spiral shaft performance.}, archive = {J_EAAI}, author = {Lin Lin and Lingyu Yue and Song Fu and Dan Liu and Yancheng Lv and Yikun Liu and Sihao Zhang and Shiwei Suo}, doi = {10.1016/j.engappai.2025.111469}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111469}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep reinforcement learning for optimization of spiral shaft design in shield machine}, volume = {158}, year = {2025}, }
• (2025). Shared mobility demand prediction via a fast spatiotemporal tensor autoregression. EAAI, 158, 111467. (www)

  @article{ , abstract = {Shared mobility is critical to urban transportation, yet its complex spatiotemporal dynamics challenge traditional prediction methods. We propose the Tucker Decomposition-based Spatiotemporal Tensor Autoregressive Model (T-STAR), which leverages tensor-structured data modeling and Tucker decomposition to efficiently capture multi-dimensional dependencies. Unlike conventional methods, T-STAR preserves high-dimensional structures by decomposing raw spatiotemporal data into a low-rank core tensor and mode-specific factor matrices, reducing complexity and enhancing interpretability by decoupling spatial, temporal, and modal interactions. Experimental results on three benchmark datasets demonstrate T-STAR's strong performance. On the Beijing Taxi Trajectory Dataset (TaxiBJ), T-STAR achieves Mean Absolute Error (MAE) of 23.53 and Root Mean Square Error (RMSE) of 37.71, improving performance by 18.5 % and 21.2 % over baseline averages. On the New York City Taxi Dataset (NYCtaxi), it records MAE of 18.18 and RMSE of 46.87, reducing errors by 22.7 % and 15.4 %. In the sparse-demand New York City Bike-Sharing Dataset (NYCbike), it maintains robust accuracy with MAE of 7.95 and RMSE of 14.32, outperforming baselines by 14.1 % and 17.9 %, respectively. Most notably, T-STAR achieves these results at high speed: on TaxiBJ, it completes a prediction in just 0.35 seconds–87 % faster than the Adaptive Graph Convolutional Recurrent Network (AGCRN) and 99.8 % faster than the Diffusion Convolutional Recurrent Neural Network (DCRNN). By retaining over 95 % of key spatiotemporal correlations through Tucker compression, T-STAR reduces prediction error by 20–30 % while delivering real-time performance, offering a scalable framework for urban traffic prediction and shared vehicle scheduling. Code and data are both available at yanhongyu0/TSTAR (github.com)}, archive = {J_EAAI}, author = {Hongyu Yan and Zhiqiang Lv and Jianbo Li and Benjia Chu and Zhihao Xu}, doi = {10.1016/j.engappai.2025.111467}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111467}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Shared mobility demand prediction via a fast spatiotemporal tensor autoregression}, volume = {158}, year = {2025}, }
• (2025). A novel aczel-alsina aggregation operators based multi-criteria group decision making approach under hesitant fuzzy set for the selection of best brand in educational institution. EAAI, 158, 111464. (www)

  @article{ , abstract = {Selecting the best brand for an educational institution requires careful evaluation of multiple criteria, such as battery life, processing speed, user friendly, and affordability. In a group decision-making environment, experts including faculty members, information technology (IT) administrators, and students collaborate to assess different brands based on these critical factors. A systematic method for evaluating personal views and efficiently ranking alternatives is to use hesitant fuzzy (HF) based aggregation approaches. Multi-criteria group decision-making (MCGDM) takes into consideration the interconnection between criteria by employing existing aggregation operators under hesitant fuzzy sets (HFSs) information. In this article, by combining the power average (PA) operator and t-norm, t-conorm operations of Aczel-Alsina with HFSs information, two aggregation operators have been developed: the hesitant fuzzy power Aczel-Alsina weighted averaging aggregation (HFPAAWAA) operator and the hesitant fuzzy power Aczel-Alsina weighted geometric aggregation (HFPAAWGA) operator. Various useful properties of those operators have been discussed. In addition, a thorough assessment has been considered by combining proximity measures and power-weighted correlation coefficients, resulting in an informed decision. The brand of laptop that is selected should ultimately fit the institution’s technological requirements, financial limitations, and long-term academic activity usability. Lastly, a sensitivity analysis of the parameters used is given and a comparative analysis with the existing aggregation operators have been discussed to demonstrate the suggested strategy.}, archive = {J_EAAI}, author = {Chittaranjan Shit and Ganesh Ghorai}, doi = {10.1016/j.engappai.2025.111464}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111464}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel aczel-alsina aggregation operators based multi-criteria group decision making approach under hesitant fuzzy set for the selection of best brand in educational institution}, volume = {158}, year = {2025}, }
• (2025). Bi-spectral concealed object detection with attention-based fusion of passive thermal infrared and terahertz imaging. EAAI, 158, 111462. (www)

  @article{ , abstract = {Thermal infrared (TIR) and terahertz (THz) spectra of electromagnetic radiation offer promising capabilities for non-intrusive detection of hidden items, providing insights into material properties and temperature distribution of objects. However, individually each modality shows limitations in resolution, penetration depth, and spectral characteristics. To address these limitations, we propose a novel approach that combines the complementary strengths of TIR and THz waves for enhanced detection performance. Specifically, we employ an attention-based fusion network to integrate TIR and THz imagery. The proposed algorithm uses a convolutional neural network (CNN) backbone to extract features from both spectra and involves an attention-based fusion mechanism to learn a set of weights that determine the importance of each feature representation for object detection. Our experimental results demonstrate that the proposed fusion network achieves state-of-the-art performance in detecting hidden items, outperforming conventional single-modality approaches. The combination of TIR and THz waves, coupled with the attention-based fusion network, provides a robust and versatile platform for non-intrusive detection of hidden items in a variety of scenarios. The method achieves state-of-the-art performance with an average accuracy (ACC) of 92.34 % and a detection rate (DR) of 91.12 % at zero false detection rate (FDR). This is the first known work to show combined capabilities of passive THz and TIR imaging for more adequate detection and classification of concealed objects.}, archive = {J_EAAI}, author = {Marcin Kowalski and Krzysztof Mierzejewski and Tomasz Pałys}, doi = {10.1016/j.engappai.2025.111462}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111462}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bi-spectral concealed object detection with attention-based fusion of passive thermal infrared and terahertz imaging}, volume = {158}, year = {2025}, }
• (2025). An artificial neural network implementation on novel tetra hybrid nanofluid model: Case study of solar wind turbines. EAAI, 158, 111459. (www)

  @article{ , abstract = {This paper introduces a novel approach for examining heat transfer in photovoltaic hybrid wind turbines through the application of neural networks and Casson-Sutterby fluid dynamics. The research indicates that the integration of agglomerative tetra hybrid nanoparticles can markedly enhance the efficacy of solar wind turbines. Tetra hybrid nanofluid, characterized by its elevated thermal conductivity, specific heat capacity, and radiative heat transfer, serves as a more efficient heat source in solar turbines. This results in heightened heat production, radiation, and thermal radiation, hence augmenting thermal energy transfer. The proposed neural network approach demonstrates the highest efficacy, as the Absolute Error (AE) falls within the range 10 − 03 t o 10 − 08 , the Mean Square Error (MSE) values lie in the range of 10 − 09 t o 10 − 10 , all the Cases have performance values fall within the range 10 − 04 t o 10 − 05 . The Error histograms values for 20 bins with zero errors are 10 − 05 t o 10 − 06 . The regression approaches to 1, showing the best fit. Tetra hybrid nanofluids exhibit superior heat transmission characteristics, and neural network provides excellent results, thereby providing valuable insights into the optimization of thermal management in solar energy applications and the enhancement of the efficiency of solar-powered vehicles. Levenberg-Marquardt Backpropagation Neural Networks (LMBNNs) is extensively utilized in engineering and fluid mechanics for addressing intricate, nonlinear issues. Future research directions encompass the integration of LMBNNs with deep learning methodologies, real-time adaptive control, sophisticated nanofluid materials, Multiphysics modeling, computational fluid dynamics and Artificial Intelligence (AI) hybridization, as well as green AI and sustainable energy systems.}, archive = {J_EAAI}, author = {Aatif Ali and Shazia Habib and Zeeshan Khan and Adel Thaljaoui and Mei Sun and Taseer Muhammad}, doi = {10.1016/j.engappai.2025.111459}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111459}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An artificial neural network implementation on novel tetra hybrid nanofluid model: Case study of solar wind turbines}, volume = {158}, year = {2025}, }
• (2025). Artificial neural network integrated SHapley additive exPlanations modeling for sodium dichromate formation. EAAI, 158, 111457. (www)

  @article{ , abstract = {This Research has focused on optimizing metallurgical processes by integrating Artificial Neural Networks with Shapley additive explanation modeling. This approach helps understand the intricate relationships and mechanisms underlying chemical processes. Neural networks provide accurate predictions based on input parameter interactions, while Shapley values identify the relative importance of each input variable and offer detailed explanations for model predictions, enhancing transparency and interpretability. In this study on roasting and leaching processes for sodium dichromate formation, the neural network - Shapley modeling framework was employed. The goal was to uncover the intricate interplay between input variables and sodium dichromate formation, providing valuable insights for process optimization and prediction. Key factors such as temperature, roasting time, reaction time, and sulfuric acid concentration were optimized in relation to the efficacy of sodium dichromate formation under different settings. The suggested neural networks model predicted optimal yields for combined roasting and leaching settings. The optimum conditions included a roasting temperature of 1046.26 °C, roasting time of 2.7 h, Cr: NaCl ratio of 1.5, leaching time of 41 min at a temperature of 40 °C, and sulfuric acid concentration of 12M. Global sensitivity analysis revealed that the yields of different metals were directly influenced by the temperature during roasting, concentration of sulfuric acid, Cr:NaCl ratio, roasting time, leaching temperature, and leaching time. These parameters were ranked in terms of sensitivity coefficients, indicating their relative importance.}, archive = {J_EAAI}, author = {M.J. Mvita and N.G. Zulu and B. Thethwayo}, doi = {10.1016/j.engappai.2025.111457}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111457}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial neural network integrated SHapley additive exPlanations modeling for sodium dichromate formation}, volume = {158}, year = {2025}, }
• (2025). Regularized three-dimensional gaussians for large-scale building reconstruction with spatial prior. EAAI, 158, 111445. (www)

  @article{ , abstract = {Existing large-scale building reconstruction methods based on Neural Radiance Field (NeRF) usually suffer from severe radiative blur and extremely long training time. The emergence of three-dimensional Gaussian Splatting (3DGS) can produce realistic rendering quality, while possessing much faster rendering speed than NeRF-based methods. However, for large-scale building reconstruction, 3DGS tends to produce blurred details and floaters due to the incomplete coverage of structural details by multi-view images and influence of illumination changes. This paper proposes a novel approach for high-quality 3D reconstruction and real-time rendering of large-scale buildings based on regularized 3D Gaussians and the building spatial prior. Considering the correlation of adjacent views, two regularizers are introduced for the predicted geometry and color to reduce artifacts caused by unseen viewpoints. Moreover, the spatial prior information of building boundaries is incorporated into the optimization process to eliminate floaters on the surface through the point cloud filtering, thereby improving the visual quality of the reconstructed model. Experiments on real-world and public datasets demonstrate that the proposed method outperforms existing methods based on NeRF and 3DGS for large-scale building reconstruction in terms of rendering quality and geometric accuracy.}, archive = {J_EAAI}, author = {Jialin Hou and Rongrong Hou and Yuequan Bao}, doi = {10.1016/j.engappai.2025.111445}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111445}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Regularized three-dimensional gaussians for large-scale building reconstruction with spatial prior}, volume = {158}, year = {2025}, }
• (2025). Transformer based lifetime interval prediction for dynamic operating proton exchange membrane fuel cells. EAAI, 158, 111444. (www)

  @article{ , abstract = {Remaining useful life is crucial for proton exchange membrane fuel cell (PEMFC). However, the complex decay mechanism makes existing methods incapable of quantifying the PEMFC decay uncertainty. To address above issues, a hybrid interval prediction method (HIPM) is proposed. First, multi-feature fusion based on incremental empirical modal decomposition (IEMD) decomposes and reorganizes the nonlinear features of the PEMFC into multiscale degradation components. Second, the temporal Transformer effectively addresses the challenge of modeling long-term dependencies in PEMFC degradation prediction. Third, a novel interval prediction method precisely quantize the uncertainty of PEMFC degradation. Experimental results show HIPM achieves a root mean square error of 0.0047 with limited training data while accurately quantifying PEMFC degradation uncertainty across all conditions.}, archive = {J_EAAI}, author = {Haolong Li and Liang Xie and Dongqi Zhao and Ze Zhou and Liyan Zhang and Qihong Chen}, doi = {10.1016/j.engappai.2025.111444}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111444}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer based lifetime interval prediction for dynamic operating proton exchange membrane fuel cells}, volume = {158}, year = {2025}, }
• (2025). A survey on potential of hybrid soft computing techniques using for assessing the geotechnical properties of soil. EAAI, 158, 111443. (www)

  @article{ , abstract = {Compression index, hydraulic conductivity, shear strength, and swelling are the engineering properties of soil used to design safe and economical design of geotechnical structures. While conventional laboratory testing remains the standard, it is time-consuming, labor, and resource-intensive. Recent advances in data-driven modeling offer promising alternatives for rapid and reliable prediction of these properties using easily obtainable index parameters and soil composition data. This study reviews hybrid prediction techniques that integrate artificial intelligence (AI) with conventional models to enhance the accuracy of geotechnical property predictions. This leads to improvement in prediction accuracy and operational efficiency in geotechnical analysis. It follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses framework and includes bibliometric analysis using cluster graphs, discussions on future challenges, and a comparative evaluation of available models. Relevant articles published between 2000 and 2025 were retrieved from the Scopus and Web of Science databases. The review highlights that predicting engineering properties using index properties is a promising approach. A paradigm shifts from conventional correlation-based models, such as linear regression and multi-linear regression, toward soft computing algorithms (artificial neural networks and machine learning), and hybrid models involving swarm intelligence, grey wolf optimisers. These techniques enable the estimation of geotechnical properties, with reported prediction accuracies ranging from 48 % to 100 %, significantly reducing the reliance on traditional testing. Through a comprehensive analysis of existing prediction methods, this research underscores the growing role of AI-driven modeling in geotechnical engineering, paving the way for faster and data-driven decision-making.}, archive = {J_EAAI}, author = {Sayali B. Rautmare and Rupa S. Dalvi and Gayatri S. Vyas}, doi = {10.1016/j.engappai.2025.111443}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111443}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A survey on potential of hybrid soft computing techniques using for assessing the geotechnical properties of soil}, volume = {158}, year = {2025}, }
• (2025). A decomposition–integration interval prediction strategy for iron ore shipping freight rates with reinforcement learning. EAAI, 158, 111442. (www)

  @article{ , abstract = {The fluctuations in iron ore shipping freight rates significantly affect market participants’ investment decisions. This study proposes a “decomposition–integration” interval prediction method for iron ore shipping freight rates. First, the trend, seasonal, and residual components of the original data are extracted. Then, the residual component with high-frequency fluctuations is predicted by developing four interval prediction models with influencing factors. Reinforcement learning-derived dynamic weighting integrates prediction outputs to reveal the underlying data relationships. Finally, with point prediction outputs on the trend and seasonal components, the interval prediction is achieved by integrating these components predictions. The empirical results illustrate that the proposed method outperforms several benchmark methods in terms of prediction interval coverage probability of 61.29% and deviation of 0.02428, indicating its effectiveness in interval prediction.}, archive = {J_EAAI}, author = {Hongyue Guo and Yijia Zhang and Yating Yu and Lidong Wang and Peng Jia and Witold Pedrycz}, doi = {10.1016/j.engappai.2025.111442}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111442}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A decomposition–integration interval prediction strategy for iron ore shipping freight rates with reinforcement learning}, volume = {158}, year = {2025}, }
• (2025). A semi-supervised multi-view siamese network with dual-contextual attention and knowledge distillation for cross-lingual low-resource paraphrase detection. EAAI, 158, 111441. (www)

  @article{ , abstract = {Paraphrase detection is a pivotal task in natural language processing (NLP), with far-reaching implications for applications of artificial intelligence (AI) such as content moderation, machine translation, and misinformation detection. Despite advances in high-resource languages, it remains challenging for low-resource languages like Arabic and Urdu due to their complex syntax, morphology, flexible word order, and scarce annotated datasets. Cross-lingual transferability, particularly between typologically distinct languages (e.g., Arabic [Semitic] and Urdu [Indo-Aryan]), is understudied. To bridge this gap, we introduce the Multi-view Siamese Dual-Contextual Paraphrase Detection Network (MSDCPDN), a semi-supervised implemented AI framework specifically designed for paraphrase detection in low-resource multilingual contexts. The novelty of MSDCPDN lies in two key innovations: (i) a dual-contextual attention mechanism within a Siamese network that simultaneously captures fine-grained lexical-semantic relationships and global syntactic dependencies, overcoming the limitations of traditional methods, and (ii) a teacher–student knowledge distillation strategy, which transfers robust linguistic representations from a high-resource Arabic model to a lightweight student model optimized for Urdu. Designed for AI applications in low-resource settings, experimental results demonstrate that MSDCPDN achieves state-of-the-art accuracy in Arabic paraphrase detection and generalizes effectively to Urdu without target-language training, demonstrating strong interlingual transfer. This work bridges the gap in cross-lingual paraphrase detection through linguistically informed AI, emphasizing scalability for languages with structural disparities. The implemented AI framework and its application of AI techniques offer practical solutions for real-world challenges in misinformation mitigation and automated content analysis, underscoring the viability of resource-efficient, transferable NLP systems.}, archive = {J_EAAI}, author = {Anil Ahmed and Degen Huang}, doi = {10.1016/j.engappai.2025.111441}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111441}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A semi-supervised multi-view siamese network with dual-contextual attention and knowledge distillation for cross-lingual low-resource paraphrase detection}, volume = {158}, year = {2025}, }
• (2025). A survey of semantic extraction for speech semantic communications: Metrics, approaches, perspectives and challenges. EAAI, 158, 111439. (www)

  @article{ , abstract = {With the rapid development of communication technology, traditional bit transmission can no longer meet the demands of emerging intelligent applications for real-time communication, low latency, and efficient transmission. As a new communication paradigm, Semantic communication (SemCom) focuses on understanding and transmitting task-relevant semantic information, making it an important direction for improving communication efficiency. Researchers have extensively studied SemCom for text and images, but the research on speech is relatively limited. Semantic extraction is a key component of SemCom, and it varies significantly across modalities. Machine learning and deep learning techniques have significantly advanced speech signal processing and semantic extraction tasks, including speech embedding and automatic speech recognition, thus enhancing end-to-end speech SemCom networks. This paper provides a survey of the research progress in semantic extraction for speech SemCom. First, the paper describes the fundamentals of SemCom and speech signal processing, including acoustic feature and speech preprocessing. Taking into account the differences in the evaluation metrics between SemCom and traditional communication, this paper then presents the semantic metrics proposed in existing studies. Next, this paper compares recent progress in speech embedding and two commonly used methods for semantic extraction from speech: one based on speech-to-text conversion and the other on direct extraction from speech signals. Assistive methods for pragmatic-level semantic extraction are also discussed. Subsequently, existing studies on speech SemCom are reviewed and compared. Finally, we discuss the challenges and future directions in SemCom, with the aim of providing references for future research.}, archive = {J_EAAI}, author = {Chong Chen and Linyu Huang}, doi = {10.1016/j.engappai.2025.111439}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111439}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A survey of semantic extraction for speech semantic communications: Metrics, approaches, perspectives and challenges}, volume = {158}, year = {2025}, }
• (2025). Learning motion-guided salience features for weakly supervised group activity recognition. EAAI, 158, 111437. (www)

  @article{ , abstract = {This paper focuses on exploring motion-guided features for weakly supervised group activity recognition (GAR). Unlike existing GAR methods that simply squeeze extracted tokens or individual features into a single vector by global pooling, limiting their ability to sufficiently represent spatial and temporal salience features in videos, we propose a Motion-Guided Network (MGN) to capture crucial motion contextual information in videos. First, we embed local correlations between the feature maps of adjacent frames to extract motion features in activities. Then, unlike previous works that simply aggregate motion and appearance features by addition or concatenation, MGN uses motion representations to guide the extraction of temporal and spatial features. We have evaluated the proposed method on sports and group activity videos. Extensive experimental results verify the effectiveness of our method. Furthermore, our method has also outperformed some approaches trained with stronger supervision in the comparative evaluation.}, archive = {J_EAAI}, author = {Zexing Du and Qing Wang}, doi = {10.1016/j.engappai.2025.111437}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111437}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Learning motion-guided salience features for weakly supervised group activity recognition}, volume = {158}, year = {2025}, }
• (2025). Behavioral decision-making of mobile robots simulating the functions of cerebellum, basal ganglia and hippocampus. EAAI, 158, 111435. (www)

  @article{ , abstract = {In facing unknown and complex environments, the ability of robots to make efficient and good behavioral decisions is an important prerequisite for them to accomplish various tasks. When robots perform tasks in changing environments or in a new environment, they generally have poor decision-making skills and difficulty acquiring stable and continuous learning capabilities. Therefore, investigating how to make robots have human-like decision-making and learning ability and be able to make better decisions facing unknown complex tasks is currently an important challenge for robots. The authors have attempted to achieve flexible behavioral decision-making in dynamic environments of mobile robots by integrating the cerebellar supervised learning function and the reinforcement learning function of the basal ganglia, but the proposed algorithms have limited ability to adapt to dynamic environments. Based on previous work, this article redesigns the cerebellum module and adopts the motivated developmental network 2 to simulate the working mechanism of the cerebellum. Motivated developmental network 2 realizes the precise characterization of intrinsic features of the complex environment through the lateral connection between hidden neurons, in order to enhance the robot decision-making ability. A memory consolidation module is designed according to the working principle of the hippocampus, which makes the robot not only able to make good behavioral decision making when facing an unknown dynamic environment, but also able to continuously acquire knowledge from the unknown environment. Meanwhile, a state–action prediction method is proposed to realize flexible behavioral decision-making based on cerebellar supervised learning and basal ganglia’s reinforcement learning in dynamic environment, which further improves its decision-making ability. Experimental results in dynamic and real environments verify the potential of the model proposed in this paper.}, archive = {J_EAAI}, author = {Qi Liu and Wei Wang and Dongshu Wang}, doi = {10.1016/j.engappai.2025.111435}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111435}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Behavioral decision-making of mobile robots simulating the functions of cerebellum, basal ganglia and hippocampus}, volume = {158}, year = {2025}, }
• (2025). Deep learning and adaptive boosting for hydroelectric power prediction using hydro-meteorological data: Insights and feature importance analysis. EAAI, 158, 111434. (www)

  @article{ , abstract = {This study explores the use of modern artificial intelligence (AI) models such as Adaptive Boosting (AdaBoost), Autoencoder Based Regression Model (Autoencoder), Deep Neural Network (DNN), Echo State Network (ESN), Light Gradient Boosting Machine (LGBM) Residual Neural Network (ResNet) and feature importance techniques to predict energy production. Daily rainfall, streamflow, temperature and energy production data between 2016 and 2019 were used in the study. According to the feature importance analysis performed using the LGBM model, streamflow, temperature, and precipitation were identified as the most critical parameters influencing energy production. This structured AI-driven approach provided robust model evaluation and reliable performance measurements. As revealed by the analysis, the LGBM model showed the most accurate result with the values of root mean square error (RMSE: 98.27), mean absolute error (MAE: 73.93), Akaike information criterion (AIC: 17,100), Nash–Sutcliffe efficiency coefficient (NSE: 0.69), Kling–Gupta efficiency (KGE: 0.77), coefficient of determination (R 2 : 0.69), mean bias error (MBE: −5.23), bias factor (BF: 1.02), and percent bias (PBIAS: 1.51), while the DNN model came in with the second-best results. These findings emphasize both the effectiveness of deep learning and ensemble AI models and their practical application in improving hydroelectric power generation forecasting. Furthermore, the study underscores the importance of accurately monitoring and managing streamflow for the optimization of hydroelectric systems.}, archive = {J_EAAI}, author = {Yakup Karakoyun and Okan Mert Katipoğlu and Ahmet Dogan}, doi = {10.1016/j.engappai.2025.111434}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111434}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning and adaptive boosting for hydroelectric power prediction using hydro-meteorological data: Insights and feature importance analysis}, volume = {158}, year = {2025}, }
• (2025). Cooperative intelligent robust adaptive methodology for uncertain non-affine nonlinear multi-agent systems: Application to multi-robot formation flight. EAAI, 158, 111433. (www)

  @article{ , abstract = {The decentralized intelligent adaptive sliding mode formation controller is designed for uncertain non-affine nonlinear multi-agent systems (MAS) under a fixed directed topology. It is assumed that the dynamics of followers are unknown in the presence of uncertainties with specified structures. The function of the agents should satisfy the Lipschitz, controllability and observability conditions. To overcome the challenge of unknown functions of the followers, it uses a fuzzy system as a universal approximator to approximate the unknown functions adaptively. Fuzzy processes are used both to accommodate iteratively the unknowable nonlinearities and to extend the expert's expertise. Compared to the investigated methodologies that concentrate on the formation of either linear or affine nonlinear systems, the proposed approach deals with the formation of the non-affine nonlinear MAS with unknown functions of the agents. Moreover, in place of approximating of the unknown feature of the subsystems in other studies, this methodology distillates on estimation of the controller to lessen computational values. The focal features of the proposed scientific methods are 1) the achievement of the formation, 2) on-line upgrading of the control parameters, 3) boundedness of all signals engaged in the closed-loop system, 4) chattering minimization, 5) alignment of the tracking error to the neighborhood of zero (uniformly ultimately bounded), and 6) robustness of the closed-loop system. To illustrate the responsiveness and fluency of the procedure, a mathematical formulation of a multi-quad rotor is regarded with promising performance.}, archive = {J_EAAI}, author = {Reza Ghasemi}, doi = {10.1016/j.engappai.2025.111433}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111433}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cooperative intelligent robust adaptive methodology for uncertain non-affine nonlinear multi-agent systems: Application to multi-robot formation flight}, volume = {158}, year = {2025}, }
• (2025). A critical systematic review of machine learning based models for water leak detection using vibroacoustic technology. EAAI, 158, 111432. (www)

  @article{ , abstract = {Detecting leaks in water pipes is a critical task for water utilities, as it helps prevent water loss and damage to infrastructure. Machine learning (ML) techniques have shown potential in detecting water pipe leaks, but there are gaps and limitations in the ML modeling process that need to be addressed. This study presents a comprehensive evaluation of ML-based techniques for leak detection in water pipes, discussing the general steps to develop an ML-based model, including data collection, data processing, and model development. The study reviews a diverse array of ML-based methods, from traditional ML techniques to advanced deep learning approaches, and introduces their applicability and limitations. Finally, current research limitations and gaps are summarized following the reviews, and potential future directions are suggested, including overcoming data scarcity, enhancing current model performance and interpretability, assessing the potential of various ML-based approaches, and improving the practicability of water management for relevant researchers, engineers, and managers. This study provides a reference for academics and industrial practitioners interested in ML-based water leak detection models, especially in signal processing techniques, model selection and development, and practical applications. The findings contribute to understanding water pipe leak detection, promoting the development of water leak detection technologies, advancing the current water leak management, enhancing the decision-making process, and benefiting water utilities, academics, and end users.}, archive = {J_EAAI}, author = {Rongsheng Liu and Tarek Zayed and Rui Xiao}, doi = {10.1016/j.engappai.2025.111432}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111432}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A critical systematic review of machine learning based models for water leak detection using vibroacoustic technology}, volume = {158}, year = {2025}, }
• (2025). An improved anomaly detection model for automated inspection of power line insulators. EAAI, 158, 111431. (www)

  @article{ , abstract = {Inspection of insulators is important to ensure reliable operation of the power system. Deep learning is being increasingly exploited to automate the inspection process by leveraging object detection models to analyse aerial images captured by drones. A purely object detection-based approach suffers from class imbalance-induced poor performance, which can be accentuated for infrequent and hard-to-detect incipient faults. This article proposes the use of anomaly detection along with object detection in a two-stage approach for incipient fault detection in a data-efficient manner. An explainable one-class classifier is adopted for anomaly detection. The one-class formulation allows learning from images of healthy insulators and reduces the reliance on plentifully available images of faulty insulators. The explainability of the model, as opposed to black-box models, is expected to promote its adoption by the industry. A modified loss function is developed that addresses computational and interpretability issues with the existing model, also allowing for the integration of other losses. The superiority of the novel loss function is demonstrated on benchmark anomaly detection dataset improving the average performance by 0.47% and worst-case performance by 2%. The models are trained for insulator inspection with two datasets – representing data-abundant and data-scarce scenarios – in unsupervised and semi-supervised settings. The model achieves area under receiver operating characteristics curve of 0.89 and 0.88 for data-abundant and data-scarce scenarios, respectively. Results suggest that including as few as five real anomalies in the training dataset significantly improves the model’s performance and enables reliable detection of rarely occurring incipient faults in insulators.}, archive = {J_EAAI}, author = {Laya Das and Blazhe Gjorgiev and Giovanni Sansavini}, doi = {10.1016/j.engappai.2025.111431}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111431}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An improved anomaly detection model for automated inspection of power line insulators}, volume = {158}, year = {2025}, }
• (2025). Advancing electric vehicle inverter multi-class transient fault detection with sequential binary classification techniques. EAAI, 158, 111430. (www)

  @article{ , abstract = {Short-duration faults in electric vehicle inverters can severely compromise system stability and passenger safety. However, these faults have received relatively limited attention in the existing literature, particularly in terms of robust detection and classification approaches. This study proposes a sequential binary classification framework for multi-fault detection systems, focusing on transient Misfire and Firethrough faults. In contrast to a multi-classification scheme, an initial binary classification distinguishes healthy from faulty states, leading to a second classifier that discriminates intra-faults, improving detection accuracy. Statistical features alongside Shannon, Rényi, and Tsallis entropy were extracted, and a Sequential Feature Selector with Maximum Mean Discrepancy identifies the most informative features. Widely applicable machine learning tools in the field of fault detection have been selected to test the sequential binary-classification scheme with a multi-classification framework. Logistic Regression, k-Nearest Neighbors, Decision Tree, Support Vector Machine, Random Forest, Categorical Boosting, and Multi-Layer Perceptron are used with Bayesian hyperparameter optimization, and TabNet is optimized with Optuna. To achieve realistic operating conditions, randomness is introduced in fault occurrence, and Gaussian noise is superimposed on the signals. Results demonstrate that the sequential binary classification method outperforms direct multi-class approaches in detecting and identifying short-duration faults, enhancing predictive maintenance strategies in power electronics. By effectively isolating brief fault events, the proposed technique offers a computationally efficient solution that strengthens the reliability of electric vehicle systems and mitigates risks of severe damage.}, archive = {J_EAAI}, author = {Ilias Palaiologou and Georgios Falekas and Nikolaos Fesakis and Athanasios Karlis}, doi = {10.1016/j.engappai.2025.111430}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111430}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advancing electric vehicle inverter multi-class transient fault detection with sequential binary classification techniques}, volume = {158}, year = {2025}, }
• (2025). Heart failure disease stratification with temporal electronic health records: Patient representation and sub-phenotypes identification. EAAI, 158, 111429. (www)

  @article{ , abstract = {Heart failure disease stratification provides health status information for distinct patient subgroups, facilitating timely interventions. However, traditional techniques struggle with the complexity and longitudinality of electronic health records and require a predetermined number of groups, limiting their applicability. To address these challenges, we introduce a novel patient representation model using a temporal dual bidirectional neural network with an attention mechanism to effectively model electronic health records. We also propose a density peak-based clustering method that identifies distinct clinical heart failure sub-phenotypes with minimal manual intervention. Our deep learning model effectively processes electronic health records to generate patient representations. Our clustering method uses an adaptive cut-off distance to reduce manual input and employs affinity propagation’s responsibility matrix to determine each patient’s local density, automatically identifying sub-phenotype centroids. Evaluation using a real-world clinical dataset and various indicators shows our methods outperform existing approaches. Importantly, they can identify clinically meaningful heart failure sub-phenotypes, with statistical test results holding practical value for clinical decision-making.}, archive = {J_EAAI}, author = {Ye Liang and Chonghui Guo and Hailin Li}, doi = {10.1016/j.engappai.2025.111429}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111429}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Heart failure disease stratification with temporal electronic health records: Patient representation and sub-phenotypes identification}, volume = {158}, year = {2025}, }
• (2025). Can foundation language models predict fluid dynamics?. EAAI, 158, 111427. (www)

  @article{ , abstract = {The application of deep learning-based, data-driven methods to fluid dynamics problems has attracted significant interest in recent years, due to their potential for faster flow predictions (model inferences) and reduced computational requirements compared to traditional Computational Fluid Dynamics (CFD) methods. Despite their success, existing supervised deep learning approaches typically require extensive model design and large amounts of high-quality training data to achieve satisfactory performance for each specific fluid flow problem. The emergence of foundation models—pre-trained on large-scale, multidisciplinary datasets and capable of solving a wide range of downstream tasks—raises the question of whether their capabilities can extend to scientific domains such as fluid dynamics. Most foundation models to date have been rooted in language tasks. This study investigates a specific instantiation of a foundation model, the Large Language Model Meta AI (Llama) 3 model, for predicting fluid flows with varying dynamical complexities. Results show that the Llama 3 foundation model, although originally designed for natural language processing, can be applied to fluid dynamics problems with simple engineering adaptations and without fine-tuning its pre-trained weights. This approach achieves improved accuracy and robustness compared to conventional fully connected or recurrent neural network models with comparable capacities and equivalent training data. Furthermore, fine-tuning the model by injecting problem-specific knowledge into the pre-trained weights further enhances its performance. These findings suggest that foundation models hold promise for fast inference and deployment of solutions to fluid dynamics problems. Whether unified with models from other domains (e.g., language models) or developed specifically for fluid dynamics, such models could become powerful tools for improving predictive accuracy and computational efficiency in engineering and scientific applications. The code is accessible at https://gitlab.com/houWGW/llm_for_fluid_dynamics .}, archive = {J_EAAI}, author = {Ganwei Wang and Sibo Cheng}, doi = {10.1016/j.engappai.2025.111427}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111427}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Can foundation language models predict fluid dynamics?}, volume = {158}, year = {2025}, }
• (2025). Probabilistic similarity preservation for distribution discrepancy reduction in domain adaptation. EAAI, 158, 111426. (www)

  @article{ , abstract = {Many existing domain adaptation methods attempt to reduce inter-domain differences without accounting for the varying weights of the feature vectors in the two domains. Since the labels estimated for the target domain data may not be accurate when calculating the conditional probability distributions, the domain adaptation methods based on these distributions are negatively impacted by these noisy labels. In this paper, a new domain adaptation method is proposed to reduce inter-domain discrepancy by considering the group relationships of samples based on the potential similarities derived from the distances between samples in source and target domains. To achieve this goal, a new similarity matrix is introduced to extract prototypes from the target domain represented as a combination of samples feature vector. These prototypes are extracted by trying to preserve the probabilistic similarities to the source domain labeled samples, limiting the impact of noisy labels when calculating the conditional probability distributions in the target domain. Following this, the conditional probability distribution divergences between the two domains for each class are reduced by imposing a common distribution function on these feature vectors. Extracting the prototypes from the target domain as well as considering similar distribution form for the feature vectors of the two domains, allow the discrepancy between the two domains to be reduced more effectively. Three different datasets are used to evaluate the proposed method: Office-31, Office-Home, and ImageCLEF. Experimental results on these datasets demonstrate that our approach achieves considerably higher accuracy compared to other recent related works.}, archive = {J_EAAI}, author = {Elham Hatefi and Hossein Karshenas and Peyman Adibi}, doi = {10.1016/j.engappai.2025.111426}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111426}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Probabilistic similarity preservation for distribution discrepancy reduction in domain adaptation}, volume = {158}, year = {2025}, }
• (2025). An advanced artificial intelligence framework integrating ensembled convolutional neural networks and vision transformers for precise soil classification with adaptive fuzzy logic-based crop recommendations. EAAI, 158, 111425. (www)

  @article{ , abstract = {This study introduces an advanced Artificial Intelligence (AI) framework for soil classification and crop recommendation, combining Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) in an ensemble approach, alongside an adaptive fuzzy logic-based decision system for crop suggestions. While existing research typically addresses soil classification or crop recommendation in isolation, this work integrates cutting-edge deep learning models and fuzzy logic to enhance both tasks. The methodology is divided into two phases: Phase 1 covers data collection, preprocessing, and augmentation using Cycle Generative Adversarial Networks (CycleGAN) to expand the curated dataset of 1189 soil images to 8,413, while Phase 2 focuses on training CNN and ViT models, ensembling these models, and developing a fuzzy logic system that considers soil type, nutrients, potential of hydrogen (pH), and climatic conditions for crop recommendations. Experimental results indicate models achieve classification accuracies of up to 89.32 % on the original dataset, improving to 91.01 % with augmented data. On the CycleGAN-augmented (CyAUG) dataset, EfficientNet v2 Large and ViT-Large/16 attain accuracies of 99.60 % and 99.73 %, respectively. Furthermore, an ensemble of these architectures achieves a perfect accuracy of 100 %. The results are also validated by K-fold cross-validation. The research also presents 'Agro Companion,' an AI-powered tool that assists farmers in soil identification and crop selection based on geological and environmental data. This framework addresses key agricultural challenges in India, offering a high-accuracy, practical solution for improving both soil classification and crop recommendation. This research delivers state-of-the-art soil classification performance and a robust AI-based crop recommendation tool to support sustainable agricultural practices.}, archive = {J_EAAI}, author = {Farhan Sheth and Priya Mathur and Amit Kumar Gupta and Sandeep Chaurasia}, doi = {10.1016/j.engappai.2025.111425}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111425}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An advanced artificial intelligence framework integrating ensembled convolutional neural networks and vision transformers for precise soil classification with adaptive fuzzy logic-based crop recommendations}, volume = {158}, year = {2025}, }
• (2025). An interpretable privacy-preserving real-time carbon emission estimation approach for heterogeneous industrial enterprises. EAAI, 158, 111420. (www)

  @article{ , abstract = {Accurate monitoring of carbon emissions from high energy-consuming enterprises is foundational to the low-carbon development of the power system. The volume trend and energy consumption structure of diverse industrial enterprises are different. Most current carbon emission accounting relies on annual energy consumption statistics, with one year or more lag. Additionally, there are severe data barriers between various enterprises, considering the consequences of protecting their carbon-related data. This paper proposes an innovative carbon emission monitoring method to address the aforementioned issues. The first step of this method is to construct an accurate and high-frequency carbon emission measurement architecture from the multi-dimensional perspective of data flow and carbon emission flow. Next, three pivotal techniques are employed in hybrid model: variational mode decomposition (VMD), temporal convolutional network (TCN), long short-term memory network with multi-head attention (LSTMMA)—collectively referred to as VMD-TCN-LSTMMA model. The proposed model can decompose data, increase dimension, and extract features of power time series to precisely estimate the direct carbon emissions of industrial enterprises. Moreover, a model training and information sharing framework based on differential privacy federated score weight algorithm (DP-FedSW) is developed to improve monitoring accuracy for enterprises without divulging raw data. A dataset from 142 users in 6 different types of high energy-consuming industries is collected to comprehensively evaluate the proposed monitoring method's performance. Experimental results demonstrate that the proposed method outperforms the conventional strategy, which can improve the stability and effectiveness of the carbon emission estimation model, ensuring the security of raw data.}, archive = {J_EAAI}, author = {Xiangyu Kong and Riwei Zhang and Bixuan Gao and Gaohua Liu and Kaijie Fang and Meimei Duan}, doi = {10.1016/j.engappai.2025.111420}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111420}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An interpretable privacy-preserving real-time carbon emission estimation approach for heterogeneous industrial enterprises}, volume = {158}, year = {2025}, }
• (2025). A dual-population based two-archive coevolution algorithm for constrained multi-objective optimization problems. EAAI, 158, 111419. (www)

  @article{ , abstract = {How to balance the objectives and constraints better is the key to solving constrained multi-objective optimization problems (CMOPs). Many evolutionary algorithms struggle to fully converge to the entire Pareto front, especially in CMOPs with narrow and complex feasible regions, which posing significant challenges in solving CMOPs. To handle this problem, the paper proposes a dual-population based two-archive coevolution algorithm (DPTAC). The main population evolves towards the true Pareto front while accounting for considering the original problem. The auxiliary population ignores the constraints and approximates unconstrained Pareto front. To assist main population in crossing larger infeasible regions, enhancing its diversity, and discovering more feasible regions, a two-archive strategy is proposed, which stores the potentially valuable non-dominated infeasible solutions and non-dominated solutions generated by the evolution of the main population and the auxiliary population respectively. In addition, a removal mechanism is introduced and integrated into the auxiliary population to reduce computational resource waste. This can help the main population have more computational resources in the late stage of evolution to find narrow feasible regions and improve the convergence of the population. Experimental results demonstrate that DPTAC outperforms 9 state-of-the-art constrained multi-objective evolutionary algorithms (CMOEAs) across 5 test suites comprising 62 benchmark functions and 6 real-world problems, confirming its superior competitiveness.}, archive = {J_EAAI}, author = {Miao Chen and Shijie Zhao and Tianran Zhang and Lei Zhang}, doi = {10.1016/j.engappai.2025.111419}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111419}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual-population based two-archive coevolution algorithm for constrained multi-objective optimization problems}, volume = {158}, year = {2025}, }
• (2025). Three-dimensional reconstruction image generation of traditional chinese painting elements. EAAI, 158, 111417. (www)

  @article{ , abstract = {This paper presents a comprehensive pipeline for generating detailed three-dimensional (3D) models from single images of traditional Chinese painting elements. This task is particularly challenging due to the lack of 3D datasets for Chinese paintings and the limited research on their 3D reconstruction. As a result, direct access to multiple views is precluded. We propose a novel method for the 3D reconstruction of T raditional C hinese P ainting E lements, termed TCPE-3D , which has three components of the process. The first component is a multi-view synthesis module named One To Six (OTX) - Multi-View Generating (MVG) Module. This module creates six fixed-view images through a series of preprocessing steps. These images are used to generate the Local Light Field Fusion (LLFF) dataset within the Neural Radiance Fields (NeRF) synthesis module. This process leads to the creation of detailed mesh structures in the final Mesh Generation module. Comparison with several state-of-the-art 3D reconstruction methods shows that our framework achieves better visualization results and higher technical specifications. Additionally, it solves the Janus problem encountered by other algorithms for Chinese painting data. Our dataset is available at https://github.com/LPDLG/3DTCP-Dataset .}, archive = {J_EAAI}, author = {Qiyao Hu and Jingyu Wang and Xianlin Peng and Tengfei Li and Rui Cao}, doi = {10.1016/j.engappai.2025.111417}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111417}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-dimensional reconstruction image generation of traditional chinese painting elements}, volume = {158}, year = {2025}, }
• (2025). Advancements in deep learning and explainable artificial intelligence for enhanced medical image analysis: A comprehensive survey and future directions. EAAI, 158, 111413. (www)

  @article{ , abstract = {With the notable advancements in deep learning, the demand for explainability and interpretability has witnessed a substantial surge, particularly within the medical domain, where critical decisions are at stake. The concept of Explainable Artificial Intelligence (XAI) has emerged as a burgeoning field focused on elucidating the opaque nature of deep learning systems by evaluating these models and providing insights into their decision-making processes. Due to their inherent lack of transparency, deep learning models often fall short of explicitly articulating the rationale behind their predictions. Consequently, integrating most deep learning algorithms into real-world clinical decision-making workflows remains limited. Our review paper presents a comprehensive analysis of XAI techniques in the context of medical imaging analysis. Through a systematic review of existing literature, we offer insights into the comparative landscape of XAI approaches, meticulously evaluating prominent medical imaging datasets and identifying gaps in prior research. We provide a concise overview of state-of-the-art interpretable models and post-hoc methods, accompanied by an assessment of suitable evaluation metrics for benchmarking these approaches. Additionally, we introduce a taxonomy for categorizing XAI methods, encompassing visual, textual, example-based, and concept-based approaches. Our discussion extends to future research directions and challenges, aiming to guide the development of innovative solutions in this rapidly evolving field.}, archive = {J_EAAI}, author = {Aqib Nazir Mir and Danish Raza Rizvi}, doi = {10.1016/j.engappai.2025.111413}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111413}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advancements in deep learning and explainable artificial intelligence for enhanced medical image analysis: A comprehensive survey and future directions}, volume = {158}, year = {2025}, }
• (2025). GOM-MMOEA: Multimodal multi-objective evolutionary algorithm based on global orchestration mechanism. EAAI, 158, 111412. (www)

  @article{ , abstract = {The core challenge of multimodal multi-objective optimization lies in identifying and discovering multiple equivalent sets of Pareto-optimal solutions, thereby offering diverse options for decision-makers. However, most existing algorithms suffer from premature convergence when tackling such problems. This issue often arises due to inadequate population diversity and ineffective global exploration mechanisms during the search process, which causes the algorithm to become trapped in local optima and hinders the exploration of other promising regions in the decision space. To address this challenge, this paper proposes a multimodal multi-objective evolutionary algorithm based on a global orchestration mechanism. First, the algorithm constructs and dynamically updates an orchestration vector to guide the search toward optimal solutions and accelerate population convergence. Second, an orchestration vector update strategy is designed to gradually diminish the influence of inferior solutions, thereby preventing convergence to local optima. During the early stages of evolution, larger increments are applied to high-quality solutions to speed up convergence, while these increments are gradually reduced over time to promote global exploration. Finally, a novel parent selection mechanism is introduced, which dynamically adjusts selection probabilities to optimize the search process while preserving population diversity. Moreover, the algorithm adopts a triple population synergistic orchestration method that simultaneously considers both the objective and decision spaces. Experimental results demonstrate that the proposed algorithm outperforms several state-of-the-art methods across a range of benchmark test problems.}, archive = {J_EAAI}, author = {Shaobo Deng and Hui Shi and Hangyu Liu and Jinyu Xu and Sujie Guan and Min Li and Zhuolei Duan}, doi = {10.1016/j.engappai.2025.111412}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111412}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {GOM-MMOEA: Multimodal multi-objective evolutionary algorithm based on global orchestration mechanism}, volume = {158}, year = {2025}, }
• (2025). Enhanced agent-oriented programming for robot teams. EAAI, 158, 111390. (www)

  @article{ , abstract = {The employment of Belief-Desire-Intention (BDI) agents offers a range of appealing features in the design of autonomous systems, including autonomy, adaptability, reactivity, and cooperation. However, when considering using BDI agents for controlling teams of multiple robots, few works explore the full potential of this approach. In this paper, we aim to integrate various BDI agent features to control a cooperative team of autonomous firefighting Unmanned Aerial Vehicles (UAVs), utilizing a layered architecture that combines the Jason BDI language with the Robot Operating System (ROS). In addition, we also present a new Mission-Management (MM) library, whose internal mechanisms enable the declarative definition of various missions, such as search-fire , extinguish-fire ; assist other UAV , and refill . Missions can be temporarily suspended, resumed, or even interrupted. The underlying mechanism orchestrates the missions’ execution, making it simple to navigate between them. Our contention is that this library helps the developer to further exploit an important feature of BDI agents: the balance between reactiveness and proactiveness. An evaluation scenario was developed extending our previous firefighting-UAV application with significantly more functionality, incorporating wind direction and the finite nature of both battery-power and fire-retardant. Our new agent also includes a more sophisticated agent-collaboration algorithm based on the well-established Contract Net Protocol (CNP). Despite these significant extensions, the MM library ensured that the extended agent did not dramatically increase (around 17%) the code size of the original agent. Our experiments also show that the proposed solution does not present performance issues when scaling the application to more agents.}, archive = {J_EAAI}, author = {Iago de Oliveira Silvestre and Leandro Buss Becker and Michael Fisher and Jomi Fred Hübner and Maiquel de Brito}, doi = {10.1016/j.engappai.2025.111390}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111390}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced agent-oriented programming for robot teams}, volume = {158}, year = {2025}, }
• (2025). UTCR-dehaze: U-net and transformer-based cycle-consistent generative adversarial network for unpaired remote sensing image dehazing. EAAI, 158, 111385. (www)

  @article{ , abstract = {To address issues of feature loss and color differences in existing unpaired dehazing methods for Remote Sensing images, we propose a method based on a U-Net and Transformer-based Cycle-Consistent Generative Adversarial Network for unpaired remote sensing image dehazing (UTCR-Dehaze). In this model, considering that paired hazy images are difficult to obtain, a cycle-consistent generative adversarial network (CycleGAN) is used to achieve remote sensing image dehazing. Due to the multi-scale features of remote sensing images, U-Net is combined with Transformer as the generator of CycleGAN. The generator learns the relationship between low-frequency and high-frequency features of the image at multiple scales. The U-Net encoder–decoder processes the high-frequency features, and the transformer at the bottleneck of U-Net learns the low-frequency feature relationship to restore image details and structures. Secondly, to further improve the details and clarity of dehazed images, a Mixed Cascade Group Attention module (MCGA) is designed. MCGA captures the global information of the image through cascade group attention and focuses on local information through Dehaze input-dependent depthwise convolution, thus better learning image features. In addition, to reduce feature loss and color differences in dehazed images, a Cycle Perceptual Identity Consistency Loss is designed, which combines perceptual and identity losses to maintain the details of input images through cycle consistency. Numerous experiments on synthetic and real remote sensing datasets show that, compared with previous methods, this method not only removes haze more accurately but also preserves image details and colors to the greatest extent.}, archive = {J_EAAI}, author = {Canlin Li and Xiangfei Zhang and Hua Wang and Zhiwen Shao and Lizhuang Ma}, doi = {10.1016/j.engappai.2025.111385}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111385}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {UTCR-dehaze: U-net and transformer-based cycle-consistent generative adversarial network for unpaired remote sensing image dehazing}, volume = {158}, year = {2025}, }
• (2025). Efficient reconfigurable system for home monitoring of the elderly via action recognition. EAAI, 158, 111383. (www)

  @article{ , abstract = {The rapid growth of the aging population poses serious challenges for our society e.g. the increase of the healthcare costs. Smart-Health Cyber–Physical Systems (CPSs) offer innovative solutions to ease this burden. This work proposes a general framework adapted in run-time to optimize the system’s overall performance, continuously monitoring system working qualities such as response time, accuracy, or energy consumption. Adaptation is achieved through the automatic deployment of different artificial intelligence (AI) based models on local edges (particularly deep learning models (DL)). Local processing is performed in embedded devices that provide short latency and real-time processing despite their limited computation capacity compared to high-end cloud servers. The paper validates this reconfigurable CPS in a challenging scenario: indoor ambient assisted living for the elderly. Our system collects lifestyle user data in a non-invasive manner to promote healthy habits and triggers alarms in case of emergency. Local edge video processing nodes identify indoor activities powered by state-of-the-art deep-learning action recognition models. The optimized embedded nodes locally reduce cost and power consumption, but the system still needs to maximize the overall performance in a changing environment. To that end, our solution enables run-time reconfiguration to adapt in terms of functionality or resource availability, offloading computation when required. The experimental section shows a real setup performing run-time adaptation with different reconfiguration policies considering average times for different daily activities. For that example, the adaptation extends the working time in more than 60% and achieves a 3x confidence in recognition for critical actions.}, archive = {J_EAAI}, author = {Daniel Deniz and Juan Isern and Jan Solanti and Pekka Jääskeläinen and Petr Hnětynka and Lubomír Bulej and Eduardo Ros and Francisco Barranco}, doi = {10.1016/j.engappai.2025.111383}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111383}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient reconfigurable system for home monitoring of the elderly via action recognition}, volume = {158}, year = {2025}, }
• (2025). An intelligent heart disease prediction framework by optimal feature integration with multi serial cascaded adaptive dilated network using heuristic improvement. EAAI, 158, 111379. (www)

  @article{ , abstract = {Problem statement Heart disease remains a leading cause of death globally, and early detection is crucial for effective intervention and improved patient outcomes. The challenge in existing approach limits the accuracy and efficiency, leading to missed diagnoses. Contribution This research work developed an adaptive system to predict cardiac illness using a heuristic assisted deep learning approach which can accurately and efficiently predict the risk of heart disease from patient data and improved patient outcomes. Initially the data was gathered and deep features from gathered data are extracted by auto encoder, then it is given to the suggested Improved Uniform Random Variable based Equilibrium Optimizer for choosing the optimal features and optimizes the weights of the autoencoder. The optimally selected features are combined for the formation of the optimal weighted features. These features are provided as the input to the Multi Serial Cascaded Adaptive Dilated Network. This network is formed by combining a Deep Temporal Convolution Network, Long Short-Term Memory, and One-Dimensional Convolution Neural Network. For enhancing the prediction accuracy, the parameters in the suggested Multi Serial Cascaded Adaptive Dilated Network are tuned by Improved Uniform Random Variable based Equilibrium Optimizer. Findings The result demonstrated that the accuracy of the developed framework is 94.40 %, which is enhanced than other existing approaches. Thus, the designed system can detect heart disease and improving patient outcomes. Application By analyzing patient data and predicting potential outcomes, the developed framework can help clinicians for personalized treatment plans that are tailored to individual needs and risk factors.}, archive = {J_EAAI}, author = {Shabana Urooj and Rajesh Arunachalam and Neha Verma and Priyanka Bhutani and Suresh Babu Kondaveeti and Aime Lay Ekuakille}, doi = {10.1016/j.engappai.2025.111379}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111379}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An intelligent heart disease prediction framework by optimal feature integration with multi serial cascaded adaptive dilated network using heuristic improvement}, volume = {158}, year = {2025}, }
• (2025). Modeling temporal continuity of spatial interactions for vessel trajectories prediction in maritime transportation systems. EAAI, 158, 111378. (www)

  @article{ , abstract = {Learning the intricate spatio-temporal interactions among vessels poses a fundamental challenge to vessel trajectory prediction (VTP). Prevailing graph neural network-based methods typically employ a three-stage independent architecture, modeling spatial and temporal dependencies separately, which not only overlooks the temporal continuity of spatial interactions but also introduces higher computational overhead. Moreover, considering interaction semantics at distinct moments equally during the modeling process may restrict the effective learning for certain key features. To address these issues, our paper proposes an innovative Heterogeneous Spatio-temporal Graph Neural Network (HSTGN) that employs hierarchical aggregations, encompassing vessel node-level, temporal edge-level, and self-attention-level aggregations, tailored for VTP. Specifically, a heterogeneous graph is first constructed through the transformation of node and edge generators, allowing vessels to interact freely with each other across temporal dimensions. A key innovation is the vessel multi-relational graph aggregation module (VMGA), which selectively aggregates edge-specific semantics and assigns adaptive weights to enhance interaction modeling. Furthermore, we devise a gated temporal convolution module (GTCN) to compensate for potential global dependencies that VMGA might not capture. Finally, HSTGN integrates VMGA and GTCN in parallel to refine spatio-temporal correlations for more precise prediction. This study enhances VTP accuracy and efficiency, enabling safer navigation and optimized maritime route planning. Extensive experiments on three real-world Automatic Identification System (AIS) datasets demonstrate that HSTGN outperforms state-of-the-art approaches, achieving a 9.47% improvement in mean absolute error (MAE), 1.43% in root-mean-square error (RMSE), 2.72% in mean absolute percentage error (MAPE), and a 43.06% increase in inference speed, while utilizing fewer model parameters.}, archive = {J_EAAI}, author = {Xiliang Zhang and Jin Liu and Chengcheng Chen and Peizhu Gong and Zhongdai Wu and Lei Guo}, doi = {10.1016/j.engappai.2025.111378}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111378}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Modeling temporal continuity of spatial interactions for vessel trajectories prediction in maritime transportation systems}, volume = {158}, year = {2025}, }
• (2025). Optimizing urban block morphology for photovoltaic power and thermal comfort in hot and humid regions. EAAI, 158, 111377. (www)

  @article{ , abstract = {This study explores the influence of urban block shape on solar energy potential and outdoor thermal comfort. A multi-objective optimization model was developed to improve both photovoltaic efficiency and thermal comfort in hot and humid regions. Using Rhino/Grasshopper parametric modeling, a digital prototype of a typical block in Foshan Shishan Town was created. Performance was evaluated through coupled simulations with Ladybug Tools. Results show that the Non-dominated Sorting Genetic Algorithm II (NSGA-II) produced Pareto front solutions. The north-south high-low bench layout increased photovoltaic efficiency by 18 % and improved thermal comfort by lowering summer temperatures by 0.8–1 °C. Additionally, placing water bodies in the southern and central parts of the block reduced ambient temperatures by 1.2–2.0 °C. Based on these findings, the study proposes guidelines for low-carbon community planning in hot and humid areas. The suggested layout—"multi-layer (south) to high (middle), multi-layer (north)"—can achieve 12–15 % energy-carbon sink synergy and a 20–25 % reduction in air conditioning load. Density-Based Spatial Clustering of Applications with Noise (DBSCAN) analysis shows that this framework has an 84 % applicability to planning schemes in similar climates. The study offers both algorithmic innovations and practical solutions for low-carbon community development.}, archive = {J_EAAI}, author = {Boyang Bai and Tianyin Li and Shiliang Wang and Hainan Yan and Junjie Dong}, doi = {10.1016/j.engappai.2025.111377}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111377}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing urban block morphology for photovoltaic power and thermal comfort in hot and humid regions}, volume = {158}, year = {2025}, }
• (2025). Research, application, and challenges of causal inference in industrial fault diagnosis: A survey. EAAI, 158, 111376. (www)

  @article{ , abstract = {Industrial fault diagnosis technologies leveraging convolutional neural networks and other advanced neural network architectures are pivotal for ensuring stable equipment operation, enhancing production efficiency, and minimizing maintenance costs. Nevertheless, these methods encounter inherent challenges due to data constraints and the complexity of production environments, particularly in identifying fault root causes and ensuring the interpretability of models. The integration of causal inference into industrial fault diagnosis offers significant promise for elucidating fault propagation pathways, revealing causal interrelations within complex systems, and advancing model interpretability. This survey presents a holistic review of research trajectories, pivotal technologies, and methodological advancements in causal inference for industrial fault diagnosis while systematically delineating the advantages and prospective challenges in this domain. First, this paper examines the limitations of conventional machine-learning approaches in fault diagnosis and traces the evolutionary trajectory of causal inference development in this context. Subsequently, the core theories and foundational technologies underpinning causal inference in industrial fault diagnosis are comprehensively discussed. Following this, the survey categorizes the existing literature according to different causal inferences to solve specific problems in industrial fault diagnosis and delves into detailed case studies, underscoring their utility in addressing distinct challenges. Finally, this survey synthesizes insights from existing literature to encapsulate the merits of causal inference in industrial fault diagnosis and to elucidate the prospective challenges it may encounter.}, archive = {J_EAAI}, author = {Bo Li and Qiang Li and Tingfeng Du and Dong Liu and Qiang Yang and Tianxiang Chen and Jing Xiong and Bo Peng and Junxiao Ren and Ji Zhao}, doi = {10.1016/j.engappai.2025.111376}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111376}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research, application, and challenges of causal inference in industrial fault diagnosis: A survey}, volume = {158}, year = {2025}, }
• (2025). Remote glacier monitoring through semantic fusion of geographic and contextual data. EAAI, 158, 111366. (www)

  @article{ , abstract = {Glacier melting, due to climate change, is a growing concern with many implications for the planet. Although remote sensing technology offers valuable solutions, the data acquired is often heterogeneous, varying in format and originating from different sensors and sources. Harmonizing this heterogeneous data is essential to ensure compatibility and integration. This paper introduces an integrated model that combines Machine Learning methods for satellite image segmentation with semantic web technologies for knowledge base construction aimed at retrieving relevant data concerning glacier monitoring implications. An ad-hoc ontology was developed to model the knowledge base about the glacier domain. At the same time, semantic segmentation allows the elicitation of relevant features combined with contextual meteorological data to populate the knowledge base. The synergy between ontology-based annotation and Deep Learning techniques for segmenting remote sensing images enables a more comprehensive assessment of glacial health, facilitating the retrieval of specific information via semantic queries. The effectiveness of the proposed system is shown by evaluating the performance of the Deep Learning models and the consistency and robustness of the ontology in processing complex queries.}, archive = {J_EAAI}, author = {Giacomo Albamonte and Giorgio Falcone and Sabrina Senatore}, doi = {10.1016/j.engappai.2025.111366}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111366}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Remote glacier monitoring through semantic fusion of geographic and contextual data}, volume = {158}, year = {2025}, }
• (2025). A multi-head attention-based bidirectional gated recurrent unit and multilayer perceptron for relation extraction model. EAAI, 158, 111365. (www)

  @article{ , abstract = {Aiming at the problems of insufficient text feature extraction, weak keyword information representation ability, and inaccurate and inadequate expression of model semantic relations in relation extraction, This paper proposes a Multi-Head Attention-based Bidirectional Gated Recurrent Unit and Multi-Layer Perceptron for relation extraction model (Att+BERT+FFBiGRU). The model can enrich the feature information of text vector, improve the representation of keyword information in sentences and enhance the expression of the relation between words. To be specific, firstly, BERT was fine-tuned by a method that added the position and part-of-speech tags of words into the text as special tags in order to improve the representation of word embedding, and a multi-head attention mechanism was introduced to make the neural network focus on the keyword features in the sentence during training. Secondly, in order to overcome the limitations of linear regression, multi-layer perceptron was introduced to solve the interaction between features. Thirdly, the multi-layer perceptron information was fed back to the BiGRU’s hidden layer to improve the accuracy of semantic information represented by the hidden layer and the quality of semantic features captured by the model. Finally, the asymmetric kernel inner product is used to capture the semantic relations between words, and the Softmax classifier is used to extract the relations. The experiments show that the F1 score of the proposed model on the datasets SemEval-2010 Task 8, TACRED and NYT-10 achieves 92.45%, 81.79% and 84.77%, which is 0.65%, 4.99% and 6.17% higher than the best model using the same dataset, respectively.}, archive = {J_EAAI}, author = {Bailin Chen and Erping Zhao and BianBaDroMa and Ngodrup}, doi = {10.1016/j.engappai.2025.111365}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111365}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-head attention-based bidirectional gated recurrent unit and multilayer perceptron for relation extraction model}, volume = {158}, year = {2025}, }
• (2025). An explainable artificial intelligence driven fall system for sensor data analysis enhanced by butterworth filtering. EAAI, 158, 111364. (www)

  @article{ , abstract = {The detection of falls is an essential component of healthcare monitoring systems, especially for older people at a greater risk of falling than younger people. To address the shortcomings of previously established methodologies, this research proposes a unique Artificial Intelligence driven sensor-based methodology that utilizes the SisFall dataset in conjunction with a Recurrent Neural Network - Long Short-Term Memory model. Two methods were considered: one using a Butterworth filter and the other without filtering. The results emphasize the significance of noise reduction in enhancing model performance. Additionally, the integration of Explainable Artificial Intelligence techniques brings transparency and interpretability to the model’s predictions, enhancing its dependability and trustworthiness in healthcare applications. Using Artificial Intelligence driven fall detection with Explainable Artificial Intelligence for transparent decision-making, this methodology presents a robust approach to improving accuracy and reducing false alarms in real-world healthcare settings. The study demonstrates that combining advanced filtering techniques with Explainable Artificial Intelligence algorithms successfully overcomes the challenges associated with traditional fall detection systems. The findings further confirm that the application of an Artificial Intelligence based Butterworth filter significantly enhances model accuracy, achieving 98.96% compared to 79.77% without filtering. These findings highlight the potential of Artificial Intelligence driven fall detection systems in healthcare, paving the way for more accurate, interpretable, and reliable monitoring solutions that can enhance elderly safety and improve real-time clinical decision-making.}, archive = {J_EAAI}, author = {Shalini J. and Ashok Kumar L.}, doi = {10.1016/j.engappai.2025.111364}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111364}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An explainable artificial intelligence driven fall system for sensor data analysis enhanced by butterworth filtering}, volume = {158}, year = {2025}, }
• (2025). Multi-step citywide traffic flow forecasting based on multiscale spatio-temporal transformer. EAAI, 158, 111362. (www)

  @article{ , abstract = {Accurate citywide traffic flow forecasting is an essential task in intelligent transportation systems. Unlike single-step forecasting, multi-step traffic flow forecasting offers extended insights that support proactive traffic management and resource allocation over longer time horizons. This paper proposes a Multi-Step Multiscale Spatial–Temporal Transformer (MS-MSTformer) for citywide traffic flow forecasting, which leverages a multiscale patch mechanism to capture both local and global spatial dependencies while integrating temporal patterns of closeness, period, and trend. Two novel cross-attention modules, namely Patch-Temporal Cross-Attention (PTCA) and Region-Temporal Cross-Attention (RTCA) are presented. These modules utilize temporal information as the query, with PTCA and RTCA focusing on patches and regions, respectively, to effectively fuse diverse spatio-temporal features. Extensive experiments on the widely used New York City Taxi (NYCTaxi) and New York City Bike (NYCBike) datasets demonstrate the MS-MSTformer’s capability to provide accurate multi-step citywide traffic flow forecasting. Specifically, the proposed model outperforms the baseline models in 11 out of 12 evaluation scenarios. On average, MS-MSTformer improves Root Mean Square Error (RMSE) by 31.13% and Mean Absolute Error (MAE) by 29.22% over the deep learning baselines. In addition, the ablation study demonstrates the contributions of both PTCA and RTCA to the proposed MS-MSTformer.}, archive = {J_EAAI}, author = {Shenkai Zhang and Taiyong Li and Yingqi Chen and Jiang Wu and Xiao Yan}, doi = {10.1016/j.engappai.2025.111362}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111362}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-step citywide traffic flow forecasting based on multiscale spatio-temporal transformer}, volume = {158}, year = {2025}, }
• (2025). Intelligent ensemble architecture for capturing transient nitrogen oxides emission spikes in real-world driving conditions. EAAI, 158, 111359. (www)

  @article{ , abstract = {Accurate prediction and monitoring of tailpipe nitrogen oxides (NOx) emissions is crucial for the design of effective automotive emission control strategies and for air quality in urban climates. This paper introduces a deep-learning soft sensing methodology designed to estimate tailpipe NOx emissions with high accuracy. The proposed method can capture both fast and slow transient behaviors of emissions, leading to accurate predictions of high spikes and smaller fluctuations. The central aspect of the proposed methodology is a temporal convolutional network (TCN) based classifier that differentiates between high and low NOx values. This differentiation allows the method to select the appropriate prediction model, specifically a TCN-based multi-head self-attention neural network. A hybrid methodology combining physics-based and data-driven approaches is employed for input feature selection, enhancing model robustness. The proposed method has been validated using an extensive dataset ( > 2300km) of field data from the City of Edmonton, Canada transit buses. Comparative analysis with other prediction models, including deep neural networks (DNN), long-short-term memory networks (LSTM), TCN, and state-of-the-art TCNs with multi-head self-attention, demonstrate the advantage of the proposed approach. Notable improvements in capturing both high and low NOx transient dynamics compared to a DNN model are achieved. Specifically, the proposed model achieves a 78% improvement in mean absolute error (MAE) for high NOx values (from 21.77 to 4.67 [milligrams per second (mg/s)]) and a 59% improvement (from 0.72 to 0.29 [mg/s]) for low NOx values compared to the DNN models. For mean square error (MSE), the improvements are 89% (from 521.85 to 54.39 [mg/s] 2 ) for high NOx values and 33% (from 1.47 to 0.98 [mg/s] 2 ) for low NOx values compared to those in the DNN models. In terms of root mean square error (RMSE), the proposed method shows a 67% (from 22.85 to 7.37 [mg/s]) improvement for high NOx values and a 32% (from 1.31 to 0.89[mg/s]) improvement for low NOx values compared to DNN models. Finally, a comparative analysis was conducted against other state-of-the-art predictive models, including LSTM and TCN, and the results show that the proposed method consistently outperforms these models. This superiority was validated through Analysis of Variance (ANOVA) and Tukey’s Honestly Significant Difference (HSD) test.}, archive = {J_EAAI}, author = {Amirreza Yasami and Mohamadali Tofigh and Bahram Bahri and Charles Robert Koch and Mahdi Shahbakhti}, doi = {10.1016/j.engappai.2025.111359}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111359}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent ensemble architecture for capturing transient nitrogen oxides emission spikes in real-world driving conditions}, volume = {158}, year = {2025}, }
• (2025). A convolutional neural network-enhanced attack detection framework with explainable artificial intelligence for internet of things-based metaverse security. EAAI, 158, 111358. (www)

  @article{ , abstract = {Metaverse integration is deeply reliant on the Internet of Things (IoT), which enables seamless data exchange between the physical and virtual worlds. By connecting sensors, wearables, and smart devices to Metaverse environments, IoT enhances interactions and immersion, essential for a fully integrated, data-driven Metaverse. However, IoT devices, due to their simple hardware and Internet access, are vulnerable to cyberattacks, making security a critical challenge for ensuring a safe infrastructure. Traditional methods struggle to combat evolving threats, requiring adaptive artificial intelligence (AI) powered solutions. This work presents a hybrid framework combining convolutional neural networks (CNN) with machine learning (ML) classifiers, categorical boosting (CatBoost), and adaptive boosting (AdaBoost), enhanced by metaheuristic optimization for better performance. A two-tier architecture is proposed for handling complex data to detect and identify attacks within IoT networks. A comparative analysis of a real-world IoT dataset demonstrates the framework’s effectiveness in detecting threats (binary classification) and identifying malicious attacks (multi-class classification). A total of six experiments are conducted using publicly available data with the best models achieving .993 accuracy for binary classifications and .877 for multi-class. Explainable AI techniques further provided insights into model decisions, guiding future data collection and system security.}, archive = {J_EAAI}, author = {Aleksandar Jokic and Miodrag Zivkovic and Luka Jovanovic and Milos Mravik and Marko Sarac and Vladimir Simic and Muhammad Attique Khan and Nebojsa Bacanin}, doi = {10.1016/j.engappai.2025.111358}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111358}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A convolutional neural network-enhanced attack detection framework with explainable artificial intelligence for internet of things-based metaverse security}, volume = {158}, year = {2025}, }
• (2025). Machine learning methods comparison for maritime wireless signal strength prediction. EAAI, 158, 111357. (www)

  @article{ , abstract = {The 5th generation mobile networks (5G) signal strength prediction techniques based on artificial intelligence (AI) demonstrate numerous advantages, such as easy implementation and high accuracy. However, what kind of machine learning model can provide superior performance in maritime applications is still unknown. In this paper, we developed twelve predictive models, encompassing eight machine learning and four linear regression approaches, for estimating reference signal receiving power (RSRP) and received signal strength indicator (RSSI) in maritime land-to-ship (L2S) scenarios. Our models are rigorously evaluated using nine metrics, trained on 5G band data collected with our self-designed equipment, and refined with carefully selected features to maximize prediction accuracy. The results show that the machine learning models are generally superior to the linear regression models in the fitting and prediction of RSRP and RSSI according to our evaluation and prediction experiments. Especially, rule-based models like decision tree regression (DTR) can accurately learn the impact of both large-scale and small-scale fading on the prediction objects from the data, and also have strong model interpretability. Our work has provided good reference value for future machine learning-based channel modeling and model evaluations.}, archive = {J_EAAI}, author = {Lisha Peng and Kun Yang and Jianming Wu and Chengyuan Wen and Tong Peng and Xin Li}, doi = {10.1016/j.engappai.2025.111357}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111357}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning methods comparison for maritime wireless signal strength prediction}, volume = {158}, year = {2025}, }
• (2025). A flexible job shop scheduling method based on heterogeneous disjunctive graph and deep reinforcement learning. EAAI, 158, 111356. (www)

  @article{ , abstract = {Automated Guided Vehicles (AGVs) have been widely applied in discrete manufacturing systems, but their introduction has increased the uncertainty and complexity of the scheduling process. In a flexible job-shop environment, achieving efficient integrated scheduling of machines and AGVs is of significant research value for enhancing production efficiency. However, existing mathematical models typically incorporate transportation time into processing time or assume it to be zero, neglecting the dynamic interaction characteristics between AGVs and machines, which limits the optimization effect of the scheduling scheme. This paper addresses the flexible job-shop scheduling problem with limited AGVs (FJSP-LA), aiming to minimize the makespan, and proposes an end-to-end optimization method based on deep reinforcement learning to achieve the coordinated optimization of processing and transportation resources. Firstly, a Heterogeneous Disjunctive Graph model is constructed, which includes three types of nodes: operations, machines, and AGVs. By integrating the information of operation nodes, machine nodes, and AGV nodes, efficient trilateral cooperative scheduling is achieved while reducing the model complexity. Secondly, a hierarchical feature extraction framework based on multi-layer perceptrons (MLPs) is designed to effectively capture the node attribute features and their dynamic constraint relationships. On this basis, a decision-making model based on the Actor–Critic architecture is constructed, and the Proximal Policy Optimization (PPO) algorithm is used for training to learn the optimal production scheduling strategy. To verify the effectiveness of the algorithm, comparative experiments were conducted on two public datasets with seven advanced scheduling algorithms. The numerical experimental results show that this method not only demonstrates good generalization ability on unknown datasets but also outperforms existing mainstream algorithms in terms of computational efficiency and scheduling performance, proving its application potential in complex flexible job-shop scheduling problems.}, archive = {J_EAAI}, author = {Xuhui Zhao and Haokai Qu and Mingchuan Zhang and Jiamei Feng and Lin Wang and Qingtao Wu}, doi = {10.1016/j.engappai.2025.111356}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111356}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A flexible job shop scheduling method based on heterogeneous disjunctive graph and deep reinforcement learning}, volume = {158}, year = {2025}, }
• (2025). A morphological information-guided detection network for various morphological objects in remote sensing imagery. EAAI, 158, 111351. (www)

  @article{ , abstract = {Optical imagery and synthetic aperture radar (SAR) imagery are critical data sources for remote sensing object detection, offering diverse feature information. However, the complex background clutter, including noise and pseudo-objects, in heterogeneous remote sensing images poses significant challenges. Existing computer vision methods rely excessively on broad-area modeling, which tends to incorporate excessive background clutter, exacerbating these issues. To address these challenges, we propose a morphology information-guided detection network (MIGNet) for heterogeneous remote sensing imagery detection. MIGNet models the long-range dependencies between pixels by utilizing cross-selective scanning to capture the spatial and morphological features of the objects. These features guide the convolutional kernel offsets, effectively mitigating the inclusion of excessive background noise during feature extraction. To further mitigate clutter interference, MIGNet’s architecture incorporates frequency-domain analysis to suppress low-frequency background noise while enhancing high-frequency object contours and details. However, while enhancing the model’s ability to distinguish between objects and backgrounds, these methods amplify the classification signals of pseudo-objects, leading to ambiguity in the representation of true and pseudo-objects. To overcome this limitation, MIGNet’s label assignment jointly classify and regress scores to determine the probability and importance of an anchor being assigned as positive or negative. Anchors are assigned higher positive weights only when both scores are high, effectively reducing interference from pseudo-objects. Experimental demonstrate that MIGNet significantly outperforms state-of-the-art methods, achieving a 1.5% improvement in average precision 0 . 5 ( A P 0 . 5 ) on the detection in optical remote (DIOR) dataset and a 1.9% increase in mean average precision (mAP) on the SAR ship detection dataset (SSDD).}, archive = {J_EAAI}, author = {Xiao Huang and Jiwei Hu and Quan Liu and Guangpeng Zhao and Qiwen Jin}, doi = {10.1016/j.engappai.2025.111351}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111351}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A morphological information-guided detection network for various morphological objects in remote sensing imagery}, volume = {158}, year = {2025}, }
• (2025). Differential out-of-distribution error guidance for unknown bearing fault diagnosis with conditional diffusion model. EAAI, 158, 111350. (www)

  @article{ , abstract = {In the modern machinery industry, bearing fault diagnosis is crucial for ensuring operational efficiency and reliability. However, methods based on closed-set assumptions face challenges in addressing the issue of unknown faults. In particular, out-of-distribution (OOD) detection is susceptible to interference from the overlap between training data distribution and the features of unknown samples, resulting in degraded diagnostic performance. To overcome this limitation, we propose a novel bearing fault diagnosis method that designs a conditional diffusion model to detect bearing faults by exploiting differential OOD error. This approach enhances the identification of unknown fault types. Our method involves training an encoder on in-distribution (ID) samples to extract feature representations and logits. These extracted features are then used to guide the diffusion model in sample reconstruction by amplifying the reconstruction error of OOD samples. Additionally, we incorporate a weighted scoring mechanism that combines the prediction confidence of a pre-trained classifier with the evaluation of the reconstructed data. We validated the proposed method through experiments on two rotating machinery datasets. The results demonstrate that our approach effectively diagnoses both known and unknown bearing faults with an accuracy of 94.3%. Under cross-domain conditions, the performance degradation is kept at 2.8%, which has good generalization. Based on the accurate classification of known faults, a robust solution is provided for the reliable identification of unknown faults including single and compound faults.}, archive = {J_EAAI}, author = {Yuting Zhang and Boyuan Yang and Zixu Zhao}, doi = {10.1016/j.engappai.2025.111350}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111350}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Differential out-of-distribution error guidance for unknown bearing fault diagnosis with conditional diffusion model}, volume = {158}, year = {2025}, }
• (2025). Koopman-based predictive tracking control. EAAI, 158, 111349. (www)

  @article{ , abstract = {Constraint handling during tracking operations is at the core of many real-world control implementations and is well understood when dynamic models of the underlying system exist, yet becomes more challenging when data-driven models are used to describe the nonlinear system at hand. We seek to combine the nonlinear modeling capabilities of a wide class of neural networks with the constraint-handling guarantees of model predictive control in a rigorous and online computationally tractable framework. The class of networks considered can be captured using Koopman operators, and are integrated into a Koopman-based predictive tracking control (KPTC) for nonlinear systems to track piecewise constant references. The effect of model mismatch between original nonlinear dynamics and its trained Koopman linear model is handled by using a constraint-tightening approach in the proposed KPTC controller. By choosing two Lyapunov functions, we prove that the solution is recursively feasible and input-to-state stable to a neighborhood of both online and offline optimal reachable steady outputs in the presence of bounded modeling errors under certain assumptions. The proposed approach has the advantage relative to existing model-based tracking approaches of enabling data-driven models to be utilized with explicit guarantees, while using efficient quadratic program solvers in online implementations. We demonstrate the proposed approach initially in simulations, and then experimentally to the problem of reference tracking by an autonomous ground vehicle.}, archive = {J_EAAI}, author = {Ye Wang and Yujia Yang and Ye Pu and Chris Manzie}, doi = {10.1016/j.engappai.2025.111349}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111349}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Koopman-based predictive tracking control}, volume = {158}, year = {2025}, }
• (2025). A deep learning approach for image time series forecasting: Study case, united states drought monitor. EAAI, 158, 111346. (www)

  @article{ , abstract = {Image time series (ITS) are a chronologically ordered images set which contains spatial and temporal patterns. The United States drought monitor provides a map collection of drought severity spatial distribution across the regions which changes over the time. This work aims to use the drought map ITS to extract inner spatiotemporal features patterns and forecast the spatial drought severity distribution classes for a future horizon that ranges from one to twelve weekly time steps by using a convolutional long short-term memory network (ConvLSTM). This approach offers a new perspective by using a set of images (ITS) as input for a deep learning model to predict the spatial drought in an image that represents the next time step with the drought distribution. The design also allow us to implement a recursive multi-step forecasting strategy to generate an horizon up to twelve ( h ) weekly drought maps. The obtained results shows that the proposed approach achieves an overall of macro F1-score metric of 0.9953 in generating the next drought map t + 1 and 0.6965 in generating the up to 12 t h drought map. ConvLSTM is more accurate in general when it is compared with convolutional neural networks, video visual transformers networks and the naïve baseline model. These findings demonstrate the approach efficacy in identifying spatiotemporal features for reliable drought map forecasts, providing a new valuable tool for drought prediction as a visual representation.}, archive = {J_EAAI}, author = {Manuel Medrano-Diaz and Hector Rodriguez-Rangel and Vicenç Puig-Cayuela and Juan J. Flores and Rodrigo Lopez-Farias and Carlos Lara-Alvarez}, doi = {10.1016/j.engappai.2025.111346}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111346}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep learning approach for image time series forecasting: Study case, united states drought monitor}, volume = {158}, year = {2025}, }
• (2025). Scene text image super-resolution with semantic-aware interaction. EAAI, 158, 111345. (www)

  @article{ , abstract = {Scene text image super-resolution aims to enhance the resolution of images containing text in various scenes, which amplifies the prominence of the text and improves its recognizability. Existing methods struggle to accurately localize text regions in high-noise environments, which hampers their ability to effectively implement targeted super-resolution. To address these issues, we propose a Text image Super-resolution Semantic-aware Interaction Network (TSSIN) by embedding a text region segmentation network. First, we propose to use a pre-trained text region segmentation network (TRSN) to extract text region information. This approach semantically guides our model to address the challenge of the model not performing targeted super-resolution processing of text in high-noise environments. Secondly, we propose a multi-modal semantic information interaction module (MSIIM) to mitigate the issue of insufficient global information exchange. Comprehensive experiments conducted on the TextZoom dataset demonstrate that our TSSIN significantly enhances image quality. Furthermore, it shows a clear superiority over state-of-the-art methods on TextZoom, achieving an average text recognition accuracy improvement of +1.0% over Transformer-Based Super-Resolution Network (TBSRN) (49.6%, 56.2%, 60.1% vs. 48.1%, 55.9%, 58.9%), +0.83% over Parallelly Contextual Attention Network (PCAN), and +1.46% over Text Prior Guided Super-Resolution (TPGSR), as evaluated by three pre-trained text recognition models. Code is available at https://github.com/ads2d/TSSIN .}, archive = {J_EAAI}, author = {Mingle Zhou and Wenlong Liu and Jin Wan and Delong Han and Min Li and Gang Li}, doi = {10.1016/j.engappai.2025.111345}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111345}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Scene text image super-resolution with semantic-aware interaction}, volume = {158}, year = {2025}, }
• (2025). 2-tuple linguistic complex intuitionistic fuzzy aggregation operators and their application in decision making problem. EAAI, 158, 111342. (www)

  @article{ , abstract = {The multiple attribute group decision making (MAGDM) issues with 2-tuple linguistic complex intuitionistic fuzzy 2 T L C I F aggregation operators are examined in this work. Next, we design several 2-tuple linguistic complex intuitionistic fuzzy aggregation operators (2TLCIFAOs) using geometric and arithmetic procedures. These suggested operators’ salient features are examined. We have then developed some methods to tackle the 2-tuple linguistic complex intuitionistic fuzzy MAGDM issues using these operators. A real-world example is given to validate the developed approach and demonstrate its overall utility and efficacy. These specific fuzzy models give solutions to the almost all of the challenges in decision making, especially 2 and 3 dimensional process. Proposed aggregation operators better to use and have accurate results. We also compare our proposed method with other existing methods. Finally, we write the conclusion of the work.}, archive = {J_EAAI}, author = {Sumaira Yasmin and Muhammad Qiyas and Darjan Karabasevic and Neelam Khan}, doi = {10.1016/j.engappai.2025.111342}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111342}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {2-tuple linguistic complex intuitionistic fuzzy aggregation operators and their application in decision making problem}, volume = {158}, year = {2025}, }
• (2025). An efficient and noise-robust framework for high-dimensional tuning in big data analytics. EAAI, 158, 111332. (www)

  @article{ , abstract = {Spark has been a widely used platform for cluster-based big data analytics and large-scale data processing. Although the performance of Spark can be optimized by tuning parameters, its high-dimensional parameter space complicates this process and motivates the development of auto-tuning research. However, despite their success in reducing the parameter space, existing approaches require an exhaustive search to determine the optimal reduced or target dimensionality, which is time-consuming. Additionally, they overlook the impact of performance noise in Spark bottlenecks, resulting in unreliable optimal configurations. These drawbacks highlight the need for a more efficient and noise-resilient tuning strategy. To address these challenges, we propose a resource-efficient and reliable configuration tuning framework for Spark that leverages subspace-based Bayesian optimization and a noise-robust acquisition function. The proposed framework optimizes the Spark configuration by effectively reducing high-dimensional parameters without requiring a time-consuming and resource-intensive determination of the target dimensionality. Moreover, it an optimal configuration with reliable performance by minimizing the effects of noise. We validate our framework using eight HiBench workloads, comparing it to state-of-the-art baselines. Experimental results show that our framework achieves up to 14.13% faster task completion time. We provide open-source code available at: https://github.com/JIEUN-12/NoRTune .}, archive = {J_EAAI}, author = {Jieun Lee and Sangmin Seo and Chanho Yeom and Huijun Jin and Sein Kwon and Sanghyun Park}, doi = {10.1016/j.engappai.2025.111332}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111332}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient and noise-robust framework for high-dimensional tuning in big data analytics}, volume = {158}, year = {2025}, }
• (2025). A group decision-making framework for public service delivery modes using Dempster–Shafer theory. EAAI, 158, 111327. (www)

  @article{ , abstract = {Public service delivery is a challenge and the goal of every government is to maximize benefits and minimize expenditures. Because of the variety and cross-cutting nature of public service delivery modes and the criteria affecting the choice of mode, it is desirable to design a rational approach to choosing them under various scenarios. In this study, we constructed a multi-criteria group decision-making framework based on public service delivery. We constructed an evaluation index system containing eight criteria by considering the drivers of the public service delivery mode. We used agglomerative hierarchical clustering to determine the similarity of multiple sources of evidence, and evidence fusion was divided into two steps: intra-cluster fusion and inter-cluster fusion. Moreover, the criterion weights were obtained by combining both subjective and objective weights. The final step of the framework employed the technique for order preference by similarity to ideal solution (TOPSIS) to identify the optimal mode. Finally, the feasibility of this decision-making framework in the decision-making of public service delivery modes was verified using an empirical analysis of old community renovation.}, archive = {J_EAAI}, author = {Shaoshuai Shang and Liguo Fei and Yuan Liu and Luning Liu}, doi = {10.1016/j.engappai.2025.111327}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111327}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A group decision-making framework for public service delivery modes using Dempster–Shafer theory}, volume = {158}, year = {2025}, }
• (2025). Mamba model guided deep visual-inertial odometry. EAAI, 158, 111326. (www)

  @article{ , abstract = {Visual-inertial odometry (VIO) aims to predict trajectories by fitting data from cameras and inertial measurement unit (IMU). In recent years, deep learning-based VIO has made significant progress. However, several challenges persist, such as the loss of key visual features during sudden linear acceleration and the efficient fusion of visual and inertial information. To address these challenges, we propose a Mamba model guided deep visual-inertial odometry algorithm, named MamVIO. Specifically, we carefully designed a module for extracting visual features, which dynamically adjusts spatiotemporal receptive field to capture the dynamic variations of key features along the temporal dimension and aggregates information from adjacent frames. Furthermore, to effectively combine visual and inertia information, we design a Mamba model based visual-inertial fusion module, which extends the Mamba model to support dual input. Extensive evaluations on the public odometry benchmark demonstrate that our algorithm achieves competitive performance, increasing average translational accuracy by 9.56% and average rotational accuracy by 8.55%, respectively, compared to baseline method.}, archive = {J_EAAI}, author = {Jiwei Liu and Qingchun Zheng and Wenpeng Ma and Peihao Zhu and Yantao Zong}, doi = {10.1016/j.engappai.2025.111326}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111326}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mamba model guided deep visual-inertial odometry}, volume = {158}, year = {2025}, }
• (2025). Addressing false alarms from high-voltage structures in subpixel fire detection. EAAI, 158, 111324. (www)

  @article{ , abstract = {False alarms in subpixel fire detection often arise when high-voltage structures, such as powerlines or towers near thermographic cameras, emit intense infrared radiation that mimics early fire signals at long distances. This paper proposes the study and statistical analysis of You Only Look Once version 8 (YOLOv8) to detect, segment, and isolate these sources of false alarms. YOLOv8 is trained on the Addressing False Alarm Situations (AFAS) dataset, which includes a variety of Long-Wave Infrared (LWIR) and Near-Infrared (NIR) imagery from both aerial and ground-level perspectives. The model achieves a mean Average Precision (mAP) of 0.784 at an Intersection over Union (IoU) threshold of 0.5. The contribution of this work lies in a detailed statistical analysis of YOLOv8 outputs, introducing, among others, the Empirical Cumulative Distribution Function (ECDF) as a metric to assess the relationship between mask overlap and detection confidence. To evaluate the model’s robustness under thermal disturbances, synthetic fires are introduced to simulate changes in the scene. The two-sample Kolmogorov–Smirnov (KS) test compares prediction distributions with and without these anomalies, important to ensure that the model performs reliably over a wide range of scenarios so that the presence of these structures can always be determined and isolated. Finally, an energy retention metric is introduced to quantify the probability that the model’s predicted masks obscure at least half of an early fire’s energy. In critical cases where the fire appears at 2, 3, and 4 pixels from the segmented structures, these probabilities are approximately 7%, 4%, and 3%, respectively.}, archive = {J_EAAI}, author = {Antonio Galván-Hernández and Víctor Araña-Pulido and Francisco Cabrera-Almeida and Pedro Quintana-Morales}, doi = {10.1016/j.engappai.2025.111324}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111324}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Addressing false alarms from high-voltage structures in subpixel fire detection}, volume = {158}, year = {2025}, }
• (2025). Graph diffusion network for multi-agent reinforcement learning in drone swarm exploration. EAAI, 158, 111322. (www)

  @article{ , abstract = {Drone swarm exploration has wide applications in rescue operations and engineering surveying. A drone swarm is a multi-agent system, and applying multi-agent reinforcement learning to such a system is an attractive topic in the field of robotics. In this paper, we propose a multi-agent reinforcement learning model that can chain-aggregate information from agents and apply it to the drone swarm via the Robot Operating System (ROS2). This model not only helps agents aggregate information with their neighbors but also enables the swarm to establish an organized structure, facilitating better cooperation and improving overall swarm performance. The model performs well in multi-drone exploration tasks, even in the presence of instability within the swarm. Experimental results demonstrate that the model enables effective cooperation among drones and achieves better global performance. Furthermore, we implemented the strategy based on our model on a physical platform to realize drone swarm exploration tasks. Although the cameras mounted on the drones have limited resolution, the swarm’s numerical advantage allows for high-quality exploration images, and the system outperforms other methods in terms of exploration efficiency and real-time data performance.}, archive = {J_EAAI}, author = {Zhiling Jiang and Chenyang Zhang and Zhan Shi and Guanghua Song}, doi = {10.1016/j.engappai.2025.111322}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111322}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph diffusion network for multi-agent reinforcement learning in drone swarm exploration}, volume = {158}, year = {2025}, }
• (2025). Lightweight binary convolutional-transformers fusion network for facial expression recognition. EAAI, 158, 111315. (www)

  @article{ , abstract = {Deep learning-based methods exploit local facial regions for facial expression recognition (FER) but overlook long-range dependencies of facial muscle movements. Moreover, deploying existing large-scale deep networks on mobile devices remains a huge challenge. To this end, this paper designs a mini-binary transformer (MiniBTR) for FER. Specifically, we first construct image patches by regions of interest (ROIs) which are associated with facial action units (AUs) and represent these patches by the histogram of oriented gradient (HOG) feature. Then, by integrating the complementary advantages of convolutional layers (Convs) and multi-head self-attention mechanism (MSA), an effective and efficient mini-transformer (MiniTR) architecture with 69K parameters is proposed to follow the human–machine collaborative strategy. Finally, we design the binary model MiniBTR by extending MiniTR with a binary operation. Extensive experimental results on four publicly available datasets demonstrate that the MiniBTR yields comparable results in terms of recognition accuracy, model size and inference speed while offering a more deployable yet high-performing alternative.}, archive = {J_EAAI}, author = {Xing Jin and Xiyin Wu and Libo Weng and Qiaolin Ye}, doi = {10.1016/j.engappai.2025.111315}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111315}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight binary convolutional-transformers fusion network for facial expression recognition}, volume = {158}, year = {2025}, }
• (2025). Enhanced facial image essence transfer via semantic guidance. EAAI, 158, 111314. (www)

  @article{ , abstract = {Generating novel facial images guided by given references (e.g., face style transfer) has attracted significant attention in recent years. However, existing approaches often struggle to achieve a balance between transfer quality and identity preservation. In this paper, we propose a novel method for reference-guided facial image essence transfer, which aims to effectively extract and transfer high-level semantic features (“essence”) from a single reference image to a source image without compromising identity preservation. Specifically, we leverage Contrastive Language-Image Pre-training (CLIP) to address this task from an image-driven perspective, exploiting CLIP’s rich understanding of diverse visual concepts. We introduce a novel essence contrastive loss designed to utilize semantic information captured from different image perspectives within the CLIP embedding space, which effectively guides the essence transfer in a semantically meaningful direction. Furthermore, to enhance feature fusion and ensure precise attribute transfer, we propose an Essence-Preserved Modulation (EPM) module that explicitly maps the CLIP embeddings of the reference image into the latent space of the generative model. Additionally, an essence degree indicator is introduced to allow flexible control over the extent of essence transferred to the generated images, significantly improving the practical applicability of our method. Extensive experiments demonstrate that our approach consistently generates realistic facial images that effectively capture diverse essence attributes from various references while simultaneously achieving superior facial identity preservation compared to existing methods.}, archive = {J_EAAI}, author = {Ailin Li and Anyong Qin and Ying Huang}, doi = {10.1016/j.engappai.2025.111314}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111314}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced facial image essence transfer via semantic guidance}, volume = {158}, year = {2025}, }
• (2025). Sequential state estimation based target tracking algorithm for unmanned underwater vehicle target grasping. EAAI, 158, 111313. (www)

  @article{ , abstract = {To address the challenges of underwater target tracking in dynamic and turbid environments, this paper proposes a Sequential State Estimation-based Tracking (SSET) algorithm that integrates temporal and spatial information for robust performance. The SSET algorithm comprises three key components: a Kalman filter-based sequential state estimation module to predict target motion and establish temporal correlations, a score head module to evaluate template reliability and optimize predictions, and a mixed-sequential-state transformer (MSST) to fuse triplet features for spatio-temporal correlation. Evaluations on open-air and underwater benchmarks demonstrate SSET’s superiority: it achieves 91.7% precision and 69.8% success rate on terrestrial datasets, outperforming state-of-the-art methods by 0.6% and 0.2%, respectively. In underwater scenarios, SSET attains 56.5% precision and 55.8% success rate, with improvements in occlusion and low-resolution conditions. Underwater grasping experiments further validate its practicality, achieving a high success rate in controlled environments.}, archive = {J_EAAI}, author = {Yanli Li and Weidong Liu and Wenbo Zhang and Le Li}, doi = {10.1016/j.engappai.2025.111313}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111313}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sequential state estimation based target tracking algorithm for unmanned underwater vehicle target grasping}, volume = {158}, year = {2025}, }
• (2025). Health- and behavior-aware energy management strategy for fuel cell hybrid electric vehicles based on parallel deep deterministic policy gradient learning. EAAI, 158, 111311. (www)

  @article{ , abstract = {To find a more optimal way to solve the energy management strategy (EMS) of fuel cell hybrid electric vehicles (FCHEVs), the majority of existing research focuses on external driving conditions, while the driver’s behavior as a more important internal influence factor also needs to be taken into account. In this paper, a health- and behavior-aware two-layer hierarchical energy management framework using an improved adaptive parallel deep deterministic policy gradient (DDPG) learning algorithm is proposed for obtaining the optimal EMS of a multi-source FCHEV. In the upper layer, machine learning approaches are employed to recognize the real-time driver’s behavior, and Pontryagin’s minimum principle is applied to calculate the optimal equivalent factor of each driver’s behavior. In the lower layer, to protect the service life of fuel cell and battery as well as increase the learning efficiency, an adaptive fuzzy filter is used, and a health- and behavior-aware multi-objective adaptive equivalent consumption minimization strategy model is constructed and solved by an improved adaptive parallel DDPG-based algorithm. Simulation results show that, the EMS obtained by the proposed DDPG algorithm can achieve the highest fuel cell (FC) working efficiency (approximate to 56%), apparently reduce the degree of degradation of battery (BAT) from 0.42% to 0.28%, and achieve a reduction of 9.24% in terms of the total cost to use compared with deep Q network (DQN)-based EMS.}, archive = {J_EAAI}, author = {Haochen Sun and Jing Li and Chun Cheng and Suzhen Shi and Jing Wang and Jingjing Lin and Yang Liu}, doi = {10.1016/j.engappai.2025.111311}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111311}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Health- and behavior-aware energy management strategy for fuel cell hybrid electric vehicles based on parallel deep deterministic policy gradient learning}, volume = {158}, year = {2025}, }
• (2025). A real-time anomaly detection method for robots based on a flexible and sparse latent space. EAAI, 158, 111310. (www)

  @article{ , abstract = {The growing demand for robots to operate effectively in diverse environments necessitates the need for robust real-time anomaly detection techniques during robotic operations. However, deep learning-based models in robotics face significant challenges due to limited training data and highly noisy signal features. In this paper, we present Sparse Masked Autoregressive Flow-based Adversarial AutoEncoder model to address these problems. This approach integrates Masked Autoregressive Flow model into Adversarial AutoEncoders to construct a flexible latent space and utilizes Sparse autoencoder to efficiently focus on important features, even in scenarios with limited feature space. Our experiments demonstrate that the proposed model achieves a 4.96% to 9.75% higher area under the receiver operating characteristic curve for pick-and-place robotic operations with randomly placed cans, compared to existing state-of-the-art methods. Notably, it showed up to 19.67% better performance in scenarios involving collisions with lightweight objects. Additionally, unlike the existing state-of-the-art model, our model performs inferences within 1 millisecond, ensuring real-time anomaly detection. These capabilities make our model highly applicable to machine learning-based robotic safety systems in dynamic environments. The code is available at https://github.com/twkang43/sparse-maf-aae .}, archive = {J_EAAI}, author = {Taewook Kang and Bum-Jae You and Juyoun Park and Yisoo Lee}, doi = {10.1016/j.engappai.2025.111310}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111310}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A real-time anomaly detection method for robots based on a flexible and sparse latent space}, volume = {158}, year = {2025}, }
• (2025). MemMambaAD: Memory-augmented state space model for multivariate time series anomaly detection. EAAI, 158, 111308. (www)

  @article{ , abstract = {Multivariate time series anomaly detection focuses on recognizing abnormal patterns to reduce system failures and improve production efficiency and product quality. Accurately detecting anomalies in data remains challenging because existing reconstruction-based methods are prone to overfitting. Recently, reconstruction methods guided by memory modules have been used to address this issue. However, these methods still suffer from insufficient feature extraction of time series prototypes and inadequate storage of normal sample prototype patterns in memory modules. To address these issues, we propose a memory-augmented state space model for multivariate time series anomaly detection. Specifically, we introduce a sequence decomposition state space model-temporal convolutional encoder, which independently extracts trend and seasonal features of multivariate time series in global and local manner, capturing the intrinsic patterns of time series more comprehensively. In addition, we propose a dynamic memory update mechanism, which flexibly updates the memory item through the memory selection mechanism, to more accurately record the prototype pattern of normal samples to improve anomaly detection performance. Extensive experiments on five benchmark datasets demonstrate that our method achieves state-of-the-art performance and reduces memory usage compared with other methods.}, archive = {J_EAAI}, author = {Gang Li and Mingchao Ge and Jin Wan and Delong Han and Min Li and Mingle Zhou}, doi = {10.1016/j.engappai.2025.111308}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111308}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MemMambaAD: Memory-augmented state space model for multivariate time series anomaly detection}, volume = {158}, year = {2025}, }
• (2025). Enhancing primitive segmentation through transformer-based cross-task interaction. EAAI, 158, 111307. (www)

  @article{ , abstract = {Point cloud primitive segmentation, which divides a point cloud into surface patches of distinct primitive types, is fundamental to three-dimensional objects processing and recognition. However, existing deep learning methods for primitive segmentation lack the capability to capture global spatial relationships across tasks, limiting the exploitation of inter-task consistency. To address this issue, we propose a novel transformer-based cross-task interaction primitive segmentation (TCIPS) method that models global spatial relationships between all tasks, leading to improved segmentation accuracy. Specifically, TCIPS leverages center offset and regional purity prediction as auxiliary tasks, providing supplementary supervision to facilitate the learning of spatial and boundary information, thereby promoting richer and more generalized feature learning. Furthermore, we design a cross-task transformer fusion module that fuses and refines features from task-specific decoders using two types of transformer blocks: the feature fusion block and the task query block. Extensive experiments and comparisons with state-of-the-art methods demonstrate the effectiveness and robustness of our approach. Codes and models are publicly available at https://github.com/MingFengHill/TCIPS .}, archive = {J_EAAI}, author = {Tao Wang and Weibin Xi and Yong Cheng and Jun Zhang and Ruochen Yin and Yang Yang}, doi = {10.1016/j.engappai.2025.111307}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111307}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing primitive segmentation through transformer-based cross-task interaction}, volume = {158}, year = {2025}, }
• (2025). WrdaGAN: A text-to-image synthesis pipeline based on wavelet representation and adaptive sample domain constraint strategy. EAAI, 158, 111305. (www)

  @article{ , abstract = {Text-to-image translation capitalizes on multimodal learning to enable convenient and efficient conversion between natural language and imagery. However, with the popularity of this task, users desire higher quality outputs from less expensive devices via semantically consistent and stable networks. We propose a generative adversarial network based on Wavelet Representation and Adaptive Sample Domain Constraint strategy, named WrdaGAN, which utilizes wavelet representation and domain adaptation strategy. It employs wavelet transform to enhance high frequency details for improving semantic consistency. The domain adaptation reconstructs and constricts the sampling space to preclude pattern collapse. In the generative module, text is encoded by a pre-trained Contrastive Language-Image Pre-Training (CLIP). The Gaussian noise vector is concatenated with the encoded text and input to a high-frequency enhancement generator (HFE-G). Specifically, the generator contains UpBlock and high-frequency enhancement module. The enhancement module leverages wavelet transform theory. The Red Green Blue (RGB) operation transforms the feature map into the actual image. In the discriminative module, the Target-Aware Discriminator (TAW-D) extracts image features, primarily regulated by the adversarial loss. Adaptive sampling and constraints incorporate reconstruction and contraction to expedite adversarial learning. WrdaGAN has low hardware requirements and can be trained on a single Nvidia 3090 Graphics Processing Unit (GPU). A range of properties of the model are tested and illustrated on the Caltech-UCSD Birds-200-2011 (CUB Bird) and Microsoft Common Objects in Context (MS COCO) datasets. On the CUB Bird and COCO datasets, the Fréchet Inception Distance (FID) reaches 12.42 and 16.34, respectively. The codes are available at https://github.com/Guanyn/WrdaGAN.git .}, archive = {J_EAAI}, author = {Yongchao Qiao and Ya’nan Guan and Shujiao Liao and Wenyuan Yang and Weiping Ding and Lin Ouyang}, doi = {10.1016/j.engappai.2025.111305}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111305}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {WrdaGAN: A text-to-image synthesis pipeline based on wavelet representation and adaptive sample domain constraint strategy}, volume = {158}, year = {2025}, }
• (2025). Low-high frequency network for spatial–temporal traffic flow forecasting. EAAI, 158, 111304. (www)

  @article{ , abstract = {Traffic flow forecasting is crucial for smart city development. Existing methods primarily focus on spatial–temporal correlation learning but often overlook the distinct temporal characteristics of traffic flow. From the temporal perspective, traffic flow can be decomposed into low frequency components-representing periodic patterns inherent in the transportation system and high frequency components-reflecting short-term variations caused by external factors. However, current approaches tend to capture high frequency components while neglecting low frequency ones, which has resulted in a performance bottleneck. To address this problem, we propose a novel framework, named L ow- H igh F requency Net work(LHFNet), for traffic flow forecasting. Our framework comprises three key components: a low frequency encoder, a high frequency encoder, and a frequency feature fusion block. The low frequency encoder employs bi-level routing attention as its core module. To enhance the stability of low frequency representations, patch embedding and patch merging operations are integrated, and the benefit of this integration is that the model complexity can be reduced while enabling longer input series. For the high frequency encoder, multilayer perceptrons are used with residual connections as the primary structure. Finally, the frequency feature fusion block dynamically integrates both frequency features through a gated selection mechanism. Extensive comparative experiments have been conducted on four real-world datasets and the results have demonstrated that LHFNet outperforms state-of-the-art models.}, archive = {J_EAAI}, author = {Qi Feng and Bo Li and Xiaohan Liu and Xiaoguang Gao and Kaifang Wan}, doi = {10.1016/j.engappai.2025.111304}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111304}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Low-high frequency network for spatial–temporal traffic flow forecasting}, volume = {158}, year = {2025}, }
• (2025). Enhancing the absolute positioning accuracy of welding robots based on joint error compensation. EAAI, 158, 111302. (www)

  @article{ , abstract = {This paper presented a novel method for improving the absolute positioning accuracy of robot end-effectors through joint space error prediction, addressing the issue of low accuracy in welding robots. A sampling method that considered both Cartesian and joint spaces was proposed, and a joint error compensation model based on Gaussian process regression (GPR) was established, utilizing the expected joint positions and joint errors as input features. A compensation strategy that integrated the error model into the robot controller was introduced. Experiments involving a laser tracker for positioning error compensation, including single-point multi-pose, spatial multi-point, and continuous welding trajectory tests, were conducted. The experimental results demonstrated that the proposed compensation method effectively reduced the mean absolute error (MAE) of welding robot positioning from approximately 0.6 mm to within 0.25 mm. This method significantly enhanced the absolute positioning accuracy of welding robot end-effectors, meeting the requirements for offline programming in welding applications.}, archive = {J_EAAI}, author = {Bingqi Jia and Haihong Pan and Yukang Cai and Lei Zhang and Xuhong Chen and Lin Chen}, doi = {10.1016/j.engappai.2025.111302}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111302}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing the absolute positioning accuracy of welding robots based on joint error compensation}, volume = {158}, year = {2025}, }
• (2025). Generative adversarial network-based imputation and personalized federated learning with convolutional neural network-long short-term memory for air quality index prediction. EAAI, 158, 111300. (www)

  @article{ , abstract = {Air quality forecasting is a critical task for mitigating the adverse effects of pollution on public health and the environment. This research introduces an advanced Generative Adversarial Network based Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) Federated Framework (GC-LFF) that addresses the limitations of traditional centralized machine learning models, which struggle with data variability and privacy concerns. The proposed framework employs advanced convolutional neural networks and long short-term memory based architectures to effectively capture spatial and temporal dependencies in decentralized air quality data. Additionally, a Generative adversarial network-based imputation module is incorporated to manage missing or corrupted sensor data, ensuring completeness without necessitating centralized data sharing. Each participating city in GC-LFF trains its model locally using its air quality data and uses a weighted Federated Averaging technique to aggregate model updates on a central server. The final global model of each node is then customized to increase the accuracy of local forecasting. The GC-LFF framework was evaluated using individualized local data-driven fine-tuning across 26 metropolitan nodes. With a Mean square error (MSE) of 499.16, a Root Mean square error (RMSE) of 19.83, a Mean Absolute Error (MAE) of 14.41, and an R-squared ( R 2 ) of 0.72 after the removal of outliers. The model’s resilience and flexibility were validated using standard deviation analysis and confidence intervals. The results demonstrate its strong performance and scalability for practical implementation.}, archive = {J_EAAI}, author = {Shweta Roy and Nisha Kandhoul and Sanjay K. Dhurandher}, doi = {10.1016/j.engappai.2025.111300}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111300}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generative adversarial network-based imputation and personalized federated learning with convolutional neural network-long short-term memory for air quality index prediction}, volume = {158}, year = {2025}, }
• (2025). A feature selection method driven by fuzzy implication granularity. EAAI, 158, 111298. (www)

  @article{ , abstract = {The data in real life are often very complex, with different types and scales, and contain a large number of redundant features. How to perform feature selection for complex data is a tricky problem. To address the issue, this paper proposes a filter-based feature selection method driven by fuzzy implication granularity (FIGFS). Firstly, the fuzzy adaptive neighborhood radius is proposed to construct the information granules, and on this basis, a series of multi-granularity fuzzy implication information measures are established to characterize the feature uncertainty. Secondly, granular consistency is proposed to capture the correlation between features and decisions at the overall and local levels respectively. Then, a new multi-criteria feature evaluation metric is constructed by combining granularity consistency and multi-granularity fuzzy implication information measures. Finally, a general forward search feature selection algorithm compatible with low-dimensional data and high-dimensional data is designed. Compared with six state-of-the-art algorithms on 24 public datasets, the results show that our method is feasible and superior.}, archive = {J_EAAI}, author = {Shaowei Yan and Jin Qian and Ying Yu and Yongting Ni}, doi = {10.1016/j.engappai.2025.111298}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111298}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A feature selection method driven by fuzzy implication granularity}, volume = {158}, year = {2025}, }
• (2025). Integrating multiple syntactic structures for enhanced aspect-based sentiment analysis. EAAI, 158, 111297. (www)

  @article{ , abstract = {Aspect-based sentiment analysis is a fine-grained sentiment analysis task that focuses on determining the sentiment polarity of specific aspect words within a sentence. Traditional methods relying solely on dependency trees as syntactic structures have limitations in representing complex sentence structures and handling domain-specific data. To address the limitations, we propose Multiple Syntactic Relational Interaction Graph Convolutional Network (MSRI-GCN). This approach combines syntactic dependency trees with aspect-specific relational graph structures derived from Large Language Models (LLMs) like Chat Generative Pre-trained Transformer, preserving global syntactic information and enriching local aspect-specific information. Our method computes the number of paths between nodes using the dependency tree’s adjacency matrix and augments this with the syntactic relations from the reconstructed relational graph. Additionally, we utilize the multi-head attention mechanism to capture semantic relations. By learning the features of aspects and contexts through these dual relations, our model adaptively integrates both syntactic and semantic information. We conducted extensive experiments on five benchmark datasets demonstrate that our model significantly outperforms existing models.}, archive = {J_EAAI}, author = {Hongtao Liu and Kang Xiong and Sanqi Wu and Pengju Cao and Kefei Cheng and Xueyan Liu}, doi = {10.1016/j.engappai.2025.111297}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111297}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating multiple syntactic structures for enhanced aspect-based sentiment analysis}, volume = {158}, year = {2025}, }
• (2025). Visual sensing approach for high-precision pressure estimation in robotic manipulation. EAAI, 158, 111296. (www)

  @article{ , abstract = {Accurate sensing of contact pressure in robotic grippers is essential for both autonomous and teleoperated manipulation. However, the direct integration of tactile sensors onto gripper surfaces often poses significant challenges. With the advancement of high-resolution cameras and computer vision technologies, vision-based pressure estimation methods have emerged as viable alternatives. This paper presents a novel high-precision vision-based pressure estimation (HiVPE) method that addresses limitations in existing approaches, particularly when dealing with posture changes, low image quality, environmental variations, and occlusion. The key contributions of this work are in the realm of artificial intelligence (AI), where a novel dual-channel feature enhancement architecture is proposed, comprising a Spatial Feature Modulation (SFM) module for detailed manipulator-specific perception and a Visual Pressure Mamba (VP-Mamba) module based on state space models for capturing long-range dependencies and global structural information. In the engineering domain, the application of this AI technology involves implementing an eye-in-hand camera system to generate pressure maps from gripper images, enabling robust pressure estimation across diverse real-world scenarios without direct sensor integration. Experimental evaluations in both controlled laboratory settings and diverse real-world environments demonstrate excellent performance, with a temporal accuracy of 96.22% and 95.48% for tendon-actuated and soft grippers, respectively, while reducing the mean absolute error by 5.6% and 19.5% compared to baseline methods. These results validate the superior performance of HiVPE in complex operational tasks and unknown environments, making it a feasible solution for practical robotic applications.}, archive = {J_EAAI}, author = {Wensheng Wang and Xueli Liu and Linxin Bai and Xiaobo Yang and Xinrong Chen}, doi = {10.1016/j.engappai.2025.111296}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111296}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Visual sensing approach for high-precision pressure estimation in robotic manipulation}, volume = {158}, year = {2025}, }
• (2025). A novel reinforcement learning framework based on simplified graph transformer for large-scale fuzzy flexible job shop scheduling problem. EAAI, 158, 111295. (www)

  @article{ , abstract = {The Large-scale Flexible Job Shop Scheduling Problem (FJSP) is a classic NP-hard problem known for its complex solution space and formidable computational challenges. This paper investigates a more realistic variant of FJSP featuring fuzzy processing times, with the objective of minimizing job completion times. Addressing the inefficiencies of metaheuristic algorithms in solving large-scale FJSP instances, characterized by low efficiency and suboptimal heuristic rules, we propose a novel reinforcement learning (RL) framework called Soft Actor-Critic based simplified graph transformers (SGFormer-SAC) as a fundamental optimization approach for addressing this problem. Firstly, we introduce the use of Graph Transformer to capture the topological structure of the fuzzy disjunctive graph and the long-term dependencies among multiple operations. Secondly, We critically demonstrate that even with a single-layer, single-headed attention mechanism, remarkable competitive performance can be achieved in large-scale disjunctive graph models. This encourages a reconsideration of the design principles behind Transformers. Finally, we reduce the quadratic complexity of the Transformer to linear complexity, significantly improving the training and inference times for large-scale fuzzy FJSP instances. Experimental results demonstrate that our model exhibits strong robustness and generalization across both generated instances and public datasets after a single training session. In generalization experiments, our approach outperforms the optimal scheduling rule (MOPNR+SPT) with an average improvement of 31.23%. For large-scale instances, the inference time does not exceed 5 s, and our model achieves an 11.4% improvement over the state-of-the-art (SOTA) CARL framework on the Beh41-45 (100 × 60) benchmark. Furthermore, analysis of variance (ANOVA) tests confirm the significant effectiveness of our method.}, archive = {J_EAAI}, author = {Wenquan Zhang and Fei Zhao and Bo Feng and Xuesong Mei}, doi = {10.1016/j.engappai.2025.111295}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111295}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel reinforcement learning framework based on simplified graph transformer for large-scale fuzzy flexible job shop scheduling problem}, volume = {158}, year = {2025}, }
• (2025). Robust broad learning system with parametrized variational mode decomposition for schizophrenia diagnosis. EAAI, 158, 111294. (www)

  @article{ , abstract = {Schizophrenia (SZ) is a significant mental disorder characterized by various neurophysiological and cognitive impairments. Early diagnosis remains challenging due to its reliance on symptom detection. However, advance signal processing algorithm is combined with machine learning technique for early detection of schizophrenia using electroencephalogram (EEG) signals efficaciously. To optimize results from biomedical signals, effective feature extraction (FE) and feature engineering are essential. In this study, parametrized variational mode decomposition (PVMD) is applied to electroencephalogram (EEG) signals to extract band-limited intrinsic mode functions (BLIMFs), which are selected using fuzzy dispersion entropy (FDE). The extracted BLIMFs are fed into deep stack autoencoder (DSAE) with a minimum reconstruction error, utilizing root mean square (RMS) as the cost function. We also demonstrate how to apply the robust broad learning system (RBLS) to classify neuro-disorders, comparing it with various broad learning system (BLS) methods for schizophrenia classification. Building on RBLS’s success, we propose a novel VMD-based BLS (VMD-BLS) technique. To address VMD-BLS’s limitations, we introduce a PVMD-DSAE based RBLS (PVMD-DSAE-RBLS). The effectiveness of PVMD-DSAE-RBLS is tested on three datasets, with results showing accuracies of 99.98%, 96.91% and 99.29% for the Poland, Kaggle, and Moscow datasets, respectively. The performance of the proposed PVMD-DSAE-RBLS method significantly outperforms compared to similar learning algorithms and state-of-the-art techniques. Finally, a reconfigurable high-speed field-programmable gate array (FPGA) embedded processor is implemented to design a computer-aided diagnosis (CAD) system, providing efficient automated diagnosis for schizophrenia patients.}, archive = {J_EAAI}, author = {Sebamai Parija and Mrutyunjaya Sahani and Susanta Kumar Rout}, doi = {10.1016/j.engappai.2025.111294}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111294}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust broad learning system with parametrized variational mode decomposition for schizophrenia diagnosis}, volume = {158}, year = {2025}, }
• (2025). An empirical evaluation of preprocessing methods for machine learning based network intrusion detection systems. EAAI, 158, 111289. (www)

  @article{ , abstract = {Despite extensive efforts in developing machine learning-based Network Intrusion Detection Systems, inconsistencies in the choice of pre-processing of training data exists across the literature. This study surveys both seminal and recent works, identifying that while the same benchmark datasets and model classes are frequently used, the data pre-processing methods vary significantly, often without in-depth discussion and justification, limiting performance and hindering fair comparison. This work aims to address this problem, by exploring the possibility of a standardised pre-processing for Network Intrusion Detection Systems benchmark datasets We perform a detailed experimental evaluation on the most common machine learning models and four prevalent Network Intrusion Detection Systems benchmark datasets, considering a total of 72 combinations of pre-processing methods for both numerical and categorical fields in the network flow data. Based on our experimental evaluation, we propose a uniform (or ’standardised’) data pre-processing approach, which consistently performs well across the considered datasets and machine learning models. We demonstrate that our proposed method can achieve up to a 11% increase in detection accuracy over commonly applied pre-processing techniques for a shallow neural network (for the purposes of this work, we deem networks with fewer than 3 hidden layers as shallow), versus less-optimal pre-processing approaches. To facilitate future research, we make our code publicly available 1 .}, archive = {J_EAAI}, author = {Liam Daly Manocchio and Siamak Layeghy and Marcus Gallagher and Marius Portmann}, doi = {10.1016/j.engappai.2025.111289}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111289}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An empirical evaluation of preprocessing methods for machine learning based network intrusion detection systems}, volume = {158}, year = {2025}, }
• (2025). Domain generalization for multi-modal face anti-spoofing via MixCropMask augmentation and deep hybrid-level fusion. EAAI, 158, 111287. (www)

  @article{ , abstract = {The domain generalization capability of face anti-spoofing (FAS) models is crucial for addressing unseen attacks. With advancements in sensor fabrication and multi-modal learning techniques, numerous multi-modal FAS approaches have emerged, but these methods face challenges when handling unseen attacks. These challenges stem from (1) data imbalance: where significant differences in data quality and multimodal feature distributions across different source domains, particularly in the context of multimodal data, hinder the model’s ability to learn consistent feature representations across all domains, and (2) modality unreliability, where sensors experience significant domain shifts across different environments, resulting in the propagation of unreliable information during cross-modal feature fusion. To address modality unreliability, we propose a Multi-Modal Domain Generalized Face Anti-Spoofing (MMDGFAS) framework, which incorporates a multi-stream network to independently process and capture the inherent heterogeneity of each modality. Within this framework, Selective Kernel Convolution is integrated to enhance the bottleneck structure, expanding the receptive field for effective multi-scale information fusion across modalities. Furthermore, we develop a deep hybrid-level fusion strategy to strengthen complementary integration between modalities, mitigating the interference caused by unreliable cross-modal information. To address modality imbalance, we propose the MixCropMask Augmentation (MCMA). MCMA generates semantically consistent and diverse samples through a variety of augmentation operation chains, thereby enhancing the quality of the data. MCMA incorporates a local masking strategy, akin to Dropout techniques, to effectively mitigate the model’s over-reliance on specific areas. Extensive domain generalization experiments conducted on four multimodal datasets demonstrate that our method outperforms state-of-the-art approaches.}, archive = {J_EAAI}, author = {Botao Zhang and Hongjiao Li}, doi = {10.1016/j.engappai.2025.111287}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111287}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Domain generalization for multi-modal face anti-spoofing via MixCropMask augmentation and deep hybrid-level fusion}, volume = {158}, year = {2025}, }
• (2025). Hypergraph-driven soft semantics flexible learning for visible–infrared person re-identification. EAAI, 158, 111286. (www)

  @article{ , abstract = {Visible–infrared person re-identification (VI-ReID) aims to match the images of the person across different modalities. The main challenge in the engineering application of VI-ReID lies in the considerable modality gap between visible and infrared images. Existing methods primarily focus on learning low-order hard semantics to reduce the modality gap, such as specific body parts (e.g., arms, legs, and hands), which are modality-specific and sensitive to cross-modality variations, leading to difficult modality alignment. However, high-order soft semantics contain more modality-invariant features that can be flexibly learned by the model to better align the cross-modality features and reduce the modality gap. To better learn soft semantics, we propose a novel, hypergraph-driven soft semantics flexible learning network (HSFLNet), which extracts hierarchical semantics and flexibly explores the relationships among the extracted soft semantics by using hypergraph neural networks (HGNNs). Specifically, first, we propose a soft semantics mining (SSM) module to capture and flexibly fuse hierarchical features, which integrates low- and high-level semantics across channel and spatial dimensions to align modalities. Second, a hypergraph-driven soft semantics flexible learning (HSFL) module is designed, which employs HGNNs to explore the relationships among soft semantics. These semantics are adaptively learned through gating mechanisms to capture modality-invariant features, thereby reducing the modality gap. We evaluate the performance of HSFLNet on SYSU-MM01, RegDB, and low-light cross-modality (LLCM) datasets, achieving a Rank-1/mean average precision of 75.09%/72.11%, 95.82%/92.42%, and 65.42%/68.55%, respectively. Experimental results demonstrate that HSFLNet outperforms state-of-the-art VI-ReID methods. Our code is available at https://github.com/Jiacheng813/HSFLNet .}, archive = {J_EAAI}, author = {Jiacheng Zhu and Hongwei Ge and Yuxuan Liu and Chunguo Wu and Jiulin Fan}, doi = {10.1016/j.engappai.2025.111286}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111286}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hypergraph-driven soft semantics flexible learning for visible–infrared person re-identification}, volume = {158}, year = {2025}, }
• (2025). Catching spinning table tennis balls in simulation with end-to-end curriculum reinforcement learning. EAAI, 158, 111285. (www)

  @article{ , abstract = {The game of table tennis is renowned for its extremely high spin rate, but most table tennis robots today struggle to handle balls with such rapid spins. To address this issue, we have proposed a series of methods, including: 1. Curriculum Reinforcement Learning (RL): This method helps the table tennis robot learn to play table tennis progressively from easy to difficult tasks. 2. Analysis of Spinning Table Tennis Ball Collisions: We have conducted a physics-based analysis to generate more realistic trajectories of spinning table tennis balls after collision. 3. Definition of Trajectory States: The definition of trajectory states aids in setting up the reward function. 4. Selection of Valid Rally Trajectories: We have introduced a scheme for selecting valid rally trajectories to ensure that the robot’s training is not influenced by abnormal trajectories. 5. Reality-to-Simulation (Real2Sim) Transfer: This scheme is employed to validate the trained robot’s ability to handle spinning balls in real-world scenarios. With Real2Sim, the deployment costs for robotic reinforcement learning can be further reduced. Moreover, the trajectory-state-based reward function is not limited to table tennis robots; it can be generalized to a wide range of cyclical tasks. To validate our robot’s ability to handle spinning balls, the Real2Sim experiments were conducted. For the specific video link of the experiment, please refer to: https://www.youtube.com/watch?v=uAx7eYAGylc .}, archive = {J_EAAI}, author = {Xiaoyi Hu and Yue Mao and Gang Wang and Qingdu Li and Jianwei Zhang and Yunfeng Ji}, doi = {10.1016/j.engappai.2025.111285}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111285}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Catching spinning table tennis balls in simulation with end-to-end curriculum reinforcement learning}, volume = {158}, year = {2025}, }
• (2025). Corrigendum to “Multi-objective optimization of lightweight and crashworthiness of automotive front-end structures based on stacked regression surrogate” [Eng. appl. artific. intellig. 155 (2025) 111138]. EAAI, 158, 111284. (www)

  @article{ , archive = {J_EAAI}, author = {Lian Fengmin and Wang Dengfeng and Zhang Zifeng and Chen Hongli and Xu Wenchao}, doi = {10.1016/j.engappai.2025.111284}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111284}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Corrigendum to “Multi-objective optimization of lightweight and crashworthiness of automotive front-end structures based on stacked regression surrogate” [Eng. appl. artific. intellig. 155 (2025) 111138]}, volume = {158}, year = {2025}, }
• (2025). Global–local adaptive resampling strategy for enhancing the performance of physics-informed neural networks. EAAI, 158, 111283. (www)

  @article{ , abstract = {In recent years, the rapid development of artificial intelligence (AI) has brought innovative solutions to the field of engineering. Physical-informed neural networks (PINNs) have provided a new paradigm for solving partial differential equations (PDEs). However, PINNs are extremely sensitive to the number and distribution of collocation points. Insufficient collocation points or uneven distribution may lead can lead to solution failure. To address this issue, this study proposes a global–local adaptive resampling strategy (GLAR) that combines Monte Carlo integration with PINNs. During the training process, the collocation points are resampled in both global and local regions according to the Monte Carlo integral values to improve the accuracy of the model. Numerical experiments show that GLAR-PINNs is comparable to existing resampling methods when dealing with linear PDEs (such as the Diffusion and Wave equations). When it comes to solving nonlinear PDEs (such as the Burgers, Korteweg–de Vries, and Allen–Cahn equations), the accuracy is improved has been improved by 14.65 times, 38.04 times, and 7.14 times, respectively. In addition, we applied this strategy to reconstruct the flow field around a two-dimensional triangular cylinder, and the relative error of the reconstructed velocity field was less than 0.001. This significantly promoting significantly promotes the application and development of artificial intelligence in the engineering field.}, archive = {J_EAAI}, author = {Lei Gao and Dan Zhang and Yaoran Chen and Xiaowei Li and Chunxin Li and Yan Peng}, doi = {10.1016/j.engappai.2025.111283}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111283}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Global–local adaptive resampling strategy for enhancing the performance of physics-informed neural networks}, volume = {158}, year = {2025}, }
• (2025). A multiple knowledge-based evolutionary algorithm for sparse large-scale multi-objective problems. EAAI, 158, 111281. (www)

  @article{ , abstract = {In real-world applications, sparse large-scale multi-objective optimization problems (LSMOPs) are prevalent. Sparse LSMOPs are the LSMOPs characterized by Pareto-optimal solutions with sparse decision variables. Nonetheless, limited attention has been given to developing general-purpose algorithms for sparse LSMOPs. Existing approaches primarily focus on detecting sparsity, whereas achieving optimal results requires accurate detection of sparse distributions and simultaneous optimization of non-zero variables. Therefore, this paper proposes a multiple knowledge-based sparse large-scale multi-objective evolutionary algorithm. Building on genetic algorithms, the proposed algorithm employs a two-layer encoding scheme, develops a knowledge-driven evolution strategy for optimizing binary vectors, and introduces an association optimization method for optimizing real vectors. The performance of the proposed algorithm is assessed using both benchmark tests and real-world applications. Experimental results demonstrate that the proposed algorithm is competitively effective in solving sparse LSMOPs.}, archive = {J_EAAI}, author = {Wanting Yang and Jianchang Liu and Yuanchao Liu and Wei Zhang and Tianzi Zheng}, doi = {10.1016/j.engappai.2025.111281}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111281}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multiple knowledge-based evolutionary algorithm for sparse large-scale multi-objective problems}, volume = {158}, year = {2025}, }
• (2025). Cutting tool life prediction and extension through generative model-augmented deep learning and laser remanufacturing techniques. EAAI, 158, 111276. (www)

  @article{ , abstract = {Predicting and extending the remaining life of cutting tools during machining processes is essential for sustainable manufacturing. Traditional prognosis methods often struggle to adapt to different working conditions over the machining process lifecycle. This paper proposes a novel framework that effectively addresses the challenges by integrating multi-source data and using deep learning techniques. The system integrates augmented-power and vibration data collected from computer numerical control machines with the following innovations: (1) A hybrid temporal convolutional network (TCN)-attention model is developed for cutting tool remaining life prognosis, which achieves the best accuracy of 98.51 % and average of 97.62 %. In addition, optimal laser shock peening parameters are selected using a deep neural network and enhanced ternary bees algorithm. (2) A time-series generative adversarial network is used for data augmentation, which increases data quantity for TCN model training. (3) Data quality is evaluated using the t-distributed stochastic neighbor embedding, Fréchet inception distance, and root mean squared error to ensure similarity between real and generated data. (4) The effectiveness of the remanufacturing approach is validated with a 28.95 % and 30.77 % increase in tool life based on finite element analysis and experimental testing, respectively. This comprehensive approach contributes to enhancing tool life prediction accuracy and optimizing sustainable remanufacturing processes, thereby enhancing production efficiency and reducing waste in machining operations.}, archive = {J_EAAI}, author = {Yuchen Liang and Yuqi Wang and Raymond Chiong and Anping Li and Jinzhong Lu}, doi = {10.1016/j.engappai.2025.111276}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111276}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cutting tool life prediction and extension through generative model-augmented deep learning and laser remanufacturing techniques}, volume = {158}, year = {2025}, }
• (2025). A lightweight detection algorithm for camellia oleifera buds, stamens, and flowers using you only look once with large selective kernel network. EAAI, 158, 111275. (www)

  @article{ , abstract = {To address the issues of occlusion and missed detections in small targets during the recognition of Camellia oleifera buds, stamens, and flowers and to improve recognition accuracy and computation speed, this study proposes a lightweight detection model based on YOLOv8s (You Only Look Once version 8 small). Firstly, we enhanced detection effectiveness by replacing the original YOLOv8s backbone with the Large Selective Kernel Network (LSKNet) and introducing the Minimum Point Distance Intersection over Union (MPD-IoU) loss function to accelerate convergence and improve recognition of overlapping targets. Secondly, for model lightweighting, we incorporated the Partial Convolution (PCConv) module to reduce parameters and floating-point operations (FLOPs) while enhancing feature representation. An additional detection head was added to improve small bud target detection. Experimental results show our improved model increased precision (P), recall (R), and mean average precision (mAP) by 0.3%, 1.2%, and 1.2% respectively over baseline YOLOv8s. Compared to mainstream models – YOLOv3-tiny, ScaledYOLOv4, YOLOv5s, YOLOv7, YOLOv8s, and Faster Region-Based Convolutional Neural Network (Faster R-CNN) – our model achieved mAP improvements of 1.8%, 1.5%, 1.4%, 2.1%, 1.2%, and 7.7% respectively. The optimized model demonstrates faster, more accurate identification of Camellia oleifera buds, stamens, and flowers, making it suitable for mobile deployment.}, archive = {J_EAAI}, author = {Fei Long and Lijun Li and Yang Liu and Haifei Chen and Yuyan Zhang and Haorui Wang}, doi = {10.1016/j.engappai.2025.111275}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111275}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight detection algorithm for camellia oleifera buds, stamens, and flowers using you only look once with large selective kernel network}, volume = {158}, year = {2025}, }
• (2025). Formal verification for multi-agent path execution in stochastic environments. EAAI, 158, 111266. (www)

  @article{ , abstract = {Multi-agent pathfinding aims to determine conflict-free paths for multiple agents in a shared environment. However, real-world uncertainties can disrupt preplanned paths, leading to delays and new conflicts. Addressing these challenges requires robust strategies for path execution and adjustment. While many multi-agent pathfinding algorithms have been proposed, this work does not introduce a new algorithm. Instead, it presents an adjustment solution based on a set of constraint rules and a priority strategy to avoid conflicts and deadlocks. Additionally, a Markov decision process model is developed, derived from the preplanned paths, and integrated with the adjustment solution to account for stochastic environmental uncertainties. A novel integrated framework is proposed for formally analyze and verify the reliability of multi-agent path execution and the robustness of the adjustment solution in stochastic environments, with formal verification achieved through a logic-based probabilistic model checker. The performance of the proposed framework is validated through various scenarios on the Flatland platform. Results demonstrate that the adjustment solution, based on the constraint rules, effectively mitigates conflicts and deadlocks, improving robustness. Furthermore, formal verification proves effective in assessing the reliability and robustness of multi-agent path execution under uncertainty.}, archive = {J_EAAI}, author = {Xia Wang and Jun Liu and Chris D. Nugent and Shaobing Xu and Yang Xu}, doi = {10.1016/j.engappai.2025.111266}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111266}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Formal verification for multi-agent path execution in stochastic environments}, volume = {158}, year = {2025}, }
• (2025). Towards pedestrian head tracking: A benchmark dataset and a multi-source data fusion network. EAAI, 158, 111265. (www)

  @article{ , abstract = {Pedestrian detection and tracking in crowded video sequences have many applications, including autonomous driving, robot navigation and pedestrian flow analysis. However, detecting and tracking pedestrians in high-density crowds face many challenges, including intra-class occlusions, complex motions, and diverse poses. Although artificial intelligence (AI) models have achieved great progress in head detection, head tracking datasets and methods are extremely lacking. Existing head datasets have limited coverage of complex pedestrian flows and scenes (e.g., pedestrian interactions, occlusions, and object interference). It is of great importance to develop new head tracking datasets and methods. To address these challenges, we present a Chinese Large-scale Cross-scene Pedestrian Head Tracking dataset (Cchead) and a Multi-source Data Fusion Network (MDFN). The dataset has features that are of considerable interest, including 10 diverse scenes of 50,528 frames with about 2,366,249 heads and 2,358 tracks. Our dataset contains diverse pedestrian moving speeds, directions, and complex crowd pedestrian flows with collision avoidance behaviors. Existing state-of-the-art (SOTA) algorithms are tested and compared on the Cchead dataset. MDFN is the first end-to-end convolutional neural network (CNN)-based head detection and tracking network that jointly trains Red, Green, Blue (RGB) frames, pixel-level motion information (optical flow and frame difference maps), depth maps, and density maps in videos. Ablation experiments confirm the significance of multi-source data fusion. Compared with SOTA pedestrian detection and tracking methods, MDFN achieves superior performance across three datasets: Cchead, Restaurant and Crowd of Heads Dataset (CroHD). To promote further development, we share our source code and trained models for global researchers: https://github.com/kailaisun/Cchead . We hope our datasets to become essential resources towards developing pedestrian tracking in dense crowds.}, archive = {J_EAAI}, author = {Kailai Sun and Xinwei Wang and Shaobo Liu and Qianchuan Zhao and Gao Huang and Chang Liu}, doi = {10.1016/j.engappai.2025.111265}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111265}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards pedestrian head tracking: A benchmark dataset and a multi-source data fusion network}, volume = {158}, year = {2025}, }
• (2025). Cascaded dimensionality reduction nonnegative matrix factorization for data representation. EAAI, 158, 111264. (www)

  @article{ , abstract = {Nonnegative matrix factorization (NMF), as a powerful dimensionality reduction technique, has attracted considerable attention for its excellent interpretability by utilizing relatively few linear combinations of basis vectors to represent the original data. The performance of its dimensionality reduction is affected by the quality and efficiency of finding a suitable collection of basis vectors. However, traditional NMF methods focus more on mining discriminative features rather than the quantity and quality of basis vectors. This may result in uncontrolled dimensionality and make it difficult to identify suitable basis vector sets, which can effectively capture the latent structure in the data. To alleviate these issues, we propose a cascaded dimensionality reduction nonnegative matrix factorization (CDRNMF) method. CDRNMF demonstrates distinctive attributes that differ from existing work as follows. (1) It subtly incorporates a feature selection mechanism into NMF, thereby establishing a novel cascaded dimensionality reduction framework that effectively retains the most representative features. (2) The dimensionality uncontrollability is effectively alleviated by constructing a feature selection matrix to assess and select basis vectors. (3) An optimization method is designed for solving CDRNMF efficiently. Numerical experiments validate that the performance of CDRNMF outperforms other state-of-the-art algorithms.}, archive = {J_EAAI}, author = {Yulei Huang and Jinlin Ma and Ziping Ma and Ke Lu}, doi = {10.1016/j.engappai.2025.111264}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111264}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cascaded dimensionality reduction nonnegative matrix factorization for data representation}, volume = {158}, year = {2025}, }
• (2025). Multi-axis vision transformer for medical image segmentation. EAAI, 158, 111251. (www)

  @article{ , abstract = {Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have shown remarkable success in medical image segmentation, but individually, they struggle to capture both local and global contexts. To address this limitation, we propose MaxViT-UNet, a hybrid encoder–decoder architecture that integrates convolutional operations and multi-axis self-attention to capture local and global context for effective medical image segmentation. Our novel Hybrid Decoder fuses upsampled decoder features with encoder skip connections and refines them using a multi-axis attention block, repeated across decoding stages for progressive segmentation refinement. Experimental evaluation on the MoNuSeg18 and MoNuSAC20 datasets demonstrates that MaxViT-UNet outperforms traditional CNN-based U-Net by 2.36% and 14.14% Dice score, respectively. Similarly it outperforms Swin-UNet by 5.31% on MoNuSeg18 and nearly doubles the Dice score on MoNuSAC20. These results confirm the generalization and effective segmentation capabilities of our hybrid architecture across diverse histopathological datasets.}, archive = {J_EAAI}, author = {Abdul Rehman Khan and Asifullah Khan}, doi = {10.1016/j.engappai.2025.111251}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111251}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-axis vision transformer for medical image segmentation}, volume = {158}, year = {2025}, }
• (2025). Modeling category and multi-level user intentions for session-based recommendation. EAAI, 158, 111248. (www)

  @article{ , abstract = {Session-based recommendation is extensively used in online services and businesses, which predicts users’ next interest with anonymous behavior sequence. Past research on session-based recommendation capturing user preference commonly focused solely on extracting the diversity of user intentions and the identification information of interacted items. However, fine-grained user intentions, which are reflected by both individual and consecutive items in session sequences, and the side information, item categories, which can help to learn the potential user intentions in session sequences have not been explored deeply. These insights inspire us to model c ategory and multi-level user i ntentions for s ession-based recommendation (CIS). Specifically, session sequences are first split pairwise between levels, and connected sequentially within levels to construct multi-level user intention graphs and item category graphs and model the fine-grained information. Moreover, we refine the representation of user intentions and item categories with graph convolutional network and iterative update process, in order to extract potential user intentions based on category semantics. Finally, we generate session embeddings based on each level of user intentions, and introduce an Integration Predicting strategy to anticipate users’ next interested item. Extensive experiment results on Diginetica dataset and Tmall dataset demonstrate that CIS is superior to other state-of-the-art baselines, with improvements of 24.38% in Hit Ratio (HR), 69.56% in Mean Reciprocal Rank (MRR) and 55.94% in Normalized Discounted Cumulative Gain (NDCG).}, archive = {J_EAAI}, author = {Shanshan Hua and Mingxin Gan and Menghan Li}, doi = {10.1016/j.engappai.2025.111248}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111248}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Modeling category and multi-level user intentions for session-based recommendation}, volume = {158}, year = {2025}, }
• (2025). Hierarchical meta alignment for cross-domain object detection. EAAI, 158, 111247. (www)

  @article{ , abstract = {Unsupervised domain adaptation (UDA) aims to adapt an object detector from a labeled source domain to an unlabeled target domain. In this task, multiple sub-tasks of different nature are involved, yet existing methods simply sum up the losses and train all the sub-tasks jointly. We, however, find that inconsistent optimization goals between different sub-tasks lead to limited adaptation performance. Specifically, from our analysis, we find notable gradient discrepancies between sub-tasks in a domain adaptive object detector, and especially significant conflicts between domain alignment and detection sub-tasks. Based on this analysis, we propose to solve UDA object detection from a multi-task learning perspective. Specifically, we divide all sub-tasks into two groups, and alleviate both inter-group and intra-group inconsistency via a novel Hierarchical Meta Alignment (HMA) method. At the first level, we construct a Meta Optimization Block (MOB) for each inter-group task pair, which is optimized via the Model-Agnostic Meta-Learning (MAML) algorithm. At the second level, all MOBs are optimized sequentially via the Reptile algorithm. Experimental results on various adaptation scenarios show that our proposed method outperforms previous methods.}, archive = {J_EAAI}, author = {Yang Li and Shanshan Zhang and Yunan Liu and Jian Yang}, doi = {10.1016/j.engappai.2025.111247}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111247}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical meta alignment for cross-domain object detection}, volume = {158}, year = {2025}, }
• (2025). Flow decomposition based spatial–temporal filtering self-attention networks for traffic flow forecasting. EAAI, 158, 111243. (www)

  @article{ , abstract = {Traffic prediction is essential for intelligent transportation systems but challenging due to complex spatial–temporal dependencies. Current methods often overlook data entanglement caused by stable flow sequences and traffic events, and attention mechanisms may introduce irrelevant spatial information. In this paper, we propose FDFSAN ( F low D ecomposition based spatial–temporal F iltering S elf- A ttention N etworks), which addresses these issues by decomposing traffic flow data into stationary and sudden components modeled through a dual-channel spatial–temporal network. Our filtering self-attention mechanism captures both temporal and spatial dependencies, integrating information from nearby and distant roads while minimizing irrelevant spatial noise. Extensive experiments on four real-world datasets show that FDFSAN outperforms state-of-the-art methods, making it suitable for urban traffic networks with dynamic spatial correlations and anomalies.}, archive = {J_EAAI}, author = {Ying Tang and Dawei Wu and Zhetao Han}, doi = {10.1016/j.engappai.2025.111243}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111243}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Flow decomposition based spatial–temporal filtering self-attention networks for traffic flow forecasting}, volume = {158}, year = {2025}, }
• (2025). Sliding window regression method for mechanical meter reading recognition in time-series images. EAAI, 158, 111240. (www)

  @article{ , abstract = {Remote meter reading is an important part of the intelligent transformation of old meters. In addition to smart meters, installing cameras outside traditional meters and transmitting images to cloud servers via Narrowband Internet of Things technology can make the intelligent transformation process more convenient. Against this backdrop, this paper proposes a sliding window-based joint recognition network to address the significant numerical errors and misrecognition issues in current meter reading methods. It combines maximum probability decoding and maximum average probability decoding to classify and recognize complete instrument images, effectively reducing numerical errors and improving accuracy. After the instrument data acquisition is systematized, combining access to historical data and current images can effectively improve the recognition accuracy rate. The network collaborates with a regression prediction model, performing regression analysis on historical data to establish equations for small-range visual-assisted regression recognition. Experiments on instrument datasets show that the algorithm achieves a 99.57% recognition accuracy, which is 4% higher than holistic methods and nearly 1% higher than single-digit methods. Moreover, it reduces the maximum error from an average of 68.5 to 1.06, and the mean absolute error and mean square error by nearly 20 times. The model is now ready for production use. This method is highly significant for industrial users, especially in old-style electricity meter reading scenarios in factories, such as in power monitoring systems of large-scale manufacturing enterprises, strongly supporting industrial automation.}, archive = {J_EAAI}, author = {Hao Xiu and Siran Hu and Yuanxin Cui and Jie He and Xiaotong Zhang and Yue Qi}, doi = {10.1016/j.engappai.2025.111240}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111240}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sliding window regression method for mechanical meter reading recognition in time-series images}, volume = {158}, year = {2025}, }
• (2025). Semantic knowledge transfer for semi-supervised medical image segmentation. EAAI, 158, 111235. (www)

  @article{ , abstract = {In semi-supervised medical image segmentation, due to possible differences in information content and distribution between labeled and unlabeled data, dealing with the two separately usually prevents knowledge transfer from labeled to unlabeled data. This prevents the model from effectively sharing learned information between the two types of data. To alleviate this problem, we train labeled and unlabeled data as a whole. Semantic mixing of labeled and unlabeled data is achieved by selecting and exchanging some of the region images of both through a mask to generate complementary input views. In addition, due to the limited labeled data, the unlabeled data has a weak ability to distinguish categories in the feature space. Traditional methods rely on pixel positions to generate positive and negative samples for contrastive learning to solve this problem, but relying on pixel position sampling can easily lead to semantic inconsistency, which affects the effect of feature learning; therefore, to address this problem, we propose an innovative labeled data-guided inter-class contrastive learning strategy, which extracts the category features from labeled and unlabeled data and exploits the accurate category information in the labeled data to guide contrastive learning, while introducing a similarity-based ranking weighting mechanism. Combining the two designs, we propose a new semantic knowledge transfer framework for semi-supervised medical image segmentation. Experiments demonstrate a significant improvement in our model compared to State of the Art (SOTA) on the Automatic Cardiac Diagnosis Challenge (ACDC) dataset and the Left Atrium (LA) dataset.}, archive = {J_EAAI}, author = {Shiwei Zhou and Haifeng Zhao and Leilei Ma and Dengdi Sun}, doi = {10.1016/j.engappai.2025.111235}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111235}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semantic knowledge transfer for semi-supervised medical image segmentation}, volume = {158}, year = {2025}, }
• (2025). Pointer type instrument reading method based on key point detection. EAAI, 158, 111220. (www)

  @article{ , abstract = {Accurate and generalizable instrument recognition is a requirement of multiple types of pointer instruments in power distribution plants. However, it is difficult to read instruments with limited visibility. A global geometric network (GGNet) key point detection model is proposed for pointer-type instrument recognition by analyzing the geometric structure of these instruments and detecting the scale key points, pointer and dial center. First, a spatial channel fusion (SCF) module is used to obtain the global sensory field, to enhance the detection effect of the pointer tip key points. Second, an ellipse-aware feature aggregator (EFA) module is designed to enhance the overall robustness of the model, adaptively capture the global geometric information of the key points on the instrument scale, and form a GGNet model, thus, the detection effect of the difficult-to-detect scale is improved in the scene of limited visibility, and high precision reading is finally achieved. The experimental results show that the algorithm achieves an accuracy of 94.0% in key point detection and an average error of only 0.37% in readings for the pointer gauge recognition task. The results comparing GGNet with four other different network models show that the proposed model performs best for a number of error metrics.}, archive = {J_EAAI}, author = {Yongjie Zhai and Wei Xu and Zhenyuan Zhao and Guotian Yang and Biqiang Du}, doi = {10.1016/j.engappai.2025.111220}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111220}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pointer type instrument reading method based on key point detection}, volume = {158}, year = {2025}, }
• (2025). Predicting eddy current losses in large generator rotor using data-driven short connection deep gaussian process. EAAI, 158, 111212. (www)

  @article{ , abstract = {In response to the increased prediction errors observed when applying deep Gaussian process (DGP) models to the eddy current loss data of large generator rotors, an explainable artificial intelligence(AI)-based DGP prediction method incorporating short connections design is proposed. This method reparameterizes multivariate normal random variables between layers and concatenates processed input-layer data to construct new inter-layer inputs for the subsequent layer, thereby establishing a short-connected DGP model. This approach effectively mitigates the accumulation of prediction errors due to increased model depth, significantly enhancing the model’s adaptability and predictive accuracy in complex data structures. Experimental results demonstrate that the short-connected DGP model achieved exceptional predictive performance for eddy current loss estimation, as evidenced by a coefficient of determination of 0.9886 and a mean squared error of 0.0026. Furthermore, as the number of layers increases, the short-connected DGP model outperforms both existing DGP models and alternative approaches in prediction accuracy. The proposed AI method not only robustly supports high-precision prediction of eddy current losses in the rotor slot filling components of large generators, but also serves as an efficient and reliable modeling tool for their design and optimization, thereby possessing significant theoretical and practical value.}, archive = {J_EAAI}, author = {Hai Guo and Jinlin Cai and Likun Wang and Junxing Zhang and Fabrizio Marignetti and Bixian Zhang}, doi = {10.1016/j.engappai.2025.111212}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111212}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predicting eddy current losses in large generator rotor using data-driven short connection deep gaussian process}, volume = {158}, year = {2025}, }
• (2025). Optimal features driven hybrid attention network for effective video summarization. EAAI, 158, 111211. (www)

  @article{ , abstract = {Video summarization (VS) has emerged as an effective method for extracting meaningful content from large video repositories. Recent advancements in visual intelligence have significantly improved the ability to summarize lengthy, raw videos into concise yet representative content. However, existing VS methods extract features from static GoogleNet Pool5 without empirical analysis at an early stage. Moreover, these approaches often lack mechanisms to jointly refine channel-wise and spatial-wise feature interactions, resulting in inadequate learning of complex visual semantics and ultimately suboptimal summarization performance. To address these limitations, we propose the Hybrid-Attention VS Network (HAVSNet), which conceptualizes VS as a keyframe selection task. Our method integrates representative intermediate features early in the network, significantly enhancing feature representation compared to conventional techniques. Furthermore, HAVSNet incorporates a hybrid-attention mechanism for advanced feature refinement: channel attention highlights the most discriminative feature maps, while an optimized spatial attention module captures and refines spatial dependencies. This enables the network to focus on the most informative and visually salient regions. Additionally, explainable AI (XAI) via heatmap visualizations further enhances interpretability by revealing how the model prioritizes salient regions, offering insights into the focus of the model and optimal feature selection. Extensive experiments demonstrate that our network outperforms state-of-the-art methods, achieving notable improvements. Comprehensive quantitative and qualitative analyses further confirm the effectiveness of the proposed network. Moreover, the proposed HAVSNet is evaluated across three training configurations of canonical, augmented, and transfer settings, showing its strong generalization and adaptability.}, archive = {J_EAAI}, author = {Habib Khan and Samee Ullah Khan and Waseem Ullah and Sung Wook Baik}, doi = {10.1016/j.engappai.2025.111211}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111211}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimal features driven hybrid attention network for effective video summarization}, volume = {158}, year = {2025}, }
• (2025). A network with enhanced ability to interact scale features with channel features. EAAI, 158, 111209. (www)

  @article{ , abstract = {Road damage must be accurately and promptly detected because road hazards have the potential to cause catastrophic traffic accidents. A Multiscale channel Shuffle Fusion-You Only Look Once(MSF-YOLO) series algorithm with multiscale cross-channel interaction data is suggested as a solution to this issue. It solves the problem of continuous change of road damage scale during vehicle traveling by enhancing the expressiveness of scale features and improves the network detection performance by enhancing the inter-channel interaction features so that the network learns the original input features as well as the input features after channel interaction with almost no additional computational cost. The experimental results show that Multiscale channel Shuffle Fusion-You Only Look Once-small (MSF-YOLO-s) accuracy on the Global Road Damage Detection Challenge (GRDDC) 2020 dataset is improved by 5.8% to reach 65.9%, while MSF-YOLO-s computation amount and number of parameters are reduced by 20.9% and 8.7%, respectively, compared to baseline. While experimental verification on the Common Objects in Context (COCO) 2017 datasets confirms its good generalizability. Ultimately, the model is implemented throughout the entire vehicle to anticipate fractures and enhance driving safety.}, archive = {J_EAAI}, author = {Peng Su and Huizi Han and Mei Liu and Siqun Ma and Jiasheng Chen}, doi = {10.1016/j.engappai.2025.111209}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111209}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A network with enhanced ability to interact scale features with channel features}, volume = {158}, year = {2025}, }
• (2025). Multi type mean field reinforcement learning for optimal resource allocation in heterogeneous network. EAAI, 158, 111207. (www)

  @article{ , abstract = {With the exponential growth in the amount of data transmitted over mobile networks, contemporary 5G communication technologies with the primary goal of improving network performance and quality of service have gained much attention. Efficient resource allocation and interference management are especially critical in large-scale wireless networks. Device-to-device (D2D) communication has become a promising technological tool to address this growing need. However, the limitation of exponentially growing solution space in large-scale ultra-dense networks makes it difficult to achieve real-time control with conventional optimization methods. To face this challenge, we propose a novel framework that combines Multi-Agent Reinforcement Learning (MARL) with Mean Field Type Game (MFTG) theory, allowing agents to operate in different action spaces. This approach extends the core principle of mean-field reinforcement learning from a single type to multiple types of interactions, effectively modeling the approximate behavior between various types of devices in heterogeneous D2D networks. Experimental results show that the proposed Multi-Type Mean-Field double deep Q-network (MTMF-Q) method outperforms benchmark methods in heterogeneous networks. In addition, the proposed method exhibits good scalability in parameters such as user density, network size and power budget, showing its potential for application in ultra-dense heterogeneous communication network scenarios.}, archive = {J_EAAI}, author = {Qiushi Sun and Yuyi Zhang and Haitao Wu and Yin Li and Ovanes Petrosian}, doi = {10.1016/j.engappai.2025.111207}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111207}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi type mean field reinforcement learning for optimal resource allocation in heterogeneous network}, volume = {158}, year = {2025}, }
• (2025). Stochastic event-triggered impulsive control of delayed networks under cyber attacks. EAAI, 158, 111191. (www)

  @article{ , abstract = {This paper investigates the problem of stochastic event-triggered impulsive control for stochastic complex networks with time delays under cyber attacks. To account for the influence of cyber attacks, the variations in event-triggered parameters induced by these attacks are incorporated into the design as stochastic event-triggered conditions. A stochastic event-triggered impulsive control strategy is then proposed for complex networks with time delays. Unlike existing impulsive control methods, which typically assume that control input instants are independent of the system state, the proposed method is state-dependent and explicitly considers the impact of cyber attacks on the impulsive control gain-an aspect often overlooked in most control approaches addressing cyber attacks. By constructing auxiliary functions and integrating graph theory with the Razumikhin method, several criteria are derived to ensure the exponential synchronization of stochastic complex networks with time delays under cyber attacks. Finally, simulation results are provided to validate the effectiveness and advantages of the proposed method.}, archive = {J_EAAI}, author = {Ni Yang and Jiakun Chen and Huan Su}, doi = {10.1016/j.engappai.2025.111191}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111191}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Stochastic event-triggered impulsive control of delayed networks under cyber attacks}, volume = {158}, year = {2025}, }
• (2025). Beyond a single solution: Liquefied natural gas process optimization using niching-enhanced meta-heuristics. EAAI, 158, 111119. (www)

  @article{ , abstract = {The escalating global energy demand and the imperative to mitigate climate change necessitate optimizing the natural gas liquefaction process to enhance energy efficiency, reduce costs, and improve sustainability. Traditional approaches typically focus on identifying a single high-quality solution. However, such methods often overlook other potential solutions that are equally good or comparable to the global optimum. These solutions provide alternative, economically viable options for engineers, enabling more insightful decisions. Motivated by this research gap, we propose the application of multi-solution meta-heuristics for the dual-effect single-mixed-refrigerant (DSMR) process, a novel approach in the field of liquefied natural gas (LNG) process optimization. To comprehensively evaluate algorithm performance, we introduce a Quantity-Quality comparative evaluation method, which ranks algorithms based on both the number and quality of solutions identified without requiring prior knowledge of the global optimum. We highlight the effectiveness of multi-solution meta-heuristics in identifying multiple feasible high-quality solutions outperforming commonly used single-solution approaches. Our assessment encompasses exergy efficiency and economic viability, with results showing that 15 out of 17 multi-solution variants surpass the popular genetic algorithm (GA) in global optimization, achieving up to 3% reduction in energy consumption compared to the state-of-the-art solution. Additionally, all niching-enhanced meta-heuristics consistently identified more high-quality solutions than single-solution meta-heuristics. An exergo-economic analysis of the top 15 solutions revealed exergy efficiency improvement of up to 37.73%, with capital cost reductions ranging from 0.27% to 5.22%, and operating cost reductions from 0.24% to 3.1%, compared to the base case. The code is available on GitHub .}, archive = {J_EAAI}, author = {Mohamed Hamdy and Shahd Gaben and Abdullah Al-Saadi and Abdulaziz Al-Ali and Majeda Khraisheh and Fares Almomani and Ponnuthurai N. Suganthan}, doi = {10.1016/j.engappai.2025.111119}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111119}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Beyond a single solution: Liquefied natural gas process optimization using niching-enhanced meta-heuristics}, volume = {158}, year = {2025}, }
• (2025). A new filter based-instance selection for high dimensional data classification. EAAI, 158, 111082. (www)

  @article{ , abstract = {Feature ranking is an essential process in data analysis, where the importance or relevance of different features is assessed to identify the most influential ones for predictive modeling or decision-making. In this paper, we introduce a novel and comprehensive method for heterogeneous feature Ranking based on Instance Selection (RIS). RIS addresses the challenges that arise when dealing with high-dimensional datasets, where noisy instances significantly impact the true relevance of each variable, thus affecting the actual ranking of variables. In this work, ”noisy instances” refer to observations that contribute to noise in the data, such as non-informative, redundant, or ambiguous instances that may hinder the learning process or obscure the underlying patterns. To overcome this issue and select a subset of instances, excluding the noisy ones, we employ the fuzzy C-means algorithm (FCM) along with a genetic algorithm-based instance selection technique. By considering the effectiveness of features in both scenarios, we obtain a comprehensive evaluation that captures their true relevance. RIS is evaluated on three simulation scenarios and nine publicly accessible datasets, achieving high average accuracies when calculated using three classifiers: support vector machine, Naive Bayes, and Random Forest. The accuracies achieved by RIS are the highest compared to the state-of-the-art approaches reviewed in this paper. Additionally, RIS selects up to 100% of the relevant features when executed on simulated data, demonstrating its capability in feature selection and data preprocessing within the field of artificial intelligence.}, archive = {J_EAAI}, author = {Basma El Mourtji and Tayeb Ouaderhman and Hasna Chamlal}, doi = {10.1016/j.engappai.2025.111082}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111082}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new filter based-instance selection for high dimensional data classification}, volume = {158}, year = {2025}, }
• (2025). Improved U-net with weighted loss function and its application in the prediction of remaining oil distribution. EAAI, 157, 111571. (www)

  @article{ , abstract = {Research on predicting the remaining oil distribution using artificial neural networks has been thoroughly conducted. However, conventional methods often perform poorly in regions exhibiting drastic changes in oil saturation. To address this limitation, the paper adopts U-net as the prediction model and improves the loss function by dynamically adjusting the loss weights of each pixel based on the extent of changes within the remaining oil distribution map. Furthermore, an exponential relationship between the weight of the loss function and the variance of oil saturation is established, and thus the loss function is modified to be saturation-variance-dependent. The findings demonstrate that modifying the weighted loss function enhances U-net's learning capability in regions with drastic changes in oil saturation, thereby effectively improving prediction performance in these challenging areas. Moreover, the optimal prediction performance of U-net is achieved when the base value of the exponential weighted function is set to 2. The proportion of predictions with R 2 greater than 0.95 is 72 % using U-net with the weighted loss function, marking a 48 % increase compared to the U-net without modifications. The study provides a promising method for prediction of remaining oil distribution, and can also serve as a valuable reference for application of the U-net in image prediction with drastic local changes.}, archive = {J_EAAI}, author = {Kang Zhou and Ke Su and Zhibin An and Yingjie Chen and Jian Hou}, doi = {10.1016/j.engappai.2025.111571}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111571}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improved U-net with weighted loss function and its application in the prediction of remaining oil distribution}, volume = {157}, year = {2025}, }
• (2025). Thermo-mechanical physics-informed deep learning for prediction of thermal stress evolution in laser metal deposition. EAAI, 157, 111554. (www)

  @article{ , abstract = {Understanding thermal stress evolution in metal additive manufacturing (AM) is crucial for producing high-quality components. Recent advancements in machine learning (ML) have shown great potential for modeling complex multiphysics problems in metal AM. While physics-based simulations face the challenge of high computational costs, conventional data-driven ML models require large, labeled training datasets to achieve accurate predictions. Unfortunately, generating large datasets for ML model training through time-consuming experiments or high-fidelity simulations is highly expensive in metal AM. To address these challenges, this study introduces a physics-informed neural network (PINN) framework that incorporates governing physical laws into deep neural networks (NNs) to predict temperature and thermal stress evolution during the laser metal deposition (LMD) process. The study also discusses enhanced accuracy and efficiency of the PINN model when supplemented with small simulation data. Furthermore, it highlights the PINN transferability, enabling fast predictions with a set of new process parameters using a pre-trained PINN model as an online soft sensor, significantly reducing computation time compared to physics-based numerical models while maintaining accuracy.}, archive = {J_EAAI}, author = {R. Sharma and Y.B. Guo}, doi = {10.1016/j.engappai.2025.111554}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111554}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Thermo-mechanical physics-informed deep learning for prediction of thermal stress evolution in laser metal deposition}, volume = {157}, year = {2025}, }
• (2025). Combining preventative maintenance planning and production scheduling with sequence dependent setups for decomposing items. EAAI, 157, 111550. (www)

  @article{ , abstract = {This paper addresses a single-machine scheduling problem involving adjustable processing durations, sequence-dependent setups for perishable items, and various preventative maintenance (PM) strategies. These measures aim to optimize machine availability or maintain a minimal level of dependability, including a set of parallel deteriorating machines. Two metaheuristic approaches are employed: a non-dominated sorting genetic algorithm (NSGA-II) and a non-dominated ranking genetic algorithm (NRGA). The model aims to find the production lot sizes and the sequence and interval of PM activities over a multi-shift planning horizon. The two heuristics algorithms aim to solve the proposed model for small and medium-scale problems. The computational efficiency of the heuristic is improved when used in combination with preventive maintenance policies to minimize the total cost. The results demonstrate the performance of the heuristic NSGA-II on a set of randomly generated test problems. The heuristic is shown to capture a substantial portion of the global Pareto-optimal solutions in instances of moderate complexity.}, archive = {J_EAAI}, author = {Elham Saghafi Oskuei and Mahdi Yousefi Nejad Attari and Ali Ala and Vladimir Simic and Dragan Pamucar}, doi = {10.1016/j.engappai.2025.111550}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111550}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Combining preventative maintenance planning and production scheduling with sequence dependent setups for decomposing items}, volume = {157}, year = {2025}, }
• (2025). MSCIAIG-net: Multi-scale spatial-channel interactive attention and interpretable guidance for fine-grained ship classification and visual inference network in remote sensing images. EAAI, 157, 111520. (www)

  @article{ , abstract = {Fine-grained ship classification (FGSC) in remote sensing images (RSI) with high visual similarity is becoming increasingly important in civil and military applications involving ship monitoring. Existing studies predominantly seek to enhance the accuracy of classification and recognition of artificial intelligence models only by designing deep network structures or algorithms, which often extract redundant information containing much background noise, ultimately degrading the robustness of the fine-tuned model. We provide a multi-scale spatial-channel interactive attention and interpretable guidance network (MSCIAIG-Net) for FGSC in RSI. This is the first study to understand and address the issue of robustness degradation resulting from parameter redundancy in FGSC models for RSI through visualized attributions. Specifically, a multi-scale spatial channel interactive attention module (MSCIAM) is designed to model and interact with the importance between multi-scale spatial and channel features. Under the guidance of interpretability, a novel fine-grained classification loss with visual interpretation is developed, which strengthens the network's ability to learn discriminative feature representations in salient regions. Additionally, an interpretability-analysis-based network pruning (IANP) strategy is introduced to reduce the noise and redundant information from the network, thereby achieving model compression and performance enhancement within a certain range without sacrificing accuracy. Furthermore, performance verification is implemented on the public dataset, Fine-grained Ship Collection-23 (FGSC-23), and our collated dataset, and MSCIAIG-Net attains the competitive accuracy and interpretability under identical conditions. Notably, for MSCIAIG-Net with Residual Network 50 Layers as the backbone network verified on FGSC-23, its overall accuracy ( OA ) and average accuracy ( AA ) improves by 6.79 % and 5.54 %, respectively.}, archive = {J_EAAI}, author = {Chengqiang Zhao and Shijie Chen and Jiashu Zhang}, doi = {10.1016/j.engappai.2025.111520}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111520}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MSCIAIG-net: Multi-scale spatial-channel interactive attention and interpretable guidance for fine-grained ship classification and visual inference network in remote sensing images}, volume = {157}, year = {2025}, }
• (2025). A lightweight hybrid neural network for remote sensing image super-resolution reconstruction. EAAI, 157, 111438. (www)

  @article{ , abstract = {To address the challenges of large model sizes and high computational complexity in existing deep learning-based super-resolution neural networks, we design a lightweight hybrid neural network for better used in applications of resource constraints or high real-time requirements, such as edge computing devices, autonomous driving, intelligent agriculture, natural disaster monitoring, and so on. The developed method can reduce the model complexity by combining lightweight convolutional neural network and transformer, while maintaining more super-resolution details. Firstly, the lightweight convolutional ghost modules replace traditional convolutions to build a lightweighted network structure, because of using the convolution with fewer channels and a cheap operation. Secondly, the high-frequency filtering multi-distillation block refines features and extracts high-frequency details through average pooling and upsampling, thereby enriching the details of the reconstructed images. And then, the shifted window transformer is used to capture global information to enhance the overall visual coherence of the reconstructed image by using the shifted window-based self-attention mechanism. Finally, a high-resolution image is obtained by feature fusion and sampling. Experimental results show that the developed method improved both objective evaluation metrics and visual effects. Compared with the state-of-the-art method, the peak signal-to-noise ratio of the developed method is increased about 0.054 dB, the structural similarity is increased by 0.003, while the learned perceptual image patch similarity is decreased by 0.029. The number of model parameters is reduced by 1.71× 10 5 , and the number of multiplication and addition operations is decreased by 8.2× 10 9 .}, archive = {J_EAAI}, author = {Qi Zhang and Fuzhen Zhu and Bing Zhu and Puying Li and Yvshuo Han}, doi = {10.1016/j.engappai.2025.111438}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111438}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight hybrid neural network for remote sensing image super-resolution reconstruction}, volume = {157}, year = {2025}, }
• (2025). Neural network-based guaranteed performance sliding mode security control of nonlinear cyber-physical systems. EAAI, 157, 111423. (www)

  @article{ , abstract = {A groundbreaking approach is proposed to ensure performance security control in nonlinear Cyber-Physical Systems (CPS), taking into account the presence of malicious cyber-attacks, unknown initial states, parameter disturbances, and external disturbances. Unlike existing CPS resilience control approaches that lack the ability to pre-set transient performance of CPS trajectory tracking error in advance, this paper introduces a novel neural network-based sliding mode resilience control approach with the capability to preset transient performance. This approach imposes dynamic performance constraint behavior on the trajectory tracking error of CPS. To achieve this goal, a stable criterion for predefined time convergence is first provided, allowing for setting an upper limit on system convergence time in advance. Based on this criterion, a neural network weight update law with predefined time convergence features is designed to enhance the speed and precision of approximating uncertain parts of the CPS system using radial basis function neural networks (RBFNN). Subsequently, the trajectory tracking error with arbitrary initial value is transformed into a new variable that has zero initial value and can converge to the actual tracking error within a specified time frame. By utilizing this new variable and neural network techniques, a sliding mode controller is designed to guarantee that the trajectory tracking error of CPS satisfies both predefined time convergence and transient performance constraints. Finally, experimental simulations and comparisons are conducted to verify the effectiveness and superiority of this method.}, archive = {J_EAAI}, author = {Chunwu Yin and Pei Yi and Xiangwei Bu}, doi = {10.1016/j.engappai.2025.111423}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111423}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural network-based guaranteed performance sliding mode security control of nonlinear cyber-physical systems}, volume = {157}, year = {2025}, }
• (2025). Regressor driven incremental instance segmentation for contraband items identification. EAAI, 157, 111421. (www)

  @article{ , abstract = {Detecting suspicious objects contained within passenger baggage is one of the most difficult tasks, even for the security experts. To address this problem, many researchers have developed computer-aided screening methods employing deep learning models to detect the presence of contraband items using X-ray imagery. However, all of these models have limitations or bottlenecks when it comes to detecting prohibited objects that are heavily obscured, cluttered, overlapping, and well concealed within the baggage. To overcome these challenges, we present a novel instance segmentation framework that transforms the conventional semantic segmentation models to perform instance-aware segmentation, via incremental learning, to automatically detect suspicious baggage items. This framework leverages knowledge distillation, enabling the model to iteratively learn and retain multiple instances of threat items. By continuously adapting, the model improves its ability to distinguish between different instances, showcasing significant advancements in security and threat detection. To overcome the under-segmentation and over-segmentation problem of the incremental instance segmentation, we introduced the use of a lightweight regressor model that can accurately identify the overlapping instances of the suspicious objects which enables the optimal selection of the segmentation model to perform the required task. Moreover, the proposed framework has been rigorously tested on two publicly available datasets on which it achieved the mask mean average precision scores of 0.54 and 0.51, respectively, at the inference stage. Similarly, the proposed method outperformed the state-of-the-art methods by 5.88%, and 8.51% in terms of mask mean average precision scores across both datasets, respectively. In addition to this, the proposed framework provides an optimal trade-off between performance and efficiency as compared to its competitors.}, archive = {J_EAAI}, author = {Ammara Nasim and Mahboob Anwar and Muhammad Owais and Samet Akcay and Divya Velayudhan and Muhammad Usman Akram and Mohammed Ghazal and Naoufel Werghi and Taimur Hassan}, doi = {10.1016/j.engappai.2025.111421}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111421}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Regressor driven incremental instance segmentation for contraband items identification}, volume = {157}, year = {2025}, }
• (2025). Unsupervised cross-modal hashing retrieval with multi-level interaction. EAAI, 157, 111418. (www)

  @article{ , abstract = {Unsupervised cross-modal hashing retrieval has gained considerable attention due to its efficiency and effectiveness in handling extensive unlabeled data. However, existing methods encounter the challenge of insufficient information interaction, leading to suboptimal retrieval performance. To tackle this issue, we propose an unsupervised cross-modal hashing retrieval with multi-level interaction (UCHRMI), which enables information interaction from coarse-grained to fine-grained levels. Specifically, we first design a cross-modal common hashing learning module to learn common hash codes while preserving modality invariance. Then, at the coarse-grained interaction level, we construct a modality-specific hashing learning module to enhance the discriminability of modality-specific hash codes through common hash code guidance and dual-contrastive learning, achieving interaction between common and specific information as well as inter-modality information. Additionally, at the fine-grained interaction level, we constrain the learning of common and modality-specific hash codes through instance-wise similarity, thereby facilitating interaction between instances. It ensures that similar samples can learn similar hash codes. Experimental results on several multimodal datasets show that the proposed method achieves outstanding performance across various retrieval tasks. The source code for this work can be found at https://github.com/szq0816/UCHRMI .}, archive = {J_EAAI}, author = {Zhenqiu Shu and Jie Zhang and Kailing Yong and Yongbing Zhang and Kaiwen Tan and Shengxiang Gao and Zhengtao Yu}, doi = {10.1016/j.engappai.2025.111418}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111418}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised cross-modal hashing retrieval with multi-level interaction}, volume = {157}, year = {2025}, }
• (2025). A cumulative absolute velocity prediction method based on surrounding strong motion records and deep learning. EAAI, 157, 111416. (www)

  @article{ , abstract = {Cumulative absolute velocity (CAV) is a critical parameter for assessing seismic destructiveness. Existing post-earthquake CAV prediction methods, such as interpolation techniques and ground motion prediction equations (GMPEs), face challenges in simultaneously leveraging historical strong-motion data and real-time observations from strong-motion stations. To address this limitation, this study proposes a novel CAV prediction method based on surrounding strong-motion records and deep learning. The method introduces a station group construction approach, where each group consists of a target station (an unmonitored location) and four surrounding stations with observed data. Using the Japanese strong-motion database, a dataset of 10,463 station groups was established to train the network model. A graph-based feature representation method, designed specifically for station groups, was implemented as the network input. Based on this, a graph neural network (GNN) model, GraphStation, was developed to predict CAV at unmonitored target locations. The performance of the proposed method was compared with interpolation methods and GMPEs, yielding the following key findings: (1) the proposed model achieves a coefficient of determination (R 2 ) of 0.91 for CAV prediction, outperforming existing methods. (2) By training on the station group database and utilizing real-time observations from surrounding stations, the method effectively integrates historical strong-motion data and real-time monitoring data, providing a robust and accurate approach for CAV prediction in regions lacking post-earthquake monitoring data.}, archive = {J_EAAI}, author = {Qingle Cheng and Chengshuai Niu and Hongyu Zhao and Jin Zhuang and Yuan Tian and Xinzheng Lu}, doi = {10.1016/j.engappai.2025.111416}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111416}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cumulative absolute velocity prediction method based on surrounding strong motion records and deep learning}, volume = {157}, year = {2025}, }
• (2025). Learning color prompt and position constraint for visual tracking. EAAI, 157, 111415. (www)

  @article{ , abstract = {The flourish of current visual tracking cannot be separated from powerful pre-trained backbone networks. Even the pre-trained networks frozen and used merely as a feature extractor can also obtain substantial tracking performance. However, how to acquire target-aware features suitable for visual tracking has always been a hot research topic to improve tracking robustness. Inspired by prompt learning, we propose the color prompt encoder to guide the acquisition of target-aware capability. Concretely, the color histogram features as a naive feature expression can provide complementary cues, so we employ color histogram features to construct the color target probability as a color prompt. Immediately after, the color prompt constructed is integrated into the unified tracking network to guide the generation of specific target feature maps. Furthermore, Discriminative Correlation Filters (DCF)-based trackers with an online update module can effectively adapt to constantly changing objects, so it is imperative to ensure that credible prediction samples are utilized to refine the tracking model online. Hence, we further devise an uncomplicated position offset constraint method based on target motion inertia to screen more reliable prediction results. Adequate experimental results reveal the validity of the color prompt encoder and position offset constraint in the DCF tracking framework. Our trackers can perform favorably against recent and far more sophisticated trackers on multiple public benchmarks. Concretely, our proposed tracker achieves a 0.815 robustness and 0.305 expected average overlap (EAO) on Visual Object Tracking (VOT) 2020 dataset, which is superior to the baseline in robustness (+2.6 %) and EAO (+0.8 %).}, archive = {J_EAAI}, author = {Xuedong He and Huiying Xu and Xinzhong Zhu and Hongbo Li and Xiao Huang and Yunliang Jiang}, doi = {10.1016/j.engappai.2025.111415}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111415}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Learning color prompt and position constraint for visual tracking}, volume = {157}, year = {2025}, }
• (2025). Optimization and classification of thermal transport on a convective surface with non-uniformly shaped ternary hybrid nanofluid flows. EAAI, 157, 111391. (www)

  @article{ , abstract = {Research background This research makes it possible to classify heat transport precisely in sophisticated cooling systems, which improves performance in energy, electronics, and aerospace applications. Integrating Support Vector Machine Learning classification and ternary hybrid nanofluids offers innovative solutions for efficient heat dissipation in complex three-dimensional geometries. Research problem /issue The current model investigates the effects of radiation, Casson, magnetic field, and non-Fourier heat flux on the behaviour of seawater-based hybrid nanofluid fluid flow dynamics and heat transfer using ternary hybrid nanoparticles, such as aluminium alloy (spherical shape), singular-walled carbon nanotubes (cylindrical shape), and multi-walled carbon nanotubes (platelet shape) in the convective surface. Method Governing partial differential equations are transformed into ordinary differential equations by self-similar transformation; solutions for the nonlinear system are obtained via a boundary value problem solver using numerical techniques Boundary Value Problem 5th-order Collocation method in matrix laboratory software. Results and conclusion Support vector machine learning was able to categorize heat transfer performance according to essential contributing elements for the suction and injection scenarios; the effect of non-dimensional flow parameters on flow profiles is shown graphically and tabularly, as the radiation parameter R increases, ternary hybrid nanofluid continuously exhibits a heat transfer rate that is roughly 46–47 % higher than hybrid nanofluid flow, industrial applications of the present study are Paper manufacture, plastics, ceramics, treatments, the automotive sector, cosmetics, and medicines, just a few of the many technical uses for these nanoparticles.}, archive = {J_EAAI}, author = {Maddina Dinesh Kumar and Nehad Ali Shah and Gurram Dharmaiah and Se-Jin Yook}, doi = {10.1016/j.engappai.2025.111391}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111391}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimization and classification of thermal transport on a convective surface with non-uniformly shaped ternary hybrid nanofluid flows}, volume = {157}, year = {2025}, }
• (2025). Distributed fault-tolerant for leader-following multi-unmanned aerial vehicle systems with faulty sensors based on belief rule base. EAAI, 157, 111388. (www)

  @article{ , abstract = {Effective fault-tolerant control (FTC) for leader-following Multi-Unmanned Aerial Vehicle (Multi-UAV) systems is crucial but challenging, especially when dealing with sensor faults in a distributed manner, including complex scenarios like overlapping failures leading to joint unobservability across multiple UAVs. This paper proposes a novel distributed FTC framework leveraging Belief Rule Base (BRB) combined with communication fusion (BRB-CF) to address these challenges. The framework integrates two key modules operating on each UAV: a decentralized BRB-based fault diagnosis (FD) module to accurately identify faulty sensor locations using local data, and a distributed reconstruction module utilizing BRB-CF. For non-overlapping faults, a communication-based fusion function compensates failures using neighbor data. Crucially, for overlapping faults where fusion is insufficient, a dedicated BRB-CF model reconstructs sensor outputs by integrating historical data with fused neighbor information. Simulation results demonstrate the proposed framework’s effectiveness, enabling the multi-UAV system to maintain formation and performance stability under both non-overlapping and overlapping sensor fault conditions, thereby enhancing system robustness and resilience.}, archive = {J_EAAI}, author = {Ruohan Yang and Xiao Gong and Zhichao Feng and Yahui Hao}, doi = {10.1016/j.engappai.2025.111388}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111388}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Distributed fault-tolerant for leader-following multi-unmanned aerial vehicle systems with faulty sensors based on belief rule base}, volume = {157}, year = {2025}, }
• (2025). Uncertainty-aware deep variational attention network: A trustworthy mechanical fault diagnostic model assisted by out-of-distribution detection. EAAI, 157, 111386. (www)

  @article{ , abstract = {Most attention mechanisms used for fault diagnosis are deterministic attention with fixed attention weights and thus can only provide overconfident point estimate predictions. When handling out-of-distribution (OOD) inputs, they typically produce untrustworthy diagnostic decisions that incorrectly classify them into any seen fault classes. Unlike deterministic approaches, variational attention treats attention weights as latent random variables drawn from their posterior distribution, enabling uncertainty estimation to reject OOD samples. However, current implementations of variational attention typically require customized structural modifications to original deterministic attention, hindering deployment across different attention architectures. Moreover, existing variational attention primarily focuses on self-attention, with limited exploration in other types of attention. To address these limitations, this paper proposes: (1) an architecture-agnostic variational attention applicable to various popular deterministic attention; (2) a Monte Carlo shaping layer ensuring attention weights’ diversity for reliable uncertainty quantification; (3) a model uncertainty estimation scheme detecting distribution shifts. These components form an uncertainty-aware deep variational attention network for OOD detection-assisted trustworthy mechanical fault diagnosis. The experimental results show that the proposed method not only gives accurate predictions for in-distribution samples but also effectively detects OOD samples.}, archive = {J_EAAI}, author = {Yiming Xiao and Haidong Shao and Haomiao Zhang and Rongming Wei and Bin Liu}, doi = {10.1016/j.engappai.2025.111386}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111386}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Uncertainty-aware deep variational attention network: A trustworthy mechanical fault diagnostic model assisted by out-of-distribution detection}, volume = {157}, year = {2025}, }
• (2025). Person re-identification based on an improved DeepLabv3+ semantic segmentation network. EAAI, 157, 111384. (www)

  @article{ , abstract = {Pedestrian re-identification is the process of recognizing and matching the same pedestrian in surveillance videos across different cameras or different time periods, which has been widely used in the field of intelligent security and security tracking. However, if the background is complex or persons are occluded, it often results in low accuracy. To solve these problems, we propose a person re-identification method based on improved DeepLabv3+ semantic segmentation, which mainly consists of two parts: semantic segmentation network and person re-identification network. Firstly, we propose an U-DeepLabv3+ semantic segmentation network by combing the advantages of the U-shaped Network(U-Net) and DeepLab version plus network(DeepLabv3+) to generate precise pedestrian masks, which can effectively eliminate background interference and provide high-quality inputs for subsequent person re-identification. Secondly, we propose an improved part-based convolutional baseline person re-identification network (IPCBReID), which leverages these masks to strengthen feature extraction and integrates an attention mechanism along with a "One-vs.-Rest" relational module to achieve deeper feature integration. Extensive experiments conducted on the Market-1501, Chinese University of Hong Kong (CUHK03), and Occluded-Duke datasets demonstrate the effectiveness of our approach. Compared with state-of-the-art methods, our IPCBReID achieves improvements in mAP by 0.3 %, 0.5 %, and 0.3 % on above three datasets respectively, and increases Rank-1 accuracy by 1.5 % and 1.6 % on Market-1501 and Occluded-Duke. These results validate the effectiveness of our method in enhancing person re-identification accuracy.}, archive = {J_EAAI}, author = {Fuzhen Zhu and Donglin Li and Huiling Li and Yong Liu and Bing Zhu}, doi = {10.1016/j.engappai.2025.111384}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111384}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Person re-identification based on an improved DeepLabv3+ semantic segmentation network}, volume = {157}, year = {2025}, }
• (2025). A modeled study of driver visual attention driven by driving tasks. EAAI, 157, 111382. (www)

  @article{ , abstract = {Visual attention is an indispensable component of driving, enabling drivers to swiftly identify critical objects within complex and dynamic traffic environments. Despite its significance, existing visual attention models predominantly focus on static or idealized driving scenarios, limiting their ability to capture attention distribution patterns in real-world, dynamic environments. Furthermore, most of these models rely heavily on data-driven approaches, extracting features exclusively from visual image data, while neglecting the profound influence of “the driver, the vehicle, and the road environment”. Consequently, these models frequently fail to effectively address the intricacies of practical driving scenarios. To bridge these gaps, this study introduces a driver visual attention prediction model that comprehensively incorporates the driving task, driver experience, and the impact of dynamic visual scenes. The proposed model leverages the advanced learning capabilities of Convolutional Neural Networks (CNN) and Vision Transformer (ViT), coupled with sequence modeling mechanisms, to effectively capture the nuanced attention allocation patterns of drivers in complex driving contexts. The model is meticulously designed to adapt to dynamically evolving driving task requirements. Experimental results demonstrate that the proposed model outperforms state-of-the-art (SOTA) visual attention prediction models across multiple benchmark evaluation metrics on the DR(eye)VE dataset, particularly excelling in dynamic driving conditions. Moreover, generalization experiments were conducted on the BDD-A and TDV datasets validate the model’s robustness and applicability across varied driving tasks and dynamic conditions.}, archive = {J_EAAI}, author = {Chuan Xu and Bo Jiang and Yukun Wang and Yan Su}, doi = {10.1016/j.engappai.2025.111382}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111382}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A modeled study of driver visual attention driven by driving tasks}, volume = {157}, year = {2025}, }
• (2025). Fuzzy clustering-based dual-channel contrastive learning for multi-behavior recommendation. EAAI, 157, 111381. (www)

  @article{ , abstract = {In recommender systems, diverse user behaviors (e.g., clicking, purchasing, sharing) provide valuable insights into user preferences. While multi-behavior recommendation models have shown promise, existing models often suffer from excessive complexity or fail to effectively capture relationships between behaviors. Two major challenges persist: (1) Most models focus primarily on user–item interactions, overlooking the dominant role of content data in real-world applications. Additionally, the high proportion of non-interacted items exacerbates the sparsity of target behavior data. (2) Current methods jointly model users and items but fail to explicitly distinguish their unique characteristics, leading to an incomplete understanding of diverse item behaviors. To address these issues, we propose Fuzzy Clustering-based Dual-Channel Contrastive Learning(a commonly used algorithm in artificial intelligence) (FCCL) model for multi-behavior recommendation. FCCL first employs a graph convolutional network to generate user and item embeddings independently, leveraging contrastive learning (CL) to capture explicit and implicit features. Subsequently, a dual-channel linear module based on fuzzy clustering is introduced to model both user interest diffusion and item provider influence. In the first layer, a user-level fuzzy clustering CL method is proposed to capture user–item similarities through a fused loss function. The second layer applies item-level hard clustering to characterize item-entity relationships, mitigating sparsity by identifying relevant items, including non-interacted ones. Finally, these tasks are integrated to enhance the quality of user and item embeddings, and a dual-channel optimization mechanism is established to optimize model parameters. Extensive experiments conducted on several public datasets demonstrate that FCCL significantly outperforms existing multi-behavior recommendation models in terms of effectiveness.}, archive = {J_EAAI}, author = {Juan Liao and Aman Jantan and Zhe Liu and Tapan Senapati and Gözde Ulutagay and Laith Abualigah and Omed Hassan Ahmed}, doi = {10.1016/j.engappai.2025.111381}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111381}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzy clustering-based dual-channel contrastive learning for multi-behavior recommendation}, volume = {157}, year = {2025}, }
• (2025). Power generation forecasting based on hybrid deep learning models for a multi-energy power generation system. EAAI, 157, 111380. (www)

  @article{ , abstract = {The accurate power generation forecasting (PGF) of the multi-energy complementary power generation system (MECPG) is great significance which lays the foundation for coordinated energy matching and optimization. This paper proposed a hybrid artificial intelligence (AI) model for the forecasting of the output in the multi-energy complementary power generation system which is designed in the prior research and the dataset primarily comprises meteorological in Lhasa, Tibet, China and operational parameters which were collected per hour, totaling 4461 samples. The hybrid model comprises the cross-correlation Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CC-ICEEMDAN) method for denoising, Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) for forecasting. This structure captures multi-dimensional features to boost the accuracy and develop learning speed and robustness. Compared with other five typical deep learning models, the proposed model is the most suitable for PGF of MECPG with Mean Squared Error (MSE) of 0.0005, Mean Absolute Percentage Error (MAPE) of 2.8, Mean Absolute Error (MAE) of 0.014, and Root Mean Squared Error (RMSE) of 0.02. This study provides a theoretical basis for the realization of artificial intelligence application which involves the real-time peak forecasting and optimal allocation for MECPG.}, archive = {J_EAAI}, author = {Ying Song and Jianqiu Zhou and Yu Peng and Jin Yan}, doi = {10.1016/j.engappai.2025.111380}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111380}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Power generation forecasting based on hybrid deep learning models for a multi-energy power generation system}, volume = {157}, year = {2025}, }
• (2025). A heterogeneous graph attention-enhanced deep reinforcement learning framework for flexible job shop scheduling problem with variable sublots. EAAI, 157, 111375. (www)

  @article{ , abstract = {Variable lot-sizing is an effective approach to improve production efficiency by splitting an operation into several sublots, which has been widely applied in flexible manufacturing systems. However, the flexibility of lot-sizing will dramatically expand the solution space, leading to excessive computation time in converging to the relative optimum. To address this challenge, this paper introduces an end-to-end deep reinforcement learning framework based on heterogeneous graph attention mechanisms (HGADRL) for flexible job shop scheduling problem with variable sublots. Unlike traditional heuristic and rule-based methods, HGADRL dynamically learns the high-dimensional nature, providing a more generalizable solution in a very short time. In HGADRL, a modified heterogeneous disjunctive graph is designed to represent the dynamic scheduling status, including operation selection and sublot division. A dual-scale graph attention network combined with two interconnected attention modules is developed, enabling the precise capture of complex interdependencies between heterogeneous vertices. This approach can significantly enhance the agent's ability to self-learn and evolve optimal policies. By leveraging local and global features extracted through the graph attention network, an actor-critic network is employed for high-quality scheduling in different states. Experimental results demonstrate that the proposed method outperforms the 12 mixed priority dispatching rules, two meta-heuristic methods and two deep reinforcement learning methods in all 500 synthetic instances. Additionally, the proposed method outperforms all compared methods across 16 unseen scales of instances and four real-world instances, demonstrating its strong generalization capabilities.}, archive = {J_EAAI}, author = {Zipeng Yang and Xinyu Li and Liang Gao and Qihao Liu}, doi = {10.1016/j.engappai.2025.111375}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111375}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A heterogeneous graph attention-enhanced deep reinforcement learning framework for flexible job shop scheduling problem with variable sublots}, volume = {157}, year = {2025}, }
• (2025). Feature refinement attention U-net for enhanced liver segmentation. EAAI, 157, 111374. (www)

  @article{ , abstract = {Developing efficient and accurate automated liver segmentation methods is of far-reaching significance for the prevention, diagnosis, treatment, and prognostic assessment of liver diseases. Artificial intelligence (AI) technology has shown immense potential in various computer vision tasks and has been increasingly adopted in medical image segmentation in recent years. However, conventional neural networks often exhibit limitations in capturing global context capture, constraining their performance in medical image segmentation. To address this challenge, we propose a Feature Refinement Attention U-Net (FRAttU-Net). The architecture is designed with two key components: a High and Low Frequency Feature Fusion and Enhancement Module (HLF 3 EM) between the encoder and decoder to integrate the multilevel features and enhance the high-level feature representation using edge cues, and an Otsu Density Refinement Module in the output layer to strengthen density information representation and improve the prediction accuracy. Extensive experiments demonstrate that FRAttU-Net achieves state-of-the-art accuracy, outperforming existing methods on multiple datasets. This performance advancement holds considerable potential in improving clinical decision-making in liver disease diagnosis and treatment planning. The code is available at: https://github.com/HUI623/ FRAttU-Net.}, archive = {J_EAAI}, author = {Hui Wang and Desheng Liu and Zhilei Zhao and Yue Wu and Anil Baris Cekderi}, doi = {10.1016/j.engappai.2025.111374}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111374}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature refinement attention U-net for enhanced liver segmentation}, volume = {157}, year = {2025}, }
• (2025). Feature-refined adaptive modulation transformer for image deraining. EAAI, 157, 111373. (www)

  @article{ , abstract = {Recent image deraining methods demonstrate impressive reconstruction performance by leveraging the global modeling capability of Transformer architecture. However, unlike convolutional approach, Transformer inherently struggles to capture high-frequency detail effectively. Furthermore, existing methods primarily focus on spatial information while largely neglecting the frequency-domain characteristics of rain streaks, which are crucial for rain removal. To address these challenges, we propose a feature-refined adaptive modulation Transformer (FRAMT), which effectively integrates spatial-domain features with frequency-domain modulation to enhance deraining performance. To accurately identify rain streaks and efficiently separate them from the background, the detail-guided attention block enhances sensitivity to high-frequency components by integrating pooling operation with convolution. To mitigate image blurring and detail loss induced by rain streaks, the local feature refinement block employs a multi-scale content decomposition strategy, utilizing a parallel multi-branch architecture to extract diverse contextual features across varying spatial scales. Additionally, the adaptive fusion modulation block incorporates a frequency selection mechanism that dynamically modulates feature response, effectively suppressing redundant information and irrelevant features. Extensive experiments conducted on widely used benchmark datasets demonstrate that the proposed method is more competitive than advanced methods.}, archive = {J_EAAI}, author = {Yeting Huang and Lei Dai and Zhihua Chen and Wenlong Hu and Shouli Wang}, doi = {10.1016/j.engappai.2025.111373}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111373}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature-refined adaptive modulation transformer for image deraining}, volume = {157}, year = {2025}, }
• (2025). A multi-head deep fusion model for recognition of cattle foraging events using sound and movement signals. EAAI, 157, 111372. (www)

  @article{ , abstract = {Monitoring feeding behaviour is a relevant task for efficient herd management and the effective use of available resources in grazing cattle. The ability to automatically recognise animals’ feeding activities through the identification of specific jaw movements allows for the improvement of diet formulation, as well as early detection of metabolic problems and symptoms of animal discomfort, among other benefits. The use of sensors to obtain signals for such monitoring has become popular in the last two decades. The most frequently employed sensors include accelerometers, microphones, and cameras, each with its own set of advantages and drawbacks. An unexplored aspect is the simultaneous use of multiple sensors with the aim of combining signals in order to enhance the precision of the estimations. In this direction, this work introduces a deep neural network based on the fusion of acoustic and inertial signals, composed of convolutional, recurrent, and dense layers. The main advantage of this model is the combination of signals through the automatic extraction of features independently from each of them. The model has emerged from an exploration and comparison of different neural network architectures proposed in this work, which carry out information fusion at different levels. Feature-level fusion has outperformed data and decision-level fusion by at least a 0.14 based on the F1-score metric. Moreover, a comparison with state-of-the-art machine learning methods is presented, including traditional and deep learning approaches. The proposed model yielded an F1-score value of 0.802, representing a 14% increase compared to previous methods. Finally, results from an ablation study and post-training quantisation evaluation are also reported.}, archive = {J_EAAI}, author = {Mariano Ferrero and José O. Chelotti and Luciano S. Martinez-Rau and Leandro D. Vignolo and Martín Pires and Julio R. Galli and Leonardo L. Giovanini and H. Leonardo Rufiner}, doi = {10.1016/j.engappai.2025.111372}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111372}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-head deep fusion model for recognition of cattle foraging events using sound and movement signals}, volume = {157}, year = {2025}, }
• (2025). Cross-domain intelligent cooperative spectrum sensing algorithm based on federated learning and swin-transformer neural network. EAAI, 157, 111370. (www)

  @article{ , abstract = {The deployment of 6th Generation (6G) mobile networks will bring revolutionary changes to organizations and users through higher speeds, lower latency, and more intelligent network management. This will require more efficient spectrum sharing to improve network performance. Cognitive Radio (CR) can be employed to help achieve this goal. However, data collection at the fusion center for cooperative spectrum sensing (CSS) creates a risk of sensing user (SU) data leakage. In this paper, an intelligent CSS algorithm for mobile communication is proposed based on Federated Learning with a Swin-Transformer. First, to achieve the cross-domain process, a Continuous Wavelet Transform (CWT) is used with the normalized time series data to obtain a time-frequency spectrum map with joint time-frequency features. Second, Federated Learning is employed to realize distributed CSS to reduce sensing overhead, improve the security of sensing data, and eliminate the SU data isolation with traditional distributed CSS. Then, a new Swin-Transformer spectrum sensing is proposed for fusion learning of the time-frequency spectrum map. The fusion of Federated Learning and Swin-Transformer neural network provides cooperation with multiple SUs to achieve efficient spectrum sensing using the K -rank fusion criterion. Compared with the Convolutional Neural Network (CNN), ViT-Transformer, Graph Neural Network (GNN) and EfficientNet algorithms, the detection probability of the proposed algorithm has increased by 45 %, and the false alarm probability has decreased by 13 %.}, archive = {J_EAAI}, author = {Lingwei Xu and Zhihe Gao and Yufang Li and T. Aaron Gulliver}, doi = {10.1016/j.engappai.2025.111370}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111370}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-domain intelligent cooperative spectrum sensing algorithm based on federated learning and swin-transformer neural network}, volume = {157}, year = {2025}, }
• (2025). Quantum-like evidence networks decision-making model. EAAI, 157, 111368. (www)

  @article{ , abstract = {For the purpose of addressing the deviations of the Sure Thing Principle observed in cognitive experiments, numerous artistic models have been developed and refined within the probabilistic framework. In spite of this, it is clear that these models still have room for improvement in terms of effectively expressing and processing global ignorance information and local ignorance information. As a result, a quantum-like evidence networks decision-making model (QLENDM) is proposed. The work we operates within the framework of evidence theory, rather than probability theory, which is different from those art models, such as quantum dynamics Markov model, quantum-like approach, quantum prospect decision theory, and quantum-like Bayesian networks. Quantum-like basic probability assignment can better model the global ignorance or local ignorance information presented by cognitive experiments than the probability distribution, which solves the issue of inadequate modeling of uncertain information within a probabilistic framework. Moreover, from a heuristic perspective, QLENDM automatically fits the parameters related to interference effect using the distance of quantum-like basic probability assignments based on focal element structure. Therefore, QLENDM with these two characteristics, in addition to its universality, has better predictive performance than the above art models. We then apply QLENDM to cognitive science and information retrieval. As a result of the experiment, it appears that compared with other models, QLENDM has the minimum average and standard deviation of fitting errors, indicating that it is better suited for addressing deviations of the Sure Thing Principle as well as predicting the behavior of participants more accurately in decision-making experiments.}, archive = {J_EAAI}, author = {Xiaozhuan Gao and Huijun Yang and Lipeng Pan and Danilo Pelusi}, doi = {10.1016/j.engappai.2025.111368}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111368}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Quantum-like evidence networks decision-making model}, volume = {157}, year = {2025}, }
• (2025). Portable fair decision making through modular approach. EAAI, 157, 111361. (www)

  @article{ , abstract = {Influenced by real-world biases, machine learning based decision systems are prone to discriminatory outcomes, which has garnered considerable attention, and led to the proposal of numerous bias mitigation methods, such as adversarial debiasing and fair representation learning. However, relying on predefined fairness standards prevents these methods from adapting to evolving fairness requirements. To address this issue, we propose a modular fairness enhancement approach. In our approach, each fairness requirement is modeled as a distinct optimization task, with the corresponding model parameters encapsulated within an independent sub-module. In addition, a main module is designed to capture the shared classification features across all fairness tasks. This multi-task learning architecture enables the decision making system to meet multiple fairness requirements without retraining. Experimental evaluations on four real datasets, by comparing portability and four fairness metrics with state-of-the-art methods. The results demonstrate that our method achieves superior portability in addressing various fairness requirements compared to the existing methods.}, archive = {J_EAAI}, author = {Fengyu Wu and Ayong Ye and Qiuling Chen and Huang Zhang and Jing Chen}, doi = {10.1016/j.engappai.2025.111361}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111361}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Portable fair decision making through modular approach}, volume = {157}, year = {2025}, }
• (2025). A classification method for online consultation on civil disputes based on deep transfer learning with small datasets. EAAI, 157, 111360. (www)

  @article{ , abstract = {With the widespread use of artificial intelligent systems to resolve civil disputes, the classification function of online consultation intelligence systems has become increasingly significant, providing accurate knowledge support for later decision-making. However, the informality of online consultation and cross-platform data decentralization, together with the small amount of data, cause typical machine learning approaches to perform poorly when dealing with such data. In light of this, this study provides a deep transfer learning-based method to solving the challenge of classifying online consultation for civil disputes with unbalanced and small datasets via cross-platform knowledge transfer. First, ontology design and modeling are integrated to normalize cross-domain feature representations and reduce the impact of linguistic ambiguity on transfer results. Second, improved marginal fisher analysis is applied to improve the model's performance of intra-class compactness and inter-class separation. Then, joint distribution adaptation with conditional feature mapping is proposed to alleviate the adverse impact of outliers on transfer learning using a conditional feature mapping mechanism. Finally, online datasets on Chinese civil disputes are used to validate the method. The results indicate that the method is more effective than other popular machine learning techniques with evaluation metrics showing accuracy 84.17 % and recall 84.40 %. The study can provide technical support for artificial intelligence systems that lack data deposition and has practical significance for improving the efficiency of civil dispute mediation.}, archive = {J_EAAI}, author = {Ning Wang and Shibo Cui and Enhui Zhao and Baiji Jin}, doi = {10.1016/j.engappai.2025.111360}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111360}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A classification method for online consultation on civil disputes based on deep transfer learning with small datasets}, volume = {157}, year = {2025}, }
• (2025). Remaining useful life transfer prediction integrated with time-varying operation condition decoupling transformation. EAAI, 157, 111355. (www)

  @article{ , abstract = {The Remaining Useful Life (RUL) prediction under time-varying operation conditions in both research and industry faces several challenges: (1) the speed signal is not always available in in-field equipment; (2) health indicators (HIs) under time-varying rotational frequencies exhibit fluctuations, which can lead to misjudgments in component failure; and (3) with fluctuating HIs resulting from time-varying operation conditions, accurately predicting the RUL becomes difficult. To address these issues, this paper proposes a novel RUL transfer prediction framework which integrates an Operation Condition Decoupling Transformation (OCDT) method. A Spectral Peak Query-based Harmonic Extraction method is firstly proposed to extract the rotational frequency harmonic. The OCDT method is then introduced to eliminate the impact of time-varying speed on the HIs and transform the HIs into an equivalent constant operation condition based on the estimated rotational frequency. A Domain Dual Adaption Transfer Prediction method with Adaptive Dynamic Weighted Loss Function is proposed to conduct cross-domain RUL prediction between the target domain of equivalent operation condition and the source domain of known operation condition. The real bearing experiments are conducted to validate the effectiveness and superiority of the proposed framework.}, archive = {J_EAAI}, author = {Lei Yang (杨磊) and Yuhe Liao (廖与禾) and Tuojian Li (李拓建) and Tao Kang (康涛) and Peng Sun (孙鹏)}, doi = {10.1016/j.engappai.2025.111355}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111355}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Remaining useful life transfer prediction integrated with time-varying operation condition decoupling transformation}, volume = {157}, year = {2025}, }
• (2025). MA-net: A multi-attention segmentation network for anisotropic prostate magnetic resonance images. EAAI, 157, 111348. (www)

  @article{ , abstract = {Prostate cancer (PCa) is the most common cancer among American men. Accurate and reliable segmentation of the prostate using magnetic resonance (MR) imaging is critical for the diagnosis and treatment of prostate cancer. Due to the inherent characteristics of prostate MR images, such as anisotropic resolution, large changes in shape and size, and low contrast between the gland and surrounding structures, the performance of traditional two-dimensional and three-dimensional based segmentation methods is limited. To address these issues, we propose a multi-attention-guided hierarchical U-shaped encoder-decoder network (MA-Net) to perform accurate and robust segmentation of prostate MR images. Specifically, an explicit decay attention (EDA) module is first used to introduce spatial distance prior knowledge into the encoding process and perform multi-scale token aggregation operations to generate rich feature representations. Secondly, a multiple cross-slice attention (MCSA) module is introduced in the skip connection. By applying multiple attention mechanisms to depth feature maps of different scales, all cross-slice information on anisotropic MR images is learned, thereby improving the segmentation performance at various positions of the prostate. Finally, the global-local interaction (GLI) module, which can interactively fuse long-range dependencies and local context information, is used to obtain a more accurate global-local feature representation and further improve the accuracy of prostate segmentation. Experimental results on three public prostate MR image datasets fully demonstrate the effectiveness and advancement of our method. In future work, we plan to integrate self-supervised pre-training, cross-modal adaptive fusion, and domain generalization methods to enhance the generalization ability of MA-Net for multi-modal and multi-center data.}, archive = {J_EAAI}, author = {Zhenglin Yi and Jinbo Chen and Zhiyong Cai and Jiatong Xiao and Jinhui Liu and Haisu Liang and Chao Quan and Xiongbing Zu and Longxiang Wu and Jiao Hu}, doi = {10.1016/j.engappai.2025.111348}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111348}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MA-net: A multi-attention segmentation network for anisotropic prostate magnetic resonance images}, volume = {157}, year = {2025}, }
• (2025). Combating false negatives in pancreatic cancer: A deep learning approach for aiding fine needle aspiration via accurate subregion identification. EAAI, 157, 111347. (www)

  @article{ , abstract = {Accurate pancreatic cancer diagnosis based upon Computed Tomography (CT) guided Fine needle aspiration (FNA), crucially depends upon segmentation and classification of cancerous subregions (head, body, and tail). This experiment proposes an Artificial Intelligence (AI) driven deep learning framework that integrates novel Pancreatic U-Network (PanUNet) for pancreatic subregion segmentation and Residual Network (ResNet50) with Squeeze-and-Excitation (SE) blocks for classification. The AI model was trained with 2895 slices and refined through data augmentation techniques. The segmentation performance was assessed with dice similarity coefficient, intersection over union, sensitivity, and specificity, whereas F1-score, precision, recall and root mean squared error were used to evaluate classification performance. The model achieved 96.46 % dice similarity coefficient and 98.96 % classification accuracy. The experimental results demonstrate enhanced feature extraction and improved classification accuracy with SE block integration. Compared to individual optimizers, a Mixed-Adaptive moment estimation- Root mean square propagation- Stochastic gradient descent (MARS) optimization technique aided in achieving superior performance in the proposed framework. An extensive comparative analysis of the proposed model against established methods showcasing significant improvements in segmentation and classification proves its potential for clinical applicability.}, archive = {J_EAAI}, author = {Ms Jasmine Chhikara and Nidhi Goel and Neeru Rathee}, doi = {10.1016/j.engappai.2025.111347}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111347}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Combating false negatives in pancreatic cancer: A deep learning approach for aiding fine needle aspiration via accurate subregion identification}, volume = {157}, year = {2025}, }
• (2025). The pseudo-analytical probability solution to parameterized Fokker–Planck equations via deep learning. EAAI, 157, 111344. (www)

  @article{ , abstract = {Efficiently solving Fokker–Planck equations (FPEs) is crucial for understanding the probabilistic evolution of stochastic particles in dynamical systems, however, analytical solutions are only attainable in specific cases. To speed up the solving process of parameterized FPEs with several system parameters, we introduce a deep learning-based method to obtain the pseudo-analytical probability solution (PAPS). Unlike previous numerical methodologies that necessitate solving the FPEs separately for each set of system parameters, the PAPS simultaneously addresses all FPEs within a predefined continuous range of system parameters during a single training phase. The approach utilizes a Gaussian mixture distribution (GMD) to represent the stationary probability density functions, namely, the solutions to FPEs. By leveraging a deep residual network, each parameter configuration of the system is mapped to the parameters of the GMD, ensuring that the weights, means, and variances of Gaussian components adaptively align with the corresponding true density functions. A grid-free algorithm is further developed to effectively train the residual network, resulting in a feasible PAPS obeying nonnegativity, normalization and boundary conditions. Extensive numerical studies validate the accuracy and efficiency of our method. This approach presents new insight to the pseudo-analytical solutions to FPEs, and promises significant acceleration in the response analysis of multi-parameter, multi-dimensional stochastic nonlinear systems.}, archive = {J_EAAI}, author = {Xiaolong Wang and Jing Feng and Gege Wang and Tong Li and Yong Xu}, doi = {10.1016/j.engappai.2025.111344}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111344}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The pseudo-analytical probability solution to parameterized Fokker–Planck equations via deep learning}, volume = {157}, year = {2025}, }
• (2025). Generative agent-based modeling with large language models for insider threat detection. EAAI, 157, 111343. (www)

  @article{ , abstract = {Insider threats pose a critical challenge in cybersecurity, as individuals within organizations misuse legitimate access to compromise sensitive systems and data. Traditional detection methods often struggle with the complexity of such threats, while Deep Learning (DL) approaches face issues like overfitting and lack of interpretability. To address these limitations, we propose a Generative Agent-Based Modeling (GABM) framework that integrates Large Language Models (LLMs) with a hierarchical multi-agent system. Our framework employs Specialized Agents to process categorized log files and generate detailed reports, which are synthesized by a Supervisor Agent for final activity classification. We validated this approach on both network-centric (PicoDomain) and behavior-rich (CERT r5.2) datasets, demonstrating its ability to handle diverse logs, model complex threats, and generalize across insider risk scenarios. The framework outperformed existing baselines, prioritizing high recall to minimize false negatives—crucial in cybersecurity contexts. While precision was comparatively lower, this trade-off supports early threat detection. An ablation study highlighted the importance of the Supervisor Agent, whose removal led to a significant drop in performance and increased false positives. These results demonstrate the potential of LLM-powered hierarchical multi-agent frameworks for scalable, interpretable, and reliable insider threat detection. Our contributions include the integration of GABM and LLMs, a hierarchical system for log analysis, and the use of Chain-of-Thought reasoning for enhanced interpretability.}, archive = {J_EAAI}, author = {Antonino Ferraro and Gian Marco Orlando and Diego Russo}, doi = {10.1016/j.engappai.2025.111343}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111343}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generative agent-based modeling with large language models for insider threat detection}, volume = {157}, year = {2025}, }
• (2025). Advanced deep learning-based electricity theft detection in smart grids using multi-dimensional analysis with convolutional autoencoder and transformer. EAAI, 157, 111333. (www)

  @article{ , abstract = {Electricity theft poses a significant threat to power grid stability, leading to substantial economic losses and safety hazards for utility providers. Therefore, effective electricity theft detection (ETD) is critical for ensuring the cost-efficiency and security of smart grids. While machine learning (ML) techniques utilizing electricity consumption (EC) data from smart meters have shown potential, traditional methods often rely on single-dimensional analysis, lacking the granularity to capture complex, periodic EC patterns, which results in suboptimal detection performance. To address these shortcomings, we propose a novel ETD system that utilizes deep learning (DL) through a combination of Convolutional Autoencoder (CAE) and Transformer models for advanced multi-dimensional analysis of EC data. We further tackle EC data imbalance with the K-means-based Synthetic Minority Oversampling Technique (K-means SMOTE). Our approach preprocesses EC data, employing the CAE to extract detailed features from 28-day consumption intervals, followed by the Transformer to analyze sequences and capture comprehensive patterns. The proposed ETD enables precise differentiation between anomalous, non-periodic theft signatures and consistent, periodic, legitimate usage. Evaluated on the State Grid Corporation of China dataset, our proposed model achieves an impressive F1 score of 0.9918, surpassing existing benchmarks. Notably, its lightweight design ensures scalability and efficiency, broadening its applicability to other smart grid anomaly detection tasks, such as identifying equipment failures or cyber threats, thus reinforcing modern power grid security.}, archive = {J_EAAI}, author = {Tien-Dat Le and Nguyen Thanh Minh Duy and Truong Hoang Bao Huy and Pham Van Phu and Hien Thanh Doan and Daehee Kim}, doi = {10.1016/j.engappai.2025.111333}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111333}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advanced deep learning-based electricity theft detection in smart grids using multi-dimensional analysis with convolutional autoencoder and transformer}, volume = {157}, year = {2025}, }
• (2025). Dual self-report foot specific outcome measurements of patients with hallux valgus based on plantar pressures. EAAI, 157, 111329. (www)

  @article{ , abstract = {This study explores the integration of pressure sensor data and self-reported outcomes from the Manchester-Oxford Foot Questionnaire (MOXFQ) and the Foot Function Index 5-point (FFI-5pt) to assess Hallux Valgus severity. Forty-six adult patients participated, with pressure footprints captured using a force platform following a three-step protocol, yielding 15 footprints per participant. Shannon entropy analysis showed moderate unpredictability in both questionnaires, with MOXFQ exhibiting slightly more randomness. A hybrid deep learning model, incorporating convolutional (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) layers, was developed to classify data based on pressure readings and questionnaire responses. The model achieved high classification accuracy (90–95 %) after just nine repetitions, indicating that minimal data is sufficient for reliable performance. The relationship between data repetitions and accuracy shows diminishing returns, offering insights for optimizing data collection. By combining objective pressure footprint data with subjective self-reports, this study offers a robust, non-invasive approach for diagnosing and monitoring Hallux Valgus. The deep learning model improves clinical decision-making and reduces patient burden, particularly for those with cognitive disabilities. This AI-based method presents a promising alternative to traditional diagnostic techniques and emphasizes the potential of artificial intelligence (AI) in enhancing clinical practices that rely on subjective assessments. Future research should focus on refining the model, expanding datasets, and exploring broader applications to validate its robustness in diverse clinical settings.}, archive = {J_EAAI}, author = {Djordje Urukalo}, doi = {10.1016/j.engappai.2025.111329}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111329}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual self-report foot specific outcome measurements of patients with hallux valgus based on plantar pressures}, volume = {157}, year = {2025}, }
• (2025). Machine learning predicted thermodynamic properties of halocarbon- and hydrocarbon-based refrigerants. EAAI, 157, 111328. (www)

  @article{ , abstract = {By using artificial intelligence (AI) in machine learning (ML) algorithms, machines can automatically learn from data, identify patterns, and make judgments with minimal human intervention. The highly accurate ML model presented in this work aims to calculate the refrigerant properties (entropy, density, specific volume, fugacity, and internal energy) of halocarbons and hydrocarbons. Chlorofluorocarbon (CFC), hydrochlorofluorocarbon (HCFC), hydrofluorocarbon (HFC), and perfluorocarbon (PFC) are the four further classes of halocarbon-based refrigerants. A large number of data points (>400,000) at various pressures and temperatures were compiled to predict thermodynamic parameters. The compiled data were used in three portions: 64 % for training, 16 % for testing, and 20 % for model validation. Categorical boosting (Catboost), decision tree (DT), random forest (RF), gradient boosting decision tree (GBDT), light boost gradient machine (LBGM), and extreme gradient boosting (XGB) are the six machine learning models that were trained. The coefficient of determination (R 2 ), mean absolute percentage error (MAPE), and mean squared error (MSE) were used to evaluate the performance of the model. The best-fit models for fugacity, entropy, density, specific volume, and internal energy for hydrocarbons were found to be RF, RF, DT, RF, and RF, in that order. The best models for halocarbons were Catboost, DT, RF, RF, and Catboost, in that order. With MSE and MAPE values less than 0.01, all best-fit models produced R 2 values larger than 0.99. The significance and effect of each input variable on the output variables are displayed in correlation heatmaps and Shapley additive explanations (SHAP) explanations. DT model required minimum computation time (in seconds) for training and prediction (0.90 and 0.03, respectively). In contrast, computation time (in seconds) for training and prediction was found to be maximum (68.89 and 3.36, respectively) for the RF model. The experimental investigation of thermodynamic properties for novel refrigerants is expensive and time-consuming. This study shows the efficacy of ML in predicting features of hydro- and halo-based refrigerants.}, archive = {J_EAAI}, author = {Chandra Prakash Gupta and Vimal Chandra Srivastava and Akshay Dvivedi}, doi = {10.1016/j.engappai.2025.111328}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111328}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning predicted thermodynamic properties of halocarbon- and hydrocarbon-based refrigerants}, volume = {157}, year = {2025}, }
• (2025). A multi-scale feature and cross-domain fusion network for image tampering localization. EAAI, 157, 111325. (www)

  @article{ , abstract = {As a phenomenal communication application based on artificial intelligence forms, the potential social harm of image tampering is prominent. Despite the significant research achievements attained by existing works, the current methods fall short in adequately extracting and integrating features from different frequency domains, and they exhibit limitations when confronted with tampered images that have undergone post-processing operations. To address these problems, we propose a multi-scale features and cross-domain fusion network (MFCF-Net). First, the simple linear iterative clustering algorithm is adopted to achieve growth, merging, and segmentation within the spatial domain. Second, uneven blur features are captured by introducing a Local Self-Attention Mechanism (LSAM) block and an Edge Feature Enhancement Module (EFEM). Third, a complete cross-domain fused feature is obtained by combining a Multi-Scale Feature Fusion (MSFF) and a Cross-Domain Fusion Module (CDFM), thereby enhancing the module's ability to recognize subtle changes in the tampered area. Finally, the detection and localization of tampered regions are achieved by utilizing the dual-matching algorithm for multi-scale feature matching. Extensive experiments indicate that the MFCF-Net possesses higher detection accuracy on six benchmark datasets and exhibits excellent performance when dealing with tampered images that have undergone post-processing operations.}, archive = {J_EAAI}, author = {Fan Deng and Tao Liu and Haibo Yu and Ruitao Yang}, doi = {10.1016/j.engappai.2025.111325}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111325}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-scale feature and cross-domain fusion network for image tampering localization}, volume = {157}, year = {2025}, }
• (2025). Stochastic multiple criteria decision-making for configuration selection of electricity-hydrogen integrated energy systems with multiple stakeholder interest coordination. EAAI, 157, 111323. (www)

  @article{ , abstract = {The development of electricity-hydrogen integrated energy systems (IESs) is an important solution for promoting the transition to clean energy. Determining the optimal system configuration is challenging due to the involvement of multiple criteria and stakeholders. This paper introduces a robust evaluation framework for the configuration selection of park-level electricity-hydrogen IESs considering multiple stakeholder interests. Firstly, eight schemes are designed for an industrial park, and configuration optimization models are constructed to obtain the optimal capacities of the components. Then, a group decision-making model based on stochastic multiple criteria acceptability analysis and multi-objective optimization on the basis of ratio analysis plus full multiplicative form (MULTIMOORA) method is developed to analyze and rank schemes, which connects the optimization with decision-making under multiple factors. The proposed framework accommodates the imprecise preferences of multiple stakeholders and coordinates their interests from technological, economic, environmental, and social aspects. Additionally, compensatory effects between multiple criteria are modeled, enhancing the robustness of decision results. Results indicate that the combination of natural gas, photovoltaic panel, hydrogen storage, and gas boiler is the winning scheme, with an energy utilization ratio of 54 % and a 76 % reduction in carbon emissions. Economically, the scheme results in a levelized cost of energy of 0.201$/kWh and creates 28 jobs annually. Scenario analysis indicates that, under current high configuration costs, a single policy aimed at reducing hydrogen storage costs is insufficient to facilitate the widespread deployment of near-zero carbon emission electricity-hydrogen IESs. It is recommended that power restriction measures be implemented simultaneously.}, archive = {J_EAAI}, author = {Zhiying Zhang and Yingying Xu and Yaoyao Yu and Xiaoyuan Chen and Huchang Liao}, doi = {10.1016/j.engappai.2025.111323}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111323}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Stochastic multiple criteria decision-making for configuration selection of electricity-hydrogen integrated energy systems with multiple stakeholder interest coordination}, volume = {157}, year = {2025}, }
• (2025). A novel approach to wind speed modeling: A fast and robust model with high generalizability. EAAI, 157, 111316. (www)

  @article{ , abstract = {Surface wind speed forecasting is a critical challenge in various engineering fields, including aerospace, electrical, civil, environmental, mechanical, and agricultural engineering. This paper presents a novel artificial intelligence (AI)-based approach for short-term wind speed forecasting. First, a geographical area with diverse topographical features, distinct atmospheric conditions, and a specific time range is selected. A neighborhood-based feature set of wind speed values is used to extract the initial training dataset. A correlation-based pruning technique is applied to refine the dataset, ensuring it remains compact yet informative. Finally, a low-complexity machine learning model is employed for efficient forecasting. Following this approach, an initial dataset containing 29 features and approximately 90 million records was generated using the fifth-generation European Centre for Medium-Range Weather Forecast atmospheric reanalysis dataset (ERA5) for the Persian Gulf region (2001–2010). A pruning method based on Spearman's correlation reduced the dataset to fewer than 52,000 records (approximately 0.06 % of the original data). A two-layer artificial neural network was subsequently trained on the pruned dataset. To evaluate generalizability, the model was tested on 18 diverse test datasets. The results demonstrated successful wind speed prediction, with mean absolute error values ranging from 0.207 to 0.538 m per second (m/s) and root mean square error values from 0.280 to 0.738 m/s. These findings highlight the model's ability to forecast wind speed with minimal error across different regions and timeframes. The simplicity of the proposed methodology, combined with its low computational demands, positions it as a promising tool for real-world applications.}, archive = {J_EAAI}, author = {Alireza Hakimi and Parvin Ghafarian and Hossein Farjami}, doi = {10.1016/j.engappai.2025.111316}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111316}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel approach to wind speed modeling: A fast and robust model with high generalizability}, volume = {157}, year = {2025}, }
• (2025). Rock burst early warning utilizing acoustic emission and electromagnetic radiation adaptive noise reduction and multi-image feature fusion: A case study. EAAI, 157, 111312. (www)

  @article{ , abstract = {Acoustic emission and electromagnetic radiation, as the typical representatives of geophysical monitoring technology, can clearly reflect the precursor information of coal and rock failure. However, the existing rock burst early warning based on them is poor in noise processing and insufficient in feature mining, resulting in low generalization and accuracy of early warning. A novel method of rock burst early warning based on adaptive noise reduction and multi-image feature fusion is proposed. Firstly, Swarm Decomposition is used to decompose the acoustic emission and electromagnetic radiation intensity signals collected from working face, and each component after Kalman filtering is reconstructed to obtain the denoised signal. Then, the image features of Recurrence Plot, Gramian Angular Field, Motif Difference Field and Relative Position Matrix are extracted from the ten types of effective and interference signals respectively, and evaluate the distribution of image features of different scene signals. Next, the Non-Subsampled Contourlet Transformation fuses the preferred image and inputs it into the Vision Transformer model to obtain the identification results of normal, interference and precursor signals. A case study of a coal mine shows that the proposed method can identify various types of precursor response, and meet the early warning requirements under different induced factors such as first square, fault structure and periodic weighting. The average early warning accuracy of acoustic emission and electromagnetic radiation monitoring were 96.32 % and 96.01 % respectively, and the comprehensive accuracy reached 96.16 %, which can accurately identify the rock burst risk in advance and provide reliable guarantee for safe mining.}, archive = {J_EAAI}, author = {Shenglei Zhao and Enyuan Wang and Jinxin Wang and Qiming Zhang and Zhonghui Li and Jiankun Xu}, doi = {10.1016/j.engappai.2025.111312}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111312}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rock burst early warning utilizing acoustic emission and electromagnetic radiation adaptive noise reduction and multi-image feature fusion: A case study}, volume = {157}, year = {2025}, }
• (2025). Vehicle real-time collision risk prediction: A multi-modal learning approach for diverse urban road scenarios based on a large-scale near-crash event dataset. EAAI, 157, 111299. (www)

  @article{ , abstract = {The effectiveness of vehicle collision avoidance systems depends on the precision of collision risk prediction models. However, current models often neglect the driver's condition, resulting in their poor ability to predict near-crash events triggered by aggressive, fatigued, or distracted driving. Additionally, current models overlook the differences in modality, type, and variability of multi-source data, leading to insufficient feature extraction from input data, which in turn limits the model's prediction accuracy. To address these issues, we developed an end-to-end pre-trained deep framework (PM-Transformer) with a Transformer, consisting of multi-module recurrent convolutional neural networks. The framework includes four modules: (1) pre-trained time series module that extracts spatiotemporal information from traffic time series using one-dimensional convolutional neural network - long short-term memory; (2) pre-trained spectral module that learns visual temporal representations from traffic spectrograms using two-dimensional convolutional neural network - long short-term memory; (3) metadata module for vectorizing traffic metadata; (4) fusion module that semantically integrates features from the three modules using a Transformer. Results show that the proposed model can achieve the same prediction accuracy as other models using only 5 % of their training sample size. Compared to other traditional models, our model improves the accuracy of risk prediction by 6 %, 9 %, and 4 %, respectively, with small sample sizes (0.5 s, 1 s, and 2 s in advance), while also maintaining the best performance with larger sample sizes. Findings of this study hold significant potential for improving the effectiveness of vehicle collision avoidance systems.}, archive = {J_EAAI}, author = {Jipu Li ( Research Assistant ) and Yi He ( Professor ) and Ye Li ( Professor ) and Helai Huang ( Professor ) and Dan Wu ( Research Assistant ) and Jieling Jin ( Research Assistant )}, doi = {10.1016/j.engappai.2025.111299}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111299}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Vehicle real-time collision risk prediction: A multi-modal learning approach for diverse urban road scenarios based on a large-scale near-crash event dataset}, volume = {157}, year = {2025}, }
• (2025). Multi-objective decision-making method for road network-level maintenance plans based on hadamard product and non-dominated sorting genetic algorithm-III algorithm. EAAI, 157, 111292. (www)

  @article{ , abstract = {Road network-level maintenance decision-making is a resource allocation process that considers multiple factors, aiming to assign appropriate maintenance activities to various segments within the network. Current top-down, bottom-up, and synchronized decision-making methods have achieved multi-objective network-level maintenance decisions. However, these methods rarely quantify the engineering feasibility of maintenance activities, resulting in maintenance failures and resource waste. This paper employs the probability measures output by a neural network model to characterize engineering feasibility. By integrating these probability measures with pavement performance indices through the Hadamard product, segment-level maintenance decisions are generated. Based on segment-level maintenance decisions outcomes, a bottom-up network-level maintenance decision-making method is established, which comprehensively considers pavement performance, funding allocation, environmental impact, and engineering feasibility. A case study is conducted using data from a highway network in China. The results indicate that the segment-level maintenance outcomes of this method ensure that maintenance activities meet feasibility requirements, while the network-level maintenance plans can balance multiple objectives to meet actual engineering needs. In comparative experiments, the highest frequency of maintenance activity selection for this method is only 0.33, which is a reduction of 33 % and 59 % compared to the 0.49 and 0.80 of the two traditional methods, respectively, indicating better diversity in maintenance activity selection and effectively avoiding excessive use of a single maintenance resource, thus achieving sustainable maintenance practices.}, archive = {J_EAAI}, author = {Zhiwei Xie and Zheng Tong and Tao Ma and Tianqing Hei}, doi = {10.1016/j.engappai.2025.111292}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111292}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective decision-making method for road network-level maintenance plans based on hadamard product and non-dominated sorting genetic algorithm-III algorithm}, volume = {157}, year = {2025}, }
• (2025). Modeling and quantifying uncertainty in nanocomposite shell vibrations using heteroscedastic gaussian processes. EAAI, 157, 111290. (www)

  @article{ , abstract = {Accurate prediction of natural vibration frequencies is essential for the dynamic analysis and safe operation of advanced engineering structures. In functionally graded carbon nanotube-reinforced composite (FG-CNT) cylindrical shells, these frequencies are highly sensitive to geometric parameters such as thickness, radius, and length. Variability in these parameters, due to manufacturing and operational conditions, leads to input-dependent variance in the frequency response, which traditional Gaussian Process (GP) models fail to capture. To address this, a novel heteroscedastic Gaussian Process (HGP) model is proposed to simultaneously predict the mean response and input-dependent variance of natural frequencies in FG-CNT shells. Frequency datasets generated from Monte Carlo simulations (MCS) are used to train and validate the HGP across multiple vibrational modes. Results show that the HGP model significantly outperforms the standard GP, with predictions closely matching MCS results and achieving an R 2 value up to 0.99. Beyond accuracy, the HGP offers notable computational advantages, reducing the simulation time by approximately fivefold compared to MCS. Sobol sensitivity analysis with first-order and total indices further indicates that the radius is the most influential parameter on frequency variability, followed by the length, with thickness having a minor impact. The small gap between first-order and total-effect indices suggests limited interaction effects among parameters. Overall, this study establishes a computationally efficient and statistically robust framework for uncertainty quantification and sensitivity analysis in FG-CNT cylindrical shells, offering valuable insights for their reliable design and optimization in engineering applications.}, archive = {J_EAAI}, author = {Duc Tien Nguyen and Dinh Gia Ninh}, doi = {10.1016/j.engappai.2025.111290}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111290}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Modeling and quantifying uncertainty in nanocomposite shell vibrations using heteroscedastic gaussian processes}, volume = {157}, year = {2025}, }
• (2025). Dense wide forward harmonic network to enhance load balancing and fault tolerance in cloud computing through deep learning-driven clustering. EAAI, 157, 111267. (www)

  @article{ , abstract = {Load balancing is used to distribute workloads across multiple servers. Olden approaches were dependent on static rules made less adaptive to dynamic system changes and prone to downtime during failures. Thus, the Dense Wide Forward Harmonic Network (Den-WideFHNet) is designed to enhance fault tolerance and load balancing. Initially, the cloud model is simulated by distributing user demands across alternative tasks, which are processed by cloud management. Tasks are assigned to Virtual Machines (VMs) in a Round-Robin manner. VMs are categorized as overloaded or underloaded depending upon their parameters, like Central Processing Unit (CPU), memory, bandwidth, frequency scaling, and Million Instructions Per Second (MIPS) using Deep Fuzzy Clustering (DFC). Tasks are prioritized within overloaded VMs and transferred to underloaded VMs to balance system load, which considers factors, like load, capacity, resource utilization, supply, demand, and fault tolerance. Task parameters, like priority and execution time, are performed for load balancing. A novel Den-WideFHNet is a fusion of Dense Network (DenseNet), Wide Residual Network (WRN), and harmonic analysis exploited to enhance load balancing. Results acquired by Den-WideFHNet are 0.059 of load, 0.927 of resource utilization, 0.931 of capacity, 2.083 of energy consumption, 14.524 of migration overhead, and 0.529 of response time for task size 400 based on 15 p.m. and 20 VM.}, archive = {J_EAAI}, author = {Susila Nagarajan and Karthick Rathinasamy and Diwan Baskaran and Siamala Devi Sivarama Moorthy}, doi = {10.1016/j.engappai.2025.111267}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111267}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dense wide forward harmonic network to enhance load balancing and fault tolerance in cloud computing through deep learning-driven clustering}, volume = {157}, year = {2025}, }
• (2025). WFC-BSN: Wavelet fusion-based conditional blind-spot network for self-supervised forward sonar denoising. EAAI, 157, 111255. (www)

  @article{ , abstract = {Forward-looking sonar is a critical technology for underwater information acquisition, but complex environment and equipment limitation introduce significant noise, degrading image quality and hindering AI-based interpretation. Current self-supervised denoising methods, primarily designed for optical images, struggle to address inherent structural noise and restore local details of imaging targets. Additionally, these methods lack ground truth for evaluation, complicating the application of reference-based metrics in sonar denoising tasks. To address these issues, this paper proposes a wavelet fusion-based conditional blind-spot network (WFC-BSN) for sonar image denoising without needing ground truth data. The core of this network is the wavelet transform fusion block (WTFB), which decomposes feature maps into multiple frequency-domain components, merging summed and differential features. This enhances the receptive field and boosts the network’s ability to recover fine details. Moreover, WFC-BSN introduces Blind-Self Displacement Loss, comprising Blind-Spot Displacement Loss and Self-Similar Displacement Loss. The former enhances noise robustness through data perturbations, and the latter improves structural reconstruction by retaining original pixel information. We also present a novel method for generating sonar ground truth by extracting noise-free feature from continuous frames. This method utilizes forward-looking sonar data collected from real ocean environment to construct a clean-noise paired dataset, enabling reference-based denoising performance evaluation. Experimental results demonstrate that whether for high- or low-frequency sonar images, WFC-BSN surpasses both traditional non-learning and existing self-supervised denoising techniques in effectiveness and detail preservation. Additionally, denoised images show improved accuracy in detection tasks, underscoring the method’s efficacy and practical value in underwater applications. Code and Dataset can be found at https://github.com/xzq-2000/WFC-BSN .}, archive = {J_EAAI}, author = {Ziqi Xia and Jie Li and Wenpei Jiao and Chunyan Zhang and Jianlei Zhang and Guangming Xie}, doi = {10.1016/j.engappai.2025.111255}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111255}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {WFC-BSN: Wavelet fusion-based conditional blind-spot network for self-supervised forward sonar denoising}, volume = {157}, year = {2025}, }
• (2025). Distributed long-time series prediction based on multiple feature decomposition and federated meta-learning under data heterogeneity. EAAI, 157, 111252. (www)

  @article{ , abstract = {Accurate and efficient traffic flow prediction is essential for intelligent transportation systems management. Traditional prediction methods have limited performance in capturing time dependent feature, especially in long time series prediction. Deep learning has limited improvement on this problem and also leads to poor model generalization due to data heterogeneity. Therefore, we propose a distributed long time series prediction framework. It primarily focuses on the enhancement of client-side local models in distributed long time series forecasting. On the one hand, in order to capture the characteristics of the client data, we design a novel multidimensional decomposition module to process the data as input to the subsequent model. To effectively address seasonality and periodicity of long time series forecasting, we propose a backbone forecasting model based on the Patch-Mixer architecture. On the other hand, to enhance the model's generalization capability across diverse tasks, we introduce an enhanced Model-Agnostic Meta-Learning (MAML) framework with improved adaptation mechanisms. This mechanism enables each client model to rapidly adapt to its local tasks through parameter distribution from the server. Furthermore, we propose an acceleration strategy based on Fast Local Initialization (FLI) to optimize the training efficiency of local models. Finally, a real heterogeneous inter-city traffic flow dataset is taken to verify the effectiveness of the proposed method. Experimental results demonstrate that our proposed method outperforms the baseline models across multiple evaluation metrics.}, archive = {J_EAAI}, author = {Meng Yang and Shuyuan Zhang and Zhanzhong Wang and Shuoqi Wang}, doi = {10.1016/j.engappai.2025.111252}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111252}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Distributed long-time series prediction based on multiple feature decomposition and federated meta-learning under data heterogeneity}, volume = {157}, year = {2025}, }
• (2025). Block information strategy for multi-modal remote sensing image registration. EAAI, 157, 111236. (www)

  @article{ , abstract = {Registration of multi-modal remote sensing image pairs (MRSI) is challenging given the distinct imaging mechanisms of multi-modal data sources, which lead to substantial geometric and radiometric distortions and inaccuracies in correspondences. To tackle this issue, we propose a novel approach that integrates local image information into feature representations through the design of local regions and the extraction of local information. The latter comprises of two key components: rank-based feature redistribution and residual information extraction utilizing a pyramid-like structure of local patches. This enhanced feature representation technique, termed Reinforced Local Information of LSS (RLILSS), embeds local information to improve the performance of the Local Self-Similarity (LSS)-based framework for MRSI registration. RLILSS strengthens feature characterization across various regions and addresses the limitations of supplementary information. This enables more reliable correspondences between images. Experimental results show that the proposed method achieves higher accuracy and better registration across diverse multi-modal datasets. Detailed analyses confirm its superiority over state-of-the-art methods in both accuracy and robustness. This approach holds significant potential for applications in automatic geographic registration and disaster area reconstruction.}, archive = {J_EAAI}, author = {Yameng Hong and Chengcai Leng and Beihua Liu and Jinye Peng and Irene Cheng and Anup Basu}, doi = {10.1016/j.engappai.2025.111236}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111236}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Block information strategy for multi-modal remote sensing image registration}, volume = {157}, year = {2025}, }
• (2025). Fast inversion of parameters on Jiles–Atherton hysteresis model based on physics-guided deep learning network. EAAI, 157, 111233. (www)

  @article{ , abstract = {The Jiles–Atherton (J–A) hysteresis model is widely used to illustrate the properties of ferromagnetic materials. However, in the inverse J–A hysteresis model, careful selection of the initial seed value is crucial for accurately obtaining the model parameters. This selection impacts both the calculation time and the model’s adaptability. To address this problem, the physics-guided Kolmogorov–Arnold Networks (KAN)-EfficientNet deep learning (DL) model is proposed to achieve fast and accurate J–A model parameters estimation. In the innovative approach, the physics constraints from the inversed J–A hysteresis model and the KAN-Linear layer are utilized to accelerate the convergence of the deep learning model and enhance its accuracy. The theoretical framework of the J–A hysteresis model and its inversed model is illustrated, alongside the development of an experimental platform for data collection to verify the proposed method’s feasibility. Data from classical J–A hysteresis numerical model and simulations are utilized to train the weights of proposed DL model. The compared algorithms are also introduced to illustrate the performance of the proposed model. The experimental data are fed into the analytical J–A hysteresis model and the pre-trained physics-guided KAN-EfficientNet model, respectively. The results show that the proposed method provides high accuracy (error less than 2.41%) in less calculation time (0.3 s) to determine the J–A hysteresis model parameters. The work introduces an innovative physics-guided DL model that significantly enhances the efficiency of J–A hysteresis model parameters estimation, thereby promoting the broader application of J–A hysteresis model in study of ferromagnetic materials.}, archive = {J_EAAI}, author = {Saibo She and Xiaochu Pang and Jun Liu and Xinnan Zheng and Kuohai Yu and Xun Zou and Ruoxuan Zhu and Rui Guo and Wuliang Yin}, doi = {10.1016/j.engappai.2025.111233}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111233}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast inversion of parameters on Jiles–Atherton hysteresis model based on physics-guided deep learning network}, volume = {157}, year = {2025}, }
• (2025). WB-YOLO: An efficient wild bat detection method for ecological monitoring in complex environments. EAAI, 157, 111232. (www)

  @article{ , abstract = {The study of bat species and their distribution is vital for understanding the origins and transmission pathways of epidemic diseases. However, the detection of wild bats faces significant challenges due to their complex natural habitats and frequent occlusions caused by social behavior. To address these issues, we propose an object detection method based on the improved You Only Look Once version 7 (YOLOv7) to achieve efficient wild bat detection (WB-YOLO). This method integrates a Vision Transformer encoder module to improve global context integration, adopts deformable convolution, and optimizes the spatial pyramid pooling structure for effective multi-scale feature fusion while reducing computational complexity. Furthermore, a hybrid attention mechanism is introduced to capture both spatial and channel information, enhancing robustness in complex environments. Experimental results in a data set of wild bat images collected in Anhui Province demonstrate that WB-YOLO achieves a precision of 90. 7%, a recall of 89. 0%, and a mean average precision (mAP) of 94. 7%, significantly outperforming other deep learning models in detecting bats in complex scenes and under occlusion. Our approach offers an efficient and accurate solution for the detection of wild bats in real time, with potential applications in ecological research and disease prevention. Code and data related to this work are publicly available at https://github.com/macandzzz/WB-YOLO .}, archive = {J_EAAI}, author = {Yang Wang and Chang Ma and Chuanxin Zhao and Huijuan Xia and Congxi Chen and Ying Zhang}, doi = {10.1016/j.engappai.2025.111232}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111232}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {WB-YOLO: An efficient wild bat detection method for ecological monitoring in complex environments}, volume = {157}, year = {2025}, }
• (2025). High-resolution multi-view stereo with multi-scale feature fusion. EAAI, 157, 111231. (www)

  @article{ , abstract = {To enhance the handling of three-dimensional reconstruction for large-scale scenes and high-resolution images, we introduce a novel multi-view high-resolution three-dimensional reconstruction approach. Our proposed method integrates a Feature Pyramid Network with the Swin Transformer for improved performance. We integrate the Swin Transformer into the feature pyramid. This integration aims to establish long-range feature dependencies, facilitate information exchange between different input positions, and enhance the global consistency of feature representation. This improves the efficiency of the feature extraction stage. Following this, we apply cost volume regularization to mitigate noise and compute depth maps. A Depth Optimization Module is employed to refine the predicted depth maps, thereby enhancing their precision. Experimental results demonstrate the efficacy of our method in generating more accurate depth information, particularly in predicting high-resolution depth maps. Our approach utilizes these depth predictions to generate point clouds, enabling precise matching and reconstruction of multi-view images. Experiments conducted on public datasets validate the effectiveness and superiority of our proposed method.}, archive = {J_EAAI}, author = {Dapeng Chen and Qi Jia and Hao Wu and Da Yu and Nanxuan Huang and Jia Liu}, doi = {10.1016/j.engappai.2025.111231}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111231}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {High-resolution multi-view stereo with multi-scale feature fusion}, volume = {157}, year = {2025}, }
• (2025). Memory guided representation learning for cross-domain face anti-spoofing. EAAI, 157, 111230. (www)

  @article{ , abstract = {Addressing generalized Face Anti-Spoofing is challenging due to the wide variety of spoofing techniques, variations in environmental conditions, and the diversity of devices used to capture images. Most approaches to enhancing the generalization ability of systems often manipulate image statistics to normalize images into a uniform representation space. However, this approach can restrict the capacity of the system to represent images accurately because image statistics vary significantly across different domains, each with unique characteristics. Recognizing that each source domain possesses distinct characteristics, we introduce an innovative approach based on memory guided representation learningto represent these characteristics in separate latent spaces. Specifically, we present a dual component framework comprising a Memory Guided Clustering Representation (MGCR)generator and a Memory Guided Mapping Representation (MGMR)classifier. Additionally, we create a memory bank filled with template and meta features, which are refined over time using a momentum update mechanism. The MGCR component employs clustering to allow new, unseen deep features close to the most similar template feature, thereby creating generalized clues for the generator. Meanwhile, the MGMR process leverages meta features to dynamically represent new domain spaces through linear combinations, bridging the divide between known and unknown domains. Our experimental results show that the proposed method significantly enhances the effectiveness under cross-domain scenarios, outperforming existing techniques.Especially in the “Leave One Out”setting, its average Half Total Error Rate exceeds that of other methods, reaching less than 10%.}, archive = {J_EAAI}, author = {Pengchao Deng and Yanhui Zhou and Zhiheng Fu and Jian Liu and Shengjun Xu and Chenyang Ge and Farid Boussaid and Mohammed Bennamoun}, doi = {10.1016/j.engappai.2025.111230}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111230}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Memory guided representation learning for cross-domain face anti-spoofing}, volume = {157}, year = {2025}, }
• (2025). A novel facial expression recognition method based on cross direction attention network. EAAI, 157, 111229. (www)

  @article{ , abstract = {Facial expression recognition (FER) is an area of growing interest in computer vision research. This paper extends the framework provided by the ‘Distract your Attention Network’ (DAN) which consists of multiple parallel branches, each branch composes of a spatial attention (SA) module followed by a channel attention (CA) module, and then these multiple branches are fused together before being passed into a classifier module. The spatial attention module of DAN has an internal channel dimension of 1, while our proposed Cross Directional Attention Network (CDAN)-I and CDAN-II contain respectively an internal channel dimension of 512 (same as the channel dimension of the input), and internal channel dimension of 1024 (double that of the channel dimension of the input). These increases in internal channel dimension allow extraction of more features, before they are being made to conform with the input channel dimension. Despite these seemingly simple modifications from that of DAN, both CDAN-I and CDAN-II are found to outperform those of DAN, a state-of-the-art FER method, on four popular FER benchmark datasets: RAF-DB (Real world Affective Face-database), AffectNet-7 (AffectNet with Seven Categories) AffectNet-8 ( AffectNet with Eight Categories), and CK+ (Cohn–Kanada Extended). Moreover, we make use of three statistical indexes for clustering analysis, and verified that the CDAN-I and CDAN-II modules have been able to increase the inter-cluster distances, and decrease the intra-cluster distances, when compared with those obtained by the backbone ResNet-18 network (Residual Network with 18 Layers) , thus providing a quantitative analysis technique in this area.}, archive = {J_EAAI}, author = {Cheng Peng and Guodong Li and Likang Lin and Bowen Zhang and Kun Zou and Sio Long Lo and Ah Chung Tsoi}, doi = {10.1016/j.engappai.2025.111229}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111229}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel facial expression recognition method based on cross direction attention network}, volume = {157}, year = {2025}, }
• (2025). Enhanced process monitoring using machine learning-based control charts for poisson-distributed data. EAAI, 157, 111227. (www)

  @article{ , abstract = {The ability to detect shifts (e.g., outliers) in process monitoring is crucial for maintaining high-quality standards and operational efficiency in industrial environments. Control Charts (CCs) provide an organized framework for recognizing and managing anomalies, generally caused by assignable factors (e.g., identifiable issues) rather than inherent process variation. Traditional CCs, such as classical Shewhart, CUSUM, and EWMA, are commonly used to monitor Poisson observations in modern industries. The classical exponentially weighted moving average (EWMA) and cumulative sum (CUSUM) CCs are individually effective at detecting small-to-moderate shifts while Shewhart CCs identify moderate-to-large shifts in the process location and/or dispersion parameters. However, the classical CCs face limitations due to their sensitivity being constrained to specific ranges of shifts. To enhance the detection abilities of classical CCs in detecting all kinds of shifts in the process location parameter, this study proposes the integration of Machine Learning (ML) techniques into CCs to optimize the shift’s detection in process location parameter across a wider range. This study generates a dataset using the statistics of classical CCs based on Poisson-distributed data, which includes both in-control (stable process) and out-of-control (unstable process) processes. This dataset is used to train ML models, which are pre-processed through normalization and feature engineering through a heuristic approach before training. The performance of ML models is evaluated using standard regression metrics, specifically mean squared error and the coefficient of determination ( R 2 -score), to ensure effective generalization across varying process conditions. After training, these models are implemented within the proposed ML-based CC ( M L C C ) schemes. Their process monitoring capabilities are then rigorously compared with traditional and existing CCs, utilizing key performance indicators such as average run length and standard deviation of run length. These metrics are computed through a Python-based algorithm developed using the Monte Carlo simulation method. For practical purposes, implementing the proposed M L C C schemes with real-life data in the food processing industry, specifically the packaging of frozen orange juice concentrate. This practical example highlights the superiority of proposed M L C C schemes in the early detection of shift(s) in process location parameter(s) against classical CCs in real-life scenarios.}, archive = {J_EAAI}, author = {Faraz Mukhtiar and Babar Zaman and Naveed Razzaq Butt}, doi = {10.1016/j.engappai.2025.111227}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111227}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced process monitoring using machine learning-based control charts for poisson-distributed data}, volume = {157}, year = {2025}, }
• (2025). SmartMatch: A semi-supervised framework for obstructive sleep apnea classification using single-lead electrocardiogram signals with limited annotations. EAAI, 157, 111226. (www)

  @article{ , abstract = {Obstructive Sleep Apnea (OSA) is a prevalent respiratory disorder with significant global health implications, affecting individuals of all ages and demographics. This underlines the demand for reliable diagnostic methods and computational models capable of accurately classifying OSA efficiently using single-lead electrocardiogram (ECG) with Limited Annotations. Recent progress in deep learning (DL) techniques, coupled with high-performance computing, has facilitated automatic classification in medical imaging, offering a viable solution for timely diagnosis. However, Traditional supervised learning (SL) requires large, expert-annotated datasets, which are costly and time-intensive to curate, which is a critical bottleneck in resource-constrained settings. To overcome these challenges, we proposed a semi-supervised learning (SSL) framework, SmartMatch. Our SSL framework minimizes reliance on annotated data by effectively leveraging unlabeled ECG signals, reducing annotation burdens while maintaining diagnostic accuracy. The proposed framework is inspired by hierarchical structures observed in real-world scenarios, where leader-follower dynamics play a crucial role in decision-making. Our approach integrates deep metric learning, employing Adaptive batch hard mining to enhance feature representation, alongside an Adaptive pseudo-labeling strategy to refine label quality, and an Adaptive temporal ensembling to stabilize learning while preserving consistency loss constraints. We conducted experiments on the PhysioNet Apnea-ECG repository (PA-ECG) dataset, which comprises 70 overnight recordings. The proposed SmartMatch achieved remarkable performance, yielding high accuracy, precision, recall, and F1-scores of 91.99% ( ± 0 . 08 ) , 91.98% ( ± 0 . 10 ) , 91.99% ( ± 0 . 11 ) , and 91.97% ( ± 0 . 10 ) per-segment, respectively. The results highlight SSL’s capability to strengthen DL models for OSA detection, particularly in applications involving home sleep apnea testing and wearable IoT devices.}, archive = {J_EAAI}, author = {Soumyajit Gayen and Nalla Maheswara rao and Deepak Kumar Sahu and J. Sivaraman and Kunal Pal and Srikanth Vasamsetti and Bala Chakravarthy Neelapu}, doi = {10.1016/j.engappai.2025.111226}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111226}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SmartMatch: A semi-supervised framework for obstructive sleep apnea classification using single-lead electrocardiogram signals with limited annotations}, volume = {157}, year = {2025}, }
• (2025). Fuzzy-based text augmentation to boost the statistical machine translation for indic languages. EAAI, 157, 111221. (www)

  @article{ , abstract = {Artificial intelligence technology has been used in various industries to give convenience to people's lives. However, real-world applications often face a critical challenge of data scarcity. Text Augmentation (TA) techniques are being investigated extensively in the field of natural language processing to solve this data scarcity and enhance model performance. For Indian Languages, data collection is challenging as they exhibit rich syntactic and morphological diversity compared to resource-rich languages like English. This diversity further compounds the problem of data scarcity, leading to poor translation quality, especially when translating from low-resource languages to resource-rich ones. This study addresses the challenge by proposing a fuzzy-based TA technique to enhance machine translation quality. The proposed approach leverages fuzzy matching to identify and utilize potential near-matches in translated sentences, thereby augmenting the available training data. Fuzzy is a lexicalized matching strategy that seeks out non-exact matches in a sentence. To evaluate the effectiveness of this method, three resource-rich Indic languages were considered, including a low-resource endangered language. Experimental results on the test set demonstrate significant and consistent improvements in the augmented dataset, achieving a +3.53 of BiLingual Evaluation Understudy (BLEU) and +6.247 of Metric for Evaluation of Translation with Explicit ORdering (METEOR) point increase over the baseline system. Furthermore, we conducted statistical analysis to confirm the significance of these results, validating the enhanced quality of the translation tasks.}, archive = {J_EAAI}, author = {Shefali Saxena and Ayush Gupta and Shweta Chauhan and Philemon Daniel}, doi = {10.1016/j.engappai.2025.111221}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111221}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzy-based text augmentation to boost the statistical machine translation for indic languages}, volume = {157}, year = {2025}, }
• (2025). Human-in-the-loop reinforcement learning for dynamic soaring: A trajectory planning and control integrated system. EAAI, 157, 111219. (www)

  @article{ , abstract = {Increasing the endurance of fixed-wing unmanned aerial vehicles has long been a designer’s dream. Inspired by the behavior of albatrosses, dynamic soaring holds significant potential for achieving perpetual unpowered flight. However, traditional approaches face challenges in implementing dynamic soaring in real-world scenarios. Prior studies simplified kinematics and control inputs, and separated design of the optimizer and controller. To this end, we propose a human-in-the-loop framework based on deep reinforcement learning to enable automated execution of dynamic soaring. First, pseudospectral method is utilized to generate the optimal trajectory for learning. Then, deep the reinforcement learning algorithm interacts with a high-fidelity environment, sending commands to the lower-level controller. Humans contribute domain expertise and design reward functions and evaluation criteria to assist the agent in trajectory search. Our results demonstrate the autonomous execution of dynamic soaring without human intervention, with enhanced energy collection in thin-shear wind fields. Importantly, our findings are more congruent with practical applications of real-world gliders compared to optimal control theory. The scope of this study also contains simple robustness tests which indicate the agent’s resistance to wind disturbances and deviations in the initial state. This research demonstrates the feasibility of integrating automatic planning and control, accelerating the transition of dynamic soaring from simulation to real-world applications.}, archive = {J_EAAI}, author = {Yufeng Lu and Fangfang Xie and Tingwei Ji}, doi = {10.1016/j.engappai.2025.111219}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111219}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Human-in-the-loop reinforcement learning for dynamic soaring: A trajectory planning and control integrated system}, volume = {157}, year = {2025}, }
• (2025). Automatic text summarization techniques: A categorization, evolution and future scope. EAAI, 157, 111216. (www)

  @article{ , abstract = {Web resources on the Internet are large textual data sources that expand exponentially on a regular basis. Users find it challenging to discover relevant information in the time frame they seek due to the overabundance of material on the Internet. As a result, there was a strong demand in the research community for developing automatic text summarization systems. Text summarization is the method of reducing a large text document into a concise version that allows users to save effort and time while still understanding the essence of the original document. This article provides scholars with a complete analysis of text summarization by covering the various features of text summarization, the techniques, their categorization, distinct components, and associated limitations. In addition, we have provided a discussion of the available techniques in chronological order from 2000 to 2025 for each category, with both extractive and abstractive summarization. Furthermore, the article also includes a detailed discussion on the datasets used and the methods of summary evaluation.}, archive = {J_EAAI}, author = {Ramesh Chandra Belwal and Atul Gupta}, doi = {10.1016/j.engappai.2025.111216}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111216}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic text summarization techniques: A categorization, evolution and future scope}, volume = {157}, year = {2025}, }
• (2025). Enhancing hydroponic farming with machine learning: Growth prediction and anomaly detection. EAAI, 157, 111214. (www)

  @article{ , abstract = {Artificial Intelligence has revolutionized precision agriculture, particularly in optimizing hydroponic farming. Our study targets Batavia lettuce cultivation, utilizing environmental and water sensors, alongside a custom camera system to capture plant growth images. The data collected undergoes extensive pre-processing, including standardization, clustering, and feature extraction via image processing techniques. Machine Learning (ML) models are employed such as linear regression, neural networks, and Random Forest (RF) classifiers, to predict growth rates and detect anomalies. Our findings demonstrate that neural networks significantly outperform linear regression in predicting growth rates, with a 73.88% reduction in mean squared error. Additionally, the RF classifier achieves high accuracy in classifying growth conditions, with an accuracy rate of 94.45%. These results suggest that integrating ML with hydroponics enhances the efficiency and reliability of crop cultivation, improving resource utilization and minimizing ecological footprints in agriculture. The implications extend to real-time monitoring and decision-making in controlled environment agriculture, potentially transforming traditional farming practices.}, archive = {J_EAAI}, author = {Ahmed Métwalli and Omar Shalash and Alia Elhefny and Nour Rezk and Fares El Gohary and Omar El Hennawy and Fouad Akrab and Adam Shawky and Ziad Mohamed and Nayira Hassan and Mostaafa Hassanen}, doi = {10.1016/j.engappai.2025.111214}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111214}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing hydroponic farming with machine learning: Growth prediction and anomaly detection}, volume = {157}, year = {2025}, }
• (2025). An adaptive output feedback formation control algorithm for multiple unmanned surface vehicles based on virtual structure and line-of-sight guidance. EAAI, 157, 111213. (www)

  @article{ , abstract = {The formation, and switching control problems for multiple unmanned surface vehicles (USVs) subject to external disturbances and model uncertainties are studied. A novel formation generation algorithm is proposed to address situations where the USV formation needs to rapidly switch its configuration to execute tasks during cooperative navigation. Specifically, an extended virtual structure method is used to generate virtual structure points that enable formation switching. Virtual leaders, based on the artificial potential field method, generate collision-free path-tracking virtual structure points to achieve formation. These virtual leaders provide reference trajectories for the USV’s trajectory guidance. Finally, an adaptive output feedback tracking control method based on line-of-sight (LOS) guidance and neural networks is developed to achieve precise and rapid one-to-one tracking of the virtual leaders by the USVs. The adaptive method compensates for the upper bounds of uncertainties in relative kinematics, and dynamic surface control is used to stabilize the yaw angle tracking error of the following USVs. A high-gain observer is constructed to estimate the speed of the USV, and combining the virtual control law, neural network approximation techniques, and a minimal parameter learning algorithm, an adaptive output feedback tracking controller is designed for the following USVs. The stability of the tracking system is confirmed through Lyapunov analysis. Simulation experiments in MATLAB with varying numbers of USVs and tasks demonstrate the effectiveness of the proposed formation control algorithm.}, archive = {J_EAAI}, author = {Luyan Sheng and Hui Ye and Yizhen Meng and Chengze Song}, doi = {10.1016/j.engappai.2025.111213}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111213}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive output feedback formation control algorithm for multiple unmanned surface vehicles based on virtual structure and line-of-sight guidance}, volume = {157}, year = {2025}, }
• (2025). SAMGCN: A spatially-augmented multi-view graph convolutional network for identifying spatial domains. EAAI, 157, 111203. (www)

  @article{ , abstract = {Recent innovations in spatial transcriptomics have enabled the measurement of gene expression profiles while preserving the spatial organization of cells. This provides extensive opportunities to explore gene expression patterns in the tissue microenvironment. However, it remains a challenge to combine spatial information with gene expression to accurately identify spatial domains. In this study, a spatially-augmented multi-view graph convolutional network for identifying spatial domains (SAMGCN) is proposed. First, SAMGCN reconstructs gene expression data by incorporating spatial neighborhood information, which enhances gene expression features. It improves the quality of gene expression data and augments the characterization of spatial domains through the construction of spatial graphs, feature graphs, and spatial expression-weighted graphs. By extracting spatial information and gene expression data via convolutional operations, SAMGCN learns multi-view-specific embeddings and employs a contrastive strategy to refine and augment spatial neighborhood relationships, addressing limitations in spatial gene expression data. An attention mechanism is then employed to flexibly merge these embeddings, generating the final spot embedding. Additionally, a zero-inflated negative binomial decoder is used to capture the global probability distribution of gene expression profiles. Finally, the performance of SAMGCN has been validated across various platforms and spatial transcriptomics datasets of different scales, demonstrating its exceptional capability to process spatial transcriptomics data.}, archive = {J_EAAI}, author = {Hao Liu and Yue Gao and Ying-Lian Gao and Cui-Na Jiao and Junliang Shang and Jin-Xing Liu}, doi = {10.1016/j.engappai.2025.111203}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111203}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SAMGCN: A spatially-augmented multi-view graph convolutional network for identifying spatial domains}, volume = {157}, year = {2025}, }
• (2025). Correlation based deep neuro-fuzzy hammerstein type wind power forecasting model considering asymmetric error characteristics. EAAI, 157, 111200. (www)

  @article{ , abstract = {Accurate and reliable wind power prediction is critical for the power system. However, most existing methods assume symmetric modeling errors, which contradict the observed asymmetric error distributions in practice. In addition, these methods often struggle with wind power series containing both linear and nonlinear elements, complicating controller design. To address these issues, we propose a novel artificial intelligence (AI) solution for wind power forecasting. In data preprocessing, a feature generation mechanism extracts approximation and detail components from wind data. In prediction stage, the deep neuro-fuzzy Hammerstein model, a cascade structure comprising a static nonlinear block followed by a dynamic linear block, is designed to capture both linear and nonlinear features from these components. Specifically, the composite signals are designed to decouple the identification of the dynamic linear block from that of the static nonlinear block. Then, the correlation analysis identifies the parameters of the linear component, while a hybrid parameter updating approach guaranteeing the global convergence of the model weights is derived based on the Lyapunov to adjust the weights of the nonlinear block. Finally, a probability density function (PDF) control based ensemble strategy is proposed to integrate the forecast results of two components for the final outputs. Additionally, the model’s special structure allows inversion of the nonlinear block, transforming the wind turbine system into a generalized linear structure to simplify controller design. Experimental evaluations on real-world wind farm datasets show our deep Hammerstein model achieves better forecasting accuracy than conventional AI-based approaches when applied to wind power prediction tasks.}, archive = {J_EAAI}, author = {Jianfang Li and Li Jia and Wei Yang and Chengyu Zhou}, doi = {10.1016/j.engappai.2025.111200}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111200}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Correlation based deep neuro-fuzzy hammerstein type wind power forecasting model considering asymmetric error characteristics}, volume = {157}, year = {2025}, }
• (2025). Varied granularity encoding based evolutionary algorithm for multi-objective intensity-modulated radiation therapy optimization. EAAI, 157, 111193. (www)

  @article{ , abstract = {Intensity-modulated radiation therapy is an interesting multi-objective optimization problem, which holds a large number of aperture shape-related variables, posing a stiff challenge to existing algorithms. To efficiently solve this problem, we propose a varied granularity encoding method in this paper, where the granularity of encoding of the shape in the multi-leaf collimator is progressively refined during the optimization. Specifically, at the beginning of the search, a coarse encoding is adopted by dividing the aperture shape-related variables into several groups and representing each group by one bit, which achieves effective search space reduction for the aperture shape. During the evolution, the granularity of encoding aperture shape-related variables is gradually varied from coarse to fine by reducing the size of each group until only one variable is contained in the group. With the proposed varied granularity encoding method, an evolutionary algorithm is suggested based on a popular evolutionary multi-objective framework (NSGA-II), where an adaptive switching method is developed to determine whether the granularity level needs to be reduced according to the convergence status of the population. The experiment empirically investigates the performance of the proposed varied granularity encoding method based evolutionary algorithm on eight clinical instances with the number of aperture shape-related variables ranging from 1932 to 3180. Compared with seven representative evolutionary algorithms and one traditional direct aperture optimization algorithm, the proposed algorithm demonstrates statistically significant improvements in hypervolume, inverted generational distance, and dose-volume histogram. The experimental results reveal that the proposed algorithm not only exhibits competitiveness but reduces computational time in radiotherapy optimization.}, archive = {J_EAAI}, author = {Langchun Si and Xingyi Zhang and Ye Tian and Ruifen Cao and Shangshang Yang and Limiao Zhang}, doi = {10.1016/j.engappai.2025.111193}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111193}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Varied granularity encoding based evolutionary algorithm for multi-objective intensity-modulated radiation therapy optimization}, volume = {157}, year = {2025}, }
• (2025). Exploring multiobjective evolutionary algorithms for designing ribonucleic acid sequences: An experimental analysis. EAAI, 157, 111189. (www)

  @article{ , abstract = {Evolutionary algorithms have proven effective in addressing the Ribonucleic Acid (RNA) inverse folding problem, a critical challenge in Biomedical Engineering. This problem, involving the discovery of a nucleotide RNA sequence that folds into a desired secondary structure, is formulated as a Multiobjective Optimization Problem. In this study, we introduce an approach incorporating three objective functions (Partition Function, Ensemble Diversity, and Nucleotides Composition) and a constraint (Similarity), utilizing a real-valued chromosome encoding. The primary focus is on analyzing and comparing the performance of four multiobjective evolutionary algorithms. We explore various crossover (Simulated Binary, Differential Evolution, One-Point, Two-Point, K-Point, and Exponential) and selection (Random and Tournament) operators, coupled with a fixed mutation operator (Polynomial). Our investigation involves 48 distinct algorithm-operator combinations, with the aim of solving a well-known benchmark set. This research makes a significant contribution to the field of Artificial Intelligence by addressing a complex problem through the lens of Multiobjective Optimization. The proposed framework not only advances our understanding of RNA inverse folding but also demonstrates the versatility of evolutionary algorithms in tackling real-world challenges in Biomedical Engineering. Our findings provide valuable insights into the behavior of different algorithmic elements and combinations, identifying optimal and suboptimal performers for future research and practical applications.}, archive = {J_EAAI}, author = {Álvaro Rubio-Largo and Nuria Lozano-García and José M. Granado-Criado}, doi = {10.1016/j.engappai.2025.111189}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111189}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploring multiobjective evolutionary algorithms for designing ribonucleic acid sequences: An experimental analysis}, volume = {157}, year = {2025}, }
• (2025). Multi-action deep reinforcement learning-based collaborative production dynamic scheduling approach for networked multi-factory under mass customization. EAAI, 157, 111187. (www)

  @article{ , abstract = {Increasing customization demand is accelerating the decentralization of manufacturing systems, elevating the distributed scheduling problem (DSP) to a prominent research focus. However, the dynamic characteristics of mass customization (MC) and the practically adaptable networked multi-factory systems remain underexplored. To address this gap, this study formulates a networked multi-factory collaborative production scheduling problem in the context of MC (MC-NMF-CPSP), and proposes a solution framework based on multi-action deep reinforcement learning (MA-DRL). The framework formulates a multiple Markov Decision Process (MMDP) incorporating a multi-dimensional state space, a two-stage action space, and a composite reward, providing a unified modeling paradigm for high-dimensional complex scheduling problems. To accurately characterize the dynamic demand, a consistent measurement model is proposed and embedded within the state space for efficient state representation. To enhance learning performance, an improved Proximal Policy Optimization (PPO) is developed, and a Multi-PPO architecture is integrated to construct a networked multi-factory collaborative production scheduling model (MaDRL-NMF-CPSM), which demonstrates exceptional coordination and adaptability under dynamic demand environments. Furthermore, a retraining mechanism is introduced to enable continual model evolution, thereby improving robustness and generalization. Extensive experiments show that the proposed method significantly outperforms several benchmark algorithms and heuristic rules in terms of convergence efficiency, scheduling performance, and dynamic adaptability, offering a novel theoretical foundation and practical solution for collaborative scheduling in networked manufacturing systems under dynamic demand.}, archive = {J_EAAI}, author = {Mingzhou Liu and Zhengfu Ding and Xi Zhang and Lin Ling and Maogen Ge}, doi = {10.1016/j.engappai.2025.111187}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111187}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-action deep reinforcement learning-based collaborative production dynamic scheduling approach for networked multi-factory under mass customization}, volume = {157}, year = {2025}, }
• (2025). Anomaly detection in ship trajectories using machine learning and dynamic time warping. EAAI, 157, 111185. (www)

  @article{ , abstract = {This research paper proposes adaptations of three state-of-the-art anomaly detection algorithms, (One-Class Support Vector Machine, Isolation Forest and Local Outlier Factor), for detecting abnormal behavior in ship trajectories in an unsupervised way. These algorithms are adapted and tested using a wide range of similarity measures built specifically for time series, such as Dynamic Time Warping. The proposed methods are first applied on synthetic Automatic Identification System datasets with available ground truth. Then, they are generalized to handle pairs of Automatic Identification System and radar trajectories to detect unexpected activities, such as route deviations, delays and entering prohibited zones. The performances of the proposed methods are shown to be competitive when compared to the state-of-the-art for abnormal ship behavior detection.}, archive = {J_EAAI}, author = {Valérian Mangé and Jean-Yves Tourneret and François Vincent and Laurent Mirambell and Fábio Manzoni Vieira}, doi = {10.1016/j.engappai.2025.111185}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111185}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Anomaly detection in ship trajectories using machine learning and dynamic time warping}, volume = {157}, year = {2025}, }
• (2025). Dynamic masking-based feature interaction modeling for e-commerce click-through rate prediction. EAAI, 157, 111184. (www)

  @article{ , abstract = {Learning feature interactions is crucial to click-through rate (CTR) prediction in e-commerce applications. The interaction between target items and user behavior history is one of the most important factors. In recent years, Deep Neural Networks (DNNs) designed to automatically learn implicit nonlinear interactions from the original sparse features have been widely adopted for industrial CTR prediction tasks. Implicit feature interactions often cannot retain the full representation capacity of the original feature interactions. Explicit modeling approaches, such as Cartesian product methods and matrix factorization-based models, either suffer from high computational costs or have limited expressive power. Recent deep learning-based methods improve feature interaction modeling but often overlook the enhancement of feature representations and fail to consider the varying importance of different interactions, limiting model performance. To address these challenges, in this paper, we report the design, implementation and deployment of F eature D ynamic M asking ( FDM ), a novel framework for modeling feature interactions in deep CTR prediction models. FDM dynamically generates unique feature representations for different interactions using learnable masks, enabling information sharing while maintaining efficiency. Additionally, it introduces a weighting module to capture the importance of different interactions. FDM framework is simple, effective and easy to deploy in real-world search and recommender systems. Since its deployment in the search system of Shopee Pte Ltd., Singapore, it has achieved a commercially significant 2.76% improvement in Order Per User (OPU) and 2.45% improvement in Gross Merchandise Volume (GMV).}, archive = {J_EAAI}, author = {Yabo Ni and Yueqiu Wu and Jingyi Li and Anxiang Zeng and Han Yu and Xiaoxiao Li}, doi = {10.1016/j.engappai.2025.111184}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111184}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic masking-based feature interaction modeling for e-commerce click-through rate prediction}, volume = {157}, year = {2025}, }
• (2025). Rotor imbalance fault classification through unified feature fusion: Combining statistical signal processing with adaptive deep learning features. EAAI, 157, 111182. (www)

  @article{ , abstract = {The most important element of efficient fault classification lies in the identification of relevant features that can serve as representations for various fault categories. For time series classification, the utilization of solely basic Statistical Features (SFs) does not yield high accuracy in classification. On the other hand, the exclusive dependence on machine learning-based feature extraction, without the incorporation of prior knowledge, compromise the model’s ability to generalize. To tackle these concerns, in this work, a hybrid methodology has been proposed, wherein the utilization of basic SFs is combined with Deep Learning (DL) model’s capability to facilitate comprehensive feature extraction, while simultaneously upholding the robustness of the model. A modified DL architecture based on the second generation wavelet decomposition theory is designed to extract Adaptive Latent Features (ALFs) from rotational time series for the purpose of classifying rotor imbalance faults. The validity of this approach is confirmed by assessing its practicality and effectiveness over a benchmark as well as on a custom dataset obtained from an experimental test-bed for rotor structural faults classification. The experiment demonstrates that the proposed method achieves 99.63% accuracy on the custom dataset, outperforming Support Vector Machine (56.89%), Extreme Gradient Boosting (72.38%), and Lifting Net (94.93%). The validation on the benchmark datasets highlights the effectiveness of the proposed method in distinguishing between different types of mechanical faults, with less than 1.5% of misclassification.}, archive = {J_EAAI}, author = {Muhammad Haseeb Arshad and Haihan Wang and Jiabao Yao and Qing Zhao}, doi = {10.1016/j.engappai.2025.111182}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111182}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rotor imbalance fault classification through unified feature fusion: Combining statistical signal processing with adaptive deep learning features}, volume = {157}, year = {2025}, }
• (2025). Feature transformation and statistical calibration for cross-domain few-shot classification. EAAI, 157, 111181. (www)

  @article{ , abstract = {The limited labeled training samples in target tasks, combined with the distribution discrepancy between the target and the model-learning source datasets, create a significant challenge in Remote Sensing Scene Classification (RSSC). Existing work attacks this problem from the perspective of feature generalizability while ignoring model bias. In this paper, we address this issue in the framework of Cross-Domain Few-Shot Classification (CDFSC) and propose an overfitting-resistant and debiased learning method, FEature Transformation and stAtistical Calibration (FETAC), to tackle this problem. In FETAC, linear and residual modules are incorporated to the backbone to transform features layer-wisely for model overfitting-resistant learning at the task level. Meanwhile, a parameter-free Lifted Laplacian Activation (LLA) unit is proposed to encourage balanced activations of features across tasks on the target domain at the dataset level. This property of balanced activations prevents the model from using only a small number of features for adaptation, thus achieving feature calibration statistically for an unbiased cross-domain feature transformation. In model architecture, FETAC is designed as a plug-and-play module on the backbone and is optimized on the target few-shot task at test time on the fly. Experiments on five distinct RSSC target domains validate the superior performance of FETAC for CDFSC tasks. Furthermore, the excellent performance exhibited by FETAC on the 1-shot CDFSC tasks further demonstrates its robust applicability in RSSC under extremely scarce labeled examples. Code will be available at https://github.com/JiaHg/fetac .}, archive = {J_EAAI}, author = {Jiafan Liu and Jin Deng and Jinrong Cui and Wei Luo}, doi = {10.1016/j.engappai.2025.111181}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111181}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature transformation and statistical calibration for cross-domain few-shot classification}, volume = {157}, year = {2025}, }
• (2025). Sustainable driving operations of urban plugin hybrid buses considering restricted emission mapping zones. EAAI, 157, 111179. (www)

  @article{ , abstract = {Road traffic is the major source of air and noise pollution. It is also one of the largest contributors to anthropogenic greenhouse gas emissions. Transport electrification can significantly reduce these negative externalities. However, electric buses do not often meet public transport requirements, due to their limited battery capacity. In contrast, plug-in hybrid electric buses offer a versatile alternative, providing zero-emission capabilities that depend on battery capacity, charging frequency, and the distribution of electric drive along a route. However, current electric drive assignment systems are oversimplified, thus not fully leveraging their potential. This work extends our previous research, which introduced a novel combinatorial optimization problem aimed at determining optimal electric drive assignment strategies for Plugin Electric Hybrid (PEH) buses. A large number of bus lines as well as zero-emission and restricted-emission zones are considered. The objectives are to maximize the buses’ electric range and minimize the overall pollution while adhering to mandatory zero-emission and restricted-emission zones. In this study, we analyze seventy bus lines from Barcelona’s urban public bus network (Spain), and tackled this large problem with two parallel implementations of the Cooperative Co-evolutionary Multi-objective Cellular Genetic Algorithm (CCMOCell). The strategies provided by both CCMOCell versions are validated against GreenK, a state of the art heuristic that only focuses on the electric range. Results demonstrate that the obtained solutions achieve up to 7.67% reduction in carbon dioxide (CO 2 ) emissions, compared to GreenK, at the cost of a slight decrease in terms of electric range, i.e. 2.28%. However, the strategy found by GreenK is unfeasible, because it exceeds the pollution threshold established for one restricted-emission zone by 635 CO 2 kilograms per day.}, archive = {J_EAAI}, author = {Marina Díaz-Jiménez and José Miguel Aragón-Jurado and Bernabé Dorronsoro and Pablo Pavón-Domínguez and Marcin Seredynski and Patricia Ruiz}, doi = {10.1016/j.engappai.2025.111179}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111179}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sustainable driving operations of urban plugin hybrid buses considering restricted emission mapping zones}, volume = {157}, year = {2025}, }
• (2025). A bearing fault diagnosis method for sample imbalance. EAAI, 157, 111171. (www)

  @article{ , abstract = {Bearing fault diagnosis of aeroengines is crucial to ensure flight safety and reduce engine maintenance costs. However, existing fault diagnosis methods face the challenge of sample imbalance, and the scarcity of some fault samples seriously affects the accuracy and reliability of the diagnostic system. To address this problem, a Variational Auxiliary Classifier Wasserstein Generative Adversarial Network (VACWGAN) is proposed to enhance scarce fault samples. This method optimizes the structure based on the Auxiliary Classification Generative Adversarial Network (ACGAN) and builds an independent classifier to enhance the generation ability of multi-target samples. At the same time, the Wasserstein distance and gradient penalty terms are combined to improve the stability during the training process, while the encoding part of the variational autoencoder is integrated to optimize the sample generation quality. This helps increase the number of scarce fault samples and enhance diagnostic accuracy. To verify the effectiveness of the VACWGAN algorithm, this paper constructed data sets with different balance ratios on the Case Western Reserve University (CWRU) and our laboratory bearing fault diagnosis test bench data sets, and applied the algorithm to enhance scarce samples to achieve data balance. The diagnostic model trained on the balanced data set shows that VACWGAN has good feasibility and practicability in handling bearing fault diagnosis with severe sample imbalance.}, archive = {J_EAAI}, author = {Zhenyu Yin and Feiqing Zhang and Cong Yin and Guangyuan Xu and Siyu Liu}, doi = {10.1016/j.engappai.2025.111171}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111171}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A bearing fault diagnosis method for sample imbalance}, volume = {157}, year = {2025}, }
• (2025). Smart grid security through fusion-enhanced federated learning against adversarial attacks. EAAI, 157, 111169. (www)

  @article{ , abstract = {The proliferation of smart grids introduces significant challenges for energy networks in managing and securing the vast data they generate. Federated Learning (FL) provides a cost-effective and privacy-preserving framework for distributed model training, addressing critical concerns around customer data privacy and security. However, FL remains vulnerable to adversarial threats, particularly data poisoning attacks, which can significantly impair model performance. This study presents a novel data poisoning attack and proposes a mitigation framework tailored for resource-constrained smart grids. The Centroid-Based Anomaly Aware Federated Averaging (CBAA-FedAvg) framework is introduced, achieving a Mean Absolute Percentage Error (MAPE) of 2.7 percent, closely aligning with baseline performance while maintaining robustness. CBAA-FedAvg integrates advanced fusion techniques, including parameter quantization (from 32-bit floating point to 8-bit fixed point) and dynamic clustering, to minimize computational complexity and optimize data processing. Furthermore, an automatic stopping criterion optimizes convergence, reducing energy consumption and computation time. Experimental results demonstrate that CBAA-FedAvg exhibits remarkable resilience to both data and model poisoning attacks, leveraging fusion strategies to enhance security and efficiency. This framework provides a scalable and effective solution for improving the fusion, security, and operational efficiency of smart grids.}, archive = {J_EAAI}, author = {Attia Shabbir and Habib Ullah Manzoor and Ahmed Zoha and Zahid Halim}, doi = {10.1016/j.engappai.2025.111169}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111169}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Smart grid security through fusion-enhanced federated learning against adversarial attacks}, volume = {157}, year = {2025}, }
• (2025). Applied an in-motion transfer alignment approach during global positioning system outages utilizing a recurrent neural network algorithm. EAAI, 157, 111167. (www)

  @article{ , abstract = {In this study, a Robust in-motion Transfer Alignment method based on the multilayer Neural Network, called RTA-NN, is proposed to correct the initial information of the slave vehicle based on the Strapdown Inertial Navigation System (SINS) and Global Positioning System (GPS) integrated navigation structure of the master vehicle, when the GPS data of the master vehicle is not available. For this purpose, first, the SINS equations of the master vehicle are extracted based on the low-grade Inertial Measurement Units (IMU) to compute the position, velocity, and attitude of the master vehicle. Then, a closed loop Kalman Filter structure is utilized to estimate the true states of the master vehicle in the presence of the GPS receiver. Next, a deep neural network, including multilayer Long-Short Term Memory (LSTM) and multilayer perceptron, is utilized to build the velocity model for the master vehicle based on the current and past samples of the master’s IMU and the SINS outputs. Finally, the performance of the proposed approach is evaluated on the navigation units of the Unmanned Surface Vehicle (USV), considered as master, and Remotely Operated Vehicle (ROV), considered as slave. The results demonstrate the effectiveness of the proposed transfer alignment approach based on the neural network, when the GPS signals are disrupted. During 100-second GPS outages, the proposed method reduces navigation errors by 0.1%, demonstrating the robustness and accuracy of the RTA-NN approach in maintaining reliable transfer alignment without GPS.}, archive = {J_EAAI}, author = {Alireza Sharifi and Ali Baniasad and Saeid Mozafari}, doi = {10.1016/j.engappai.2025.111167}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111167}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Applied an in-motion transfer alignment approach during global positioning system outages utilizing a recurrent neural network algorithm}, volume = {157}, year = {2025}, }
• (2025). Dynamical multiple polynomial-based neural networks classifier realized with the aid of dropfilter and dual statistical selection. EAAI, 157, 111164. (www)

  @article{ , abstract = {Polynomial neural networks (PNN) have emerged as an effective regression modeling methodology in computational intelligence, relying on its interpretable polynomial nodes to fit complex nonlinear data relationships and the adaptive nature of self-organizing networks. To break the bottleneck of PNN structure in the field of multi-classification, this study designs a dynamical multiple polynomial-based neural networks (DMPNN) classifier, focusing on developing a flexible polynomial network classification methodology that enhances predictive capabilities without sacrificing the advantages of PNN structures. Our approach effectively addresses the challenges of multi-class classification with uncertain class boundaries and reduces computational complexity, which is achieved through the synergy of several proposed techniques. Three key issues underpin the proposed DMPNN: (a) The integration of PNN regression models using the one-against-all strategy can provide effective and scalable solutions to multi-class classification problems, especially for uncertain class boundary issues. (b) The dual statistical selection (DSS) approach aims to eliminate redundant inputs during data processing, reduce the computational burden, and increase the variety of neural network nodes in the model neuron selection stage. (c) The synergy of regularization methods including the ℓ2 norm-based method (ℓ2-LSM) and the DropFilter, is exploited to mitigate potential overfitting in coefficient estimation and enhance the generalization capabilities of the proposed classifier. A series of ablation experiments and parameter analysis were conducted to demonstrate the stability and reliability of the proposed model. Then, we applied DMPNN to 17 publicly available datasets and two engineering applications: Recycling of black plastic wastes and phased resolved partial discharge. The performance results show that the DMPNN model outperforms five classical classifiers and four state-of-the-art (SOTA) classifiers on 78.94% of the datasets. This highlights the unique ability of the proposed DMPNN to enhance predictive accuracy while maintaining model simplicity and interpretability.}, archive = {J_EAAI}, author = {Zhen Wang and Sung-Kwun Oh and Zunwei Fu and Seok-Beom Roh and Witold Pedrycz}, doi = {10.1016/j.engappai.2025.111164}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111164}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamical multiple polynomial-based neural networks classifier realized with the aid of dropfilter and dual statistical selection}, volume = {157}, year = {2025}, }
• (2025). Multi-modal semantic feature alignment medical cross-modal hashing. EAAI, 157, 111158. (www)

  @article{ , abstract = {Medical cross-modal retrieval facilitates semantic similarity searches across various modalities of medical cases. For example, it can quickly identify pertinent ultrasound images using ultrasound reports, or retrieve matching ultrasound reports using ultrasound images. Consequently, medical cross-modal hashing retrieval technology has become a significant and pioneering area of research. However, the field encounters considerable challenges. Firstly, there are significant semantic differences between different categories of medical images, coupled with subtle and often indistinguishable visual variations. Secondly, there is a significant semantic gap between different modalities of medical data. To address these challenges, this paper presents a novel multi-modal semantic feature alignment medical cross-modal hashing (MSACH) framework. The approach begins with a staged training strategy that effectively combines modality feature extraction with hash function learning, refining low-dimensional features enriched with essential semantic information. Subsequently, a feature extraction module is carefully constructed using a transformer encoder. Three pretraining tasks enable the encoder to derive semantic information from diverse modalities. Hash function learning is then conducted, utilising manifold, balance, and linear classification network constraints to enhance the discriminative power of the hash codes, thus significantly improving the retrieval accuracy across various medical data modalities. Comprehensive experimental results from three medical datasets confirm that MSACH surpasses several leading cross-modal hashing methods in terms of precision and effectiveness.}, archive = {J_EAAI}, author = {Qinghai Liu and Qianlin Wu and Lun Tang and Liming Xu and Qianbin Chen}, doi = {10.1016/j.engappai.2025.111158}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111158}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modal semantic feature alignment medical cross-modal hashing}, volume = {157}, year = {2025}, }
• (2025). Erasure-based interaction network for red-green-blue and thermal object detection and a unified benchmark. EAAI, 157, 111155. (www)

  @article{ , abstract = {Recently, many breakthroughs have been made in the field of video object detection, but the performance is still limited due to the imaging limitations of RGB (red-green-blue) sensors in adverse illumination conditions. To alleviate this issue, this work introduces a new computer vision task called RGBT (red-green-blue and thermal) video object detection by introducing the thermal modality that is insensitive to adverse illumination conditions. To promote the research and development of RGBT video object detection, we design a novel Erasure-based Interaction Network (EINet) and establish a comprehensive benchmark dataset for this task. Traditional methods often leverage temporal information by using many auxiliary frames, and thus have a large computational burden. Considering thermal images exhibit less noise than RGB ones, we develop a negative activation function that is used to erase the noise of RGB features with the help of thermal image features. Furthermore, with the benefits from thermal images, we rely only on a small temporal window to model the spatio temporal information to greatly improve efficiency while maintaining detection accuracy. Our dataset consists of 50 pairs of RGBT video sequences with complex backgrounds, various objects and different illuminations, which are collected in real traffic scenarios. Extensive experiments on the proposed dataset demonstrate the effectiveness and efficiency of EINet. Compared with existing detectors, EINet achieves a relatively balanced performance with a detection accuracy of 46.3% and a speed of 92.6 frames per second. This project will be released to the public for free academic usage at https://github.com/tzz-ahu .}, archive = {J_EAAI}, author = {Qishun Wang and Zhengzheng Tu and Chenglong Li and Hongshun Wang and Kunpeng Wang}, doi = {10.1016/j.engappai.2025.111155}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111155}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Erasure-based interaction network for red-green-blue and thermal object detection and a unified benchmark}, volume = {157}, year = {2025}, }
• (2025). Selecting cloud providers of infrastructure as a service: A picture fuzzy symmetry point of criterion-based expert-driven model. EAAI, 157, 111132. (www)

  @article{ , abstract = {This study addresses the infrastructure as a service cloud provider selection process from a decision-making perspective, with the primary motivation to fill the existing gap in the literature regarding decision support systems for IaaS provider selection. The research aims to develop a reliable and practical decision support system that can assist organizations in improving their decision-making processes. By addressing this gap, the study seeks to enhance the effectiveness and applicability of decision-making frameworks in the context of infrastructure as a service provider selection. In pursuit of this objective, a hybrid model known as the picture fuzzy - symmetry point of criterion - mixed aggregation by comprehensive normalization technique is developed, which allows for the simultaneous evaluation of both qualitative and quantitative parameters. Qualitative criteria are assessed based on expert opinions using picture fuzzy sets and picture fuzzy Yager weighted arithmetic average aggregation operator. The picture fuzzy symmetry point of criterion method is introduced to determine criteria weighting. This aggregation operator is used to aggregate expert opinions based on qualitative parameters. The picture fuzzy mixed aggregation by comprehensive normalization technique method is proposed to rank infrastructure as a service providers. To demonstrate the practical application of the hybrid model, a case study involving the infrastructure as a service provider selection process of a company in the logistics industry is conducted. The reliability of the proposed method is validated through two sensitivity analyses based on parameters affecting the alternative rankings, while its consistency is confirmed by comparing the results with alternative methods.}, archive = {J_EAAI}, author = {Galip Cihan Yalçın and Karahan Kara and Can Saygıner and Vladimir Simic and Dragan Pamucar}, doi = {10.1016/j.engappai.2025.111132}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111132}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Selecting cloud providers of infrastructure as a service: A picture fuzzy symmetry point of criterion-based expert-driven model}, volume = {157}, year = {2025}, }
• (2025). Towards fusing gaze estimation and object prediction: What are you looking at?. EAAI, 157, 111113. (www)

  @article{ , abstract = {Gaze Object Prediction (GOP) is a novel task that aims to evaluate which object is being looked at by humans. There are many applications for this task within the image understanding topic, but unfortunately, there are not many works that address it. In this paper, we propose a model named Gaze Estimation and Object Prediction (GEOP) that performs GOP, fusing the object detection and gaze following tasks. Specifically, we use object detection features to carry out both tasks, avoiding the extra computational cost added by using different branches. In addition, we include a dual gaze regression approach to improve the gaze following performance. A deeply analysis of the influence of auxiliary prediction heads to improve the detection accuracy has been carried out to optimize the proposed architecture. In this work, we also propose a novel metric called Average Precision Bounding Box Looked at (AP BL ) that explores further the GOP task. Extensive experiments on the Gaze On Object (GOO) dataset prove the performance of the proposed model in object detection, gaze following, and GOP tasks, demonstrating that the proposal outperforms other state-of-the-art approaches on both average error (Distance and Angle) and GOP metrics.}, archive = {J_EAAI}, author = {Álvaro Nieva-Suárez and Marta Marrón-Romera and Cristina Losada-Gutiérrez and Irene Guardiola-Luna}, doi = {10.1016/j.engappai.2025.111113}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111113}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards fusing gaze estimation and object prediction: What are you looking at?}, volume = {157}, year = {2025}, }
• (2025). Physics-informed neural networks in heat transfer-dominated multiphysics systems: A comprehensive review. EAAI, 157, 111098. (www)

  @article{ , abstract = {This article presents a comprehensive review of the role of Physics-Informed Neural Networks (PINNs) in engineering applications dominated by heat transfer, synthesizing bibliometric analysis with practical engineering perspectives. PINNs, which integrate separable neural networks and parameterized multiphysics models, offer a robust framework for solving both forward and inverse heat transfer problems. These models demonstrate particular efficacy in complex multiphysics environments, such as combustion modeling and geothermal forecasting, circumventing the need for extensive mesh generation required by traditional methods. Key applications explored include thermal management in electronics, monitoring of cogeneration systems, geothermal heat production, lifespan management of batteries and transformers, food drying processes, material thermal property identification and thermal oversight in additive manufacturing. Recent advancements, including the incorporation of numerical methods, adaptive activation functions, and transfer learning, have significantly improved the practicality of PINNs. However, their training cost remains a notable drawback when compared to the efficiency of conventional numerical simulation methods for regular, large-scale simulations. Emerging trends in deep learning and probabilistic approaches suggest promising future directions, such as potential integration with Graph Neural Networks and meta-learning to enhance modeling capabilities. While traditional numerical simulation methods remain preferable for short-term, well-defined scenarios, PINNs or hybrid approaches are recommended for long-term, innovative solutions requiring adaptability. This review provides a structured framework for engineers and researchers to apply PINNs in heat transfer-dominated applications and delineates future pathways for advancing data-driven, physics-constrained modeling strategies.}, archive = {J_EAAI}, author = {Zhuang Zhao and Ye Wang and Weijian Zhang and Zhenggang Ba and Lin Sun}, doi = {10.1016/j.engappai.2025.111098}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111098}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed neural networks in heat transfer-dominated multiphysics systems: A comprehensive review}, volume = {157}, year = {2025}, }
• (2025). Shear capacity prediction and reliability analysis of corroded reinforced concrete beams using deep generative modeling and ensemble learning. EAAI, 157, 111085. (www)

  @article{ , abstract = {This study develops a machine learning framework to predict the shear capacity of corroded reinforced concrete (CRC) beams, enhancing structural reliability assessments. A Variational Autoencoder (VAE) generated a synthetic dataset of 10,000 samples, addressing the challenges of limited and varied experimental data on corrosion. Comparative analyses showed the VAE outperformed Generative Adversarial Networks (GANs) based on entropy, Kullback–Leibler divergence, and Fréchet Inception Distance (FID), indicating higher data quality and realism. Five machine learning models – Extreme Gradient Boosting (XGBoost), Gradient Boosting Machine (GBM), Categorical Boosting (CatBoost), Adaptive Boosting (AdaBoost), and a back-propagation neural network (BPNN) – were trained using this data. XGBoost demonstrated superior accuracy, achieving an R 2 of 0.96 on synthetic data and 0.85 on real data. Shapley Additive Explanations (SHAP) identified critical factors such as concrete strength and stirrup corrosion, impacting shear capacity. A reliability-based approach calibrated a global resistance factor of 1.10, ensuring CRC beams designed with the XGBoost model meet a reliability index of 3.8. This approach significantly advances predictive capabilities and reliability assessments for aging infrastructure management.}, archive = {J_EAAI}, author = {L. Simwanda and A.B David and O.B. Olalusi and I.B. Muhit and M. Sykora}, doi = {10.1016/j.engappai.2025.111085}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111085}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Shear capacity prediction and reliability analysis of corroded reinforced concrete beams using deep generative modeling and ensemble learning}, volume = {157}, year = {2025}, }
• (2025). Particle swarm optimized fuzzy extended state observer design for disturbance rejection in sliding mode attitude control of a quadrotor. EAAI, 157, 111018. (www)

  @article{ , abstract = {The paper provides a novel approach to extended state observer design for disturbance rejection in sliding mode attitude control problem of a quadrotor. The original contribution of the work lies in employing Takagi-Sugeno fuzzy logic to improve the robustness in the observer design for the proposed system. The well-known Particle Swarm Optimization algorithm is employed in the quadrotor control system to optimize the observed states and parameters of used fuzzy sets, with an aim to improve both the tracking performance and noise figure of the observer response in presence of external disturbance and parametric noise. Additional contribution includes a data-driven approach to knowledge discovery in the design of Takagi-Sugeno fuzzy observer. The present study also examines the scope of replacing traditional extended state observer by the proposed fuzzy observer in a dual channel disturbance rejection proposal of a recent seminal work. The study reveals that high frequency disturbances are better compensated by the fuzzy observer than its traditional model. Experiments undertaken with Gaussian noise as disturbance inputs further reveal that the fuzzy state observer in the presence of the existing low frequency disturbance compensator, proposed in the seminal work, exhibits relatively improved noise figure and tracking performance in comparison to the seminal work itself. The chattering in the control signal is also significantly reduced by the incorporation of the proposed Takagi-Sugeno fuzzy extended state observer in the sliding mode control architecture. Because of high robustness of the proposed extended fuzzy state observer, it exhibits amazing performance in sliding mode control architecture of quadrotors, outperforming the existing approaches in both tracking and noise rejection.}, archive = {J_EAAI}, author = {Dipanjan Konar and Amitava Chatterjee}, doi = {10.1016/j.engappai.2025.111018}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111018}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Particle swarm optimized fuzzy extended state observer design for disturbance rejection in sliding mode attitude control of a quadrotor}, volume = {157}, year = {2025}, }
• (2025). Multi-scale pseudo-labels filtering and key pixels adversarial alignment for domain adaptive object detection. EAAI, 157, 111014. (www)

  @article{ , abstract = {Domain Adaptive Object Detection (DAOD) aims to adapt a detector trained on the labeled source domain to perform effectively on the unlabeled target domain. Most existing DAOD methods commonly employ self-training strategy and adversarial learning. However, the methods based on the self-training strategy often generate unreliable pseudo-labels (e.g., missing detections and false positives) due to the absence of target domain semantics, resulting in suboptimal models. Meanwhile, aligning a large number of unimportant pixels hampers adversarial learning’s ability to capture domain-invariant semantic information. Therefore, we propose the Multi-scale Adversarial Teacher Framework (MATF) based on the common self-training framework (teacher–student framework). Specifically, to mitigate the impact of unreliable pseudo-labels, we propose the Multi-scale Negatives Filtering (MNF) module in the teacher model, which prevents missing detections from damaging the model by filtering out unreliable negative samples at multiple feature scales. In addition, we propose the Multi-scale Key-Pixel Discriminator (MKPD), which predicts the distributions of key pixels at multiple feature scales emphasizing the adversarial alignment of key pixels rather than unimportant pixels. Finally, to generate higher-quality pseudo-labels, we introduce the discriminator into the student model to reduce the model’s bias towards the source domain. Unlike most methods, we construct our method on a computationally efficient but less work-intensive one-stage detector. Extensive experiments conducted on DAOD benchmark datasets such as Cityscapes, FoggyCityscapes, KITTI, and Sim10k demonstrate the strong adaptability and effectiveness of MATF.}, archive = {J_EAAI}, author = {Xing Wei and Jiong Xia and Fan Yang and Chong Zhao and Cang Liu and Ge Wang and Yanke Chen and Yang Lu}, doi = {10.1016/j.engappai.2025.111014}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {111014}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale pseudo-labels filtering and key pixels adversarial alignment for domain adaptive object detection}, volume = {157}, year = {2025}, }
• (2025). Supervisory risk control with application to industrial drone inspection. EAAI, 157, 110970. (www)

  @article{ , abstract = {This article develops and experimentally tests a supervisory risk controller used to increase the safety of drone operations. Its task is to monitor the state of the drone and environment and to use this information to automatically change safety-critical parameters in real-time during operation. The main contribution of this article is to demonstrate how methods from the risk sciences can be combined with methods from the field of artificial intelligence to design risk awareness into automated systems. More precisely a system theoretic process analysis (STPA) is performed to identify how the system can fail. The results of the STPA are used to build a Dynamic Decision Network (DDN), which is used as an online risk model. An optimization algorithm then uses the online risk model to find the optimal parameter configuration that ensures an acceptable risk level. A case study of a tethered industrial inspection drone is considered. Through experimental trials, it is demonstrated how the supervisory risk controller is able to identify the state of the drone and the environment by combining information from multiple measurements over time and how it uses this information to modify maximum speed, safety distance, and maximum vertical acceleration such that the risk level remains acceptable. When no parameter set can ensure an acceptable risk level then a recommendation of aborting the mission is sent to the human operator.}, archive = {J_EAAI}, author = {Sverre Velten Rothmund and Christoph Alexander Thieme and Ingrid Bouwer Utne and Tor Arne Johansen}, doi = {10.1016/j.engappai.2025.110970}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {110970}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Supervisory risk control with application to industrial drone inspection}, volume = {157}, year = {2025}, }
• (2025). A new consensus method for group decision making based on mixed adjustment. EAAI, 157, 110883. (www)

  @article{ , abstract = {In the consensus process, decision makers consider compromise thresholds when adjusting their opinions to achieve more reasonable results. However, previous consensus methods based on limited compromise threshold have drawbacks such as ignoring the fact that experts have different abilities in different attributes, not considering public opinion, excessive adjustments. Due to the varying abilities of each expert in different attributes, in order to avoid excessive adjustments and develop personalized adjustment strategies, a mixed consensus adjustment method based on linguistic intuitionistic fuzzy sets is proposed, considering difference of expert ability and expert limited compromise threshold under different attributes. First of all, considering different fields that different experts are good at, this paper will determine the weight of each expert under different attributes according to the ability of experts. Secondly, the objective weight and subjective weight of attributes are calculated by public opinion and expert opinion respectively, and the comprehensive attribute weight is determined by the minimum relative entropy model. Then, based on expert ability, this paper defines the unit adjustment cost and limited compromise threshold of experts under different attributes, and then proposes a new mixed consensus adjustment method. Finally, the effectiveness and rationality of the proposed method are demonstrated through an examples of selecting the best tourist destination and comparative analysis.}, archive = {J_EAAI}, author = {Wei Yang and Yifan Lian and Jiarong Shi}, doi = {10.1016/j.engappai.2025.110883}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {110883}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new consensus method for group decision making based on mixed adjustment}, volume = {157}, year = {2025}, }
• (2025). Diabetes risk assessment model based on unbalanced public health examination data. EAAI, 157, 110877. (www)

  @article{ , abstract = {Timely risk assessment is crucial for the prevention, treatment, and management of diabetes. In this study, we propose an intelligent risk assessment model framework for diabetes based on public health examination data, focusing on both clinical examination and personal lifestyle dimensions. First, to address the multi-feature issue in examination data, a progressive correlation-based feature selection method was established to select risk features. Fourteen and six key features were selected for the clinical examination and personal lifestyle dimensions, respectively, with the features being concise, transparent, and highly interpretable. Second, to address the issue of sample imbalance and the medical focus on minority class samples, we designed an under-sampling ensemble classification iterative boosting method using light gradient boosting machine as the base classifier. This method combines the advantages of ensemble learning and undersampling, and progressively improves model performance through adaptive sampling mechanisms and multiple rounds of iterative learning. Compared to other balancing methods, our approach demonstrated superior overall performance, achieving accuracies of 89.02% and 87.53% in the clinical examination and personal lifestyle dimensions, respectively. Finally, to enhance the practicality of predictions, we designed a web-based visual risk grading scorecard. On an independent test set, the accuracies for the clinical examination and personal lifestyle dimensions reached 85.74% and 85.48%, respectively, indicating that the information loss after binning was relatively low and that the features effectively captured factors related to diabetes risk. The proposed diabetes risk assessment model framework demonstrates good practicality and lays a solid foundation for diabetes risk warning and decision support.}, archive = {J_EAAI}, author = {Liangjun Jiang and Jing Wang and Jie Xie and Zhenhua Xia and Juan Li and Haimei Gong and Lei Wang}, doi = {10.1016/j.engappai.2025.110877}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {110877}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Diabetes risk assessment model based on unbalanced public health examination data}, volume = {157}, year = {2025}, }
• (2025). A hybrid fermatean fuzzy approach based on a novel score function for solid waste management. EAAI, 157, 110876. (www)

  @article{ , abstract = {Solid waste management (SWM) is vital for sustainable development, necessitating the evaluation of diverse SWM techniques under uncertain conditions. However, existing Multi-Criteria Decision-Making (MCDM) approaches often struggle with complex uncertainty and fail to effectively combine subjective and objective criteria. To address this gap, we propose a novel hybrid method that integrates the REmoval Effect of Criterion (MEREC), Stepwise Weight Assessment Ratio Analysis (SWARA), and Operational Competitiveness Rating (OCRA) techniques, complemented by a newly developed score function under the Fermatean Fuzzy Set (FFS) setting. By fusing SWARA and MEREC, the model achieves balanced weight assessment, while OCRA precisely evaluates the alternatives. We test this approach using two case studies on plastic waste disposal and urban SWM, supported by rigorous statistical assessments such as Spearman correlation, Kendall's Tau, and Bootstrap analysis. Our findings reveal that “incineration” is consistently the optimal solution, demonstrating the method's effectiveness in handling uncertainty. Sensitivity, comparative, statistical, and runtime analyses confirm the framework's robustness, precision, and practicality, offering enhanced ranking consistency and reliability over existing MCDM techniques. The innovative FFS-based score function further refines decision-making by integrating environmental, social, and economic factors crucial for sustainability. By optimizing SWM practices, the proposed method addresses real-world complexities while supporting sustainable development goals. Moreover, it offers a flexible platform adaptable to other complex domains where uncertainty is prevalent, making a significant contribution to the broader field of MCDM and advancing sustainable resource management strategies. This integrated approach furthers robust and holistic MCDM evaluations, enabling resilient, future-oriented decision-making in SWM and beyond.}, archive = {J_EAAI}, author = {Palash Dutta and Alakananda Konwar and Rajnish Kumar and Niladri Palit}, doi = {10.1016/j.engappai.2025.110876}, journal = {Engineering Applications of Artificial Intelligence}, month = {10}, pages = {110876}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid fermatean fuzzy approach based on a novel score function for solid waste management}, volume = {157}, year = {2025}, }
• (2025). Universal multimodal aggregation network with adaptive enhancement and semantic guidance for salient object detection. EAAI, 156, 111371. (www)

  @article{ , abstract = {Salient object detection (SOD) is crucial for various computer vision applications, and recent advancements in SOD methods have shown great promise. However, most methods are designed for single modal and specific scenarios, which cannot accurately identify and detect salient objects in complex and challenging scenarios when faced with limited available information, and have poor robustness and generalization. Meanwhile, these methods cannot fully leverage the complementary information of multimodal features and lack the guidance of semantic information, which results in incomplete or blurred segmentation and affects overall detection accuracy. To address these challenges, a universal multimodal aggregation network (UMANet) with adaptive enhancement and semantic guidance for triple-modal SOD is proposed. In UMANet, red-green-blue, depth, and thermal infrared images are first simultaneously introduced to provide the enough available information. Subsequently, the adaptive hybrid feature enhancement module is established to achieve feature enhancement by adaptively extracting complementary information, and the global semantic mining module is introduced to deeply mine semantic information by progressively integrating features from multiple modalities. Next, the semantic guided cross scale deep feature fusion module is developed to efficiently achieve the fusion of enhanced multimodal features guided by global semantic information. Finally, the progressive residual decoding module is used to implement feature decoding progressively from higher to lower levels, which further refines the salient object boundaries. Extensive experiments demonstrate that UMANet achieves high detection accuracy and strong generalization, making it adaptable to various application scenarios and surpassing other state-of-the-art SOD methods.}, archive = {J_EAAI}, author = {Qiancheng Li and Chuancang Ding and Baoxiang Wang and Jun Wang and Weiguo Huang and Zhongkui Zhu}, doi = {10.1016/j.engappai.2025.111371}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111371}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Universal multimodal aggregation network with adaptive enhancement and semantic guidance for salient object detection}, volume = {156}, year = {2025}, }
• (2025). A wavelet disentanglement and topological semantic neural network for traffic flow forecasting. EAAI, 156, 111367. (www)

  @article{ , abstract = {Existing spatio-temporal models struggle to capture spatial dependencies in traffic flow that involve both local topological structures and semantic relations shaped by multiple factors. To this end, a novel Wavelet disentanglement and Topological Semantic Neural Network (WTSNet) for accurate traffic flow forecasting is presented. We employ Discrete Wavelet Transform (DWT) to decouple traffic flow into stable trends and event fluctuations. To separately model the stable trends and event fluctuations, the dual-channel spatial semantic layer is designed. It can effectively handle traffic flow affected by multiple modalities and solve the limitations of traditional models in capturing global semantic dependencies. To better capture both spatial topological dependencies and temporal dependencies the spatial topological and temporal layer is also designed in WTSNet. The designed layer can enhance the model's ability to learn both local topological dependencies and global semantic features in traffic flow. Extensive experimental results on four real public transportation datasets show that the proposed approach has better performance.}, archive = {J_EAAI}, author = {Tao Cui and Yudong Lu and Di Dong and Chongguang Ren and Zhijian Qu and Panjing Li}, doi = {10.1016/j.engappai.2025.111367}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111367}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A wavelet disentanglement and topological semantic neural network for traffic flow forecasting}, volume = {156}, year = {2025}, }
• (2025). Image steganography with high embedding capacity based on multi-target adversarial attack. EAAI, 156, 111341. (www)

  @article{ , abstract = {Deep learning-based steganography techniques utilizing generative adversarial networks have attracted considerable attention due to their ability to produce realistic images that serve as effective carriers for hidden information. However, as the capacity for embedding information increases, the quality of the generated covert images tends to decline significantly. To address these challenges and enhance both the quality of covert images and data-hiding performance, we propose a high-capacity image steganography method known as Multi-Target Adversarial Image Steganography (MTAIS). This method leverages a multi-target adversarial attack technique to effectively conceal high-capacity secret information within images. The proposed scheme adapts the fully connected layer of the recognition model and transforms the undirected adversarial attack into a directed adversarial attack targeting multiple outputs, allowing for fine-tuning of the base model without extensive retraining. We conducted comprehensive experiments to benchmark the proposed scheme against several established deep learning-based steganography schemes. The results indicate that the proposed scheme consistently outperforms its competitors across various evaluation metrics. Notably, our method preserves the quality of the cover image, ensuring visual integrity while achieving a high capacity for embedding secret information. The experimental results underscore the advantages of the proposed scheme in terms of performance and efficiency, establishing it as a robust solution for high-capacity image steganography.}, archive = {J_EAAI}, author = {Xiaolong Liu and Minghuang Shen and Jiayi Liu and Qiwen Wu}, doi = {10.1016/j.engappai.2025.111341}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111341}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Image steganography with high embedding capacity based on multi-target adversarial attack}, volume = {156}, year = {2025}, }
• (2025). Compressors quantitative health assessment based on local spatial-temporal relations and moving graph convolutional network. EAAI, 156, 111330. (www)

  @article{ , abstract = {Health assessment plays a crucial role in ensuring compressor safety, improving performance, and extending the operating range. While current research on the condition of compressors has predominantly focused on fault diagnosis and condition estimation, methods for quantitatively assessing its health state are still relatively lacking. Existing condition assessment methods are unsuitable for dealing with time-series data with spatial coupling characteristics, thereby hindering the ability to achieve reliable quantitative assessments. To address this issue, this paper proposes a local spatial-temporal convolution-based moving Graph Convolutional Network (LSTC-GCN). Firstly, by considering additional temporal connections, the structure of the spatial-temporal graph is constructed, decomposing long time series with spatial relationships into local spatial-temporal relations. Then, a spatial-temporal graph convolutional unit is developed to capture high-level features in these local spatial-temporal graphs through moving convolutions, followed by attention pooling for feature simplification and concentration. The model was validated using simulated operational data from compressors, and efficiency was used as a health factor to establish assessment standards. The LSTC-GCN model achieves recall, precision, F1-score, and accuracy of 87.1 %, 89.6 %, 85.9 % and 92.4 %, respectively. Validation was performed in an online assessment scenario, where the assessment of the method was not affected when the number of monitorable parameters was reduced. This study provides insights for health assessment of compressors and similar equipment.}, archive = {J_EAAI}, author = {Kanru Cheng and Yuzhang Wang and Jiao Li and Yueheng Wang and Houqi Wei and Chaoran Yang}, doi = {10.1016/j.engappai.2025.111330}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111330}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Compressors quantitative health assessment based on local spatial-temporal relations and moving graph convolutional network}, volume = {156}, year = {2025}, }
• (2025). Hybrid and multiple ensemble metamodel-based evaluation for operating tunnel performance in three-dimensional spatially variable soils. EAAI, 156, 111321. (www)

  @article{ , abstract = {In recent years, the Random Finite Element Method (RFEM) has gained prominence in geotechnical engineering for assessing the inherent spatial variability in the mechanical properties of both natural and processed soils. Nevertheless, RFEM often demands more extensive computational resources than deterministic finite element analysis, as it is coupled with Monte-Carlo simulations (MCS). To mitigate this computational burden, metamodeling techniques have emerged as a popular approach. This paper proposes a novel and hybrid Support Vector Regression (SVR) metamodel by fusing the RFEM analysis. The metamodel can efficiently generate the original finite element method predicted quantities with limited training by utilizing input random field features, which encapsulate high-dimensional information pertaining to spatially variable soil stiffness parameters. Furthermore, based on ensemble learning, the Bagging and Adaboost algorithms were used to develop a multiple SVR (M-SVR) ensemble learning metamodel to enhance prediction reliability. Simultaneously, considering the limitation that machine learning prediction can only provide a single value, the prediction results with confidence intervals based on Bagging ensemble algorithms were also developed to quantify the uncertainty of machine learning predictions in regression analysis. The consistency between SVR and M-SVR predictions and RFEM calculations is demonstrated through a problem involving the failure probability evaluation of tunnel longitudinal performance induced by ground surface surcharge in three-dimensional spatially variable soils. The substantial improvement in efficiency with the adoption of the SVR and M-SVR, as compared to RFEM, underscores the immense potential of machine learning algorithms in conducting geotechnical reliability analyses involved with spatial variability.}, archive = {J_EAAI}, author = {Ning Tian and Jinsong Huang and Jian Chen and Kaiwei Tian and Peng Wu}, doi = {10.1016/j.engappai.2025.111321}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111321}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid and multiple ensemble metamodel-based evaluation for operating tunnel performance in three-dimensional spatially variable soils}, volume = {156}, year = {2025}, }
• (2025). Multi-class classification of paint/coating defects using transfer learning. EAAI, 156, 111320. (www)

  @article{ , abstract = {Paints and Coatings are integral to various industries, offering not only aesthetic enhancements but also protection against environmental factors like ultraviolet radiation, corrosion, and wear. However, defects compromising coating quality are common, and manual inspection methods are often inefficient, error-prone, and labour-intensive. In this study, we address these challenges by leveraging vision-based deep learning techniques for automated classification of 10 most common classes of industrial paints and coating defects. Although our dataset was limited to just 100 images (10 per class) for few-shot and 332 images (30–35 per class) for moderate-shot scenario, Vision Transformer (ViT) model achieved an impressive average accuracy of 86 % and 93 % respectively, outperforming Visual Geometry Group Network (VGG) (82 %), Densely Connected Convolutional Network (DenseNet) (91 %), and Efficient Network (EfficientNet) (79 %) in the moderate shot classification. These models’ robustness was further evaluated using a challenging non-trivial augmented test dataset designed to mimic real-world scenario. These test images were generated by mapping them onto curved surfaces, introducing distortions, and capturing from angled perspective. Remarkably, these models demonstrated robust performance with a maximum accuracy drop of only 7 %. To enhance adaptability to novel and unseen defects, Meta-Learning was also employed, improving the generalization capability of the models across both familiar and previously unseen defect types. This study offers valuable guidance for researchers working on vision-based image classification applications and provides significant insights into the potential of Artificial Intelligence (AI) driven quality control systems for the paint/coating industry, as well as in manufacturing, automotive, aerospace and civil infrastructure maintenance.}, archive = {J_EAAI}, author = {Priyankkumar Dhrangdhariya and Parvesh Saini and Soumyadipta Maiti and Beena Rai}, doi = {10.1016/j.engappai.2025.111320}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111320}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-class classification of paint/coating defects using transfer learning}, volume = {156}, year = {2025}, }
• (2025). Machine learning prediction of flow-induced scour depth around isolated pier comparing stand-alone and ensemble models. EAAI, 156, 111318. (www)

  @article{ , abstract = {Scour at bridge piers significantly contributes to bridge collapses worldwide, prompting interest in accurate scour estimation. Recent research favors machine learning (ML) over classical models for accuracy. This study aims to develop a robust framework to predict scour depth by evaluating various ML models and their ensemble to identify the best methods. It also seeks to analyze the impact of input parameters through the interpretable ML method and sensitivity analysis. This study analyzes 871 datapoints from 51 sources in clear-water conditions, identifying eight key parameters through dimensional analysis, like flow intensity, shallowness; sediment coarseness, gradation; constriction ratio; time; Froude and Reynolds numbers. Various methods, including subset regression, permutation importance, sequential feature selection, and recursive feature elimination, are used to evaluate how three different combinations of 6–8 parameters improve model accuracy. The study replaces the traditional randomized-splitting with stratified-splitting for more representative analysis. It rigorously compares 42 stand-alone, 18 ensembles and 11 empirical models to improve representativeness. Categorical Boosting performs best among stand-alone models outperforming all empirical models. Stacking regressor, combining the best seven stand-alone models with six input parameters, outperforms Categorical Boosting with the highest efficiency. The best stand-alone model demonstrates physical consistency of hydraulic flow-structure interactions using SHapley Additive exPlanations. Sensitivity analysis using Permutation Importance reveals flow shallowness as the most important parameter, closely following nondimensional-time. This study introduces ensemble ML models to improve scour prediction accuracy and provide insights into physical aspects of scouring. It assists bridge designers and technicians globally in understanding how each parameter impacts predictions.}, archive = {J_EAAI}, author = {Buddhadev Nandi and Gaurav Patel and Subhasish Das}, doi = {10.1016/j.engappai.2025.111318}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111318}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning prediction of flow-induced scour depth around isolated pier comparing stand-alone and ensemble models}, volume = {156}, year = {2025}, }
• (2025). An effective hybrid algorithm with log-sum regularization and modified momentum for restricted boltzmann machine. EAAI, 156, 111317. (www)

  @article{ , abstract = {Restricted Boltzmann Machine (RBM) is a strong representation and generative model for unsupervised feature extraction in deep learning. Deep architectures can effectively enhance the capability for image feature expression in image recognition. However, deep architectures have two main disadvantages: 1) they are prone to overfitting, and 2) they are computationally expensive. We propose an effective hybrid algorithm with log-sum regularization and modified momentum for RBM, aiming to overcome overfitting and enhance computational efficiency. Firstly, we improve the gradient approximation by incorporating bias modification. Secondly, we incorporate the log-sum norm into RBM to achieve sparse representation, allowing the sparsity of each hidden unit to be learned adaptively. Thirdly, different combinations of the momentum term are used in the pre-training and fine-tuning phases of the RBM to accelerate the network convergence. Recognition experiments on the MNIST, CMU-PIE, and Extended Yale B databases demonstrate the advantages of our method over other deep learning algorithms in terms of recognition performance and computational efficiency. Denoising experiments for removing the random noise indicate good image denoising capabilities of our proposed method over other unsupervised learning algorithms in natural image data, simulated seismic data, and real seismic data. Based on these image recognition and image denoising results, this novel combination of the improved gradient approximation with modified momentum and the log-sum regularization for training RBM is an effective method for learning robust generative and representation models. Additionally, our research provides a reference for using RBMs to develop engineering applications of artificial intelligence.}, archive = {J_EAAI}, author = {Huihui Shen and Hongwei Li and Zhiguo Gong and Qing Li}, doi = {10.1016/j.engappai.2025.111317}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111317}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An effective hybrid algorithm with log-sum regularization and modified momentum for restricted boltzmann machine}, volume = {156}, year = {2025}, }
• (2025). Underwater target pose recognition: A deep learning approach based on sonar signals. EAAI, 156, 111309. (www)

  @article{ , abstract = {Underwater target pose recognition has become a significant research focus in ocean exploration, resource investigation, and military applications. Traditional methods based on physical models and rule-based matching struggle with noise interference and dynamic underwater conditions. In this study, we propose an artificial intelligence-based approach, employing a multi-task deep learning model to enhance underwater target pose estimation. A synthetic sonar frequency response dataset was generated by simulating the backscattering characteristics of ellipsoidal targets under various incident angles. A multi-layer neural network was designed to simultaneously perform ellipsoid ratio classification and incidence angle estimation, utilizing a shared feature extraction framework for joint classification and regression learning. Experimental results demonstrate that the proposed model achieves a classification accuracy of 100 % under standard conditions and a mean absolute error (MAE) of 0.0595° in angle estimation. Even under significant noise interference (10 % noise added), the model maintains a classification accuracy of 99.5 % and an MAE of 0.3805°. In extreme conditions with high noise and strong signal attenuation, the model achieves 99 % classification accuracy and an MAE of 0.4328°, demonstrating its robustness and adaptability to complex underwater environments. These findings demonstrate that deep learning serves as a robust alternative to traditional physics-based modeling, significantly enhancing the precision and reliability of underwater target recognition. Future research will integrate real-world sonar data and explore advanced AI architectures such as convolutional neural networks (CNNs) and Transformers for enhanced feature extraction and generalization.}, archive = {J_EAAI}, author = {Jikai Yang and Ziyan Gu and Peijun Li and Zihan Li and Wei Li}, doi = {10.1016/j.engappai.2025.111309}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111309}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Underwater target pose recognition: A deep learning approach based on sonar signals}, volume = {156}, year = {2025}, }
• (2025). A systematic review on advancement and challenges in multi-fault diagnosis of rotating machines. EAAI, 156, 111306. (www)

  @article{ , abstract = {Industries are heavily dependent on rotating components, which are gradually deteriorate over time, resulting in system failure and financial losses. Numerous investigations have been conducted utilizing various methodologies to identify incipient faults in rotating components. But more focus is given to single component fault diagnosis. It is highly unlikely that one component fault will occur at a time in real-world situations. Multiple components and sub-component faults take place simultaneously in a system. The signature of multi-fault conditions is chaotic, and diagnosis is difficult with signal processing techniques as fault information is modulated. Some studies have been done to identify the multi-faults of rotating machines using different approaches. However, a dedicated literature review on multi-fault diagnosis has not been presented properly. Therefore, this paper discusses the advancement in multi-fault diagnosis by implementing a systematic literature review using the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. This article reports the article selection strategy, inclusion and exclusion criteria, etc. in a transparent manner. It also discusses the trends, techniques and approaches for multi-fault diagnosis. It is found that, bearing and gear multi-fault are simulated a lot with constant speed conditions. Vibration based analysis with data driven approaches are used enormously for multi-fault diagnosis. After analysis, this article reports the research gaps, and challenges and gives recommendations for the wide audience to work in the future. This systematic literature review will be helpful for researchers to find solutions that are more practical and can be implemented in industrial scenarios.}, archive = {J_EAAI}, author = {R.K. Mishra and Anurag Choudhary and S. Fatima and A.R. Mohanty and B.K. Panigrahi}, doi = {10.1016/j.engappai.2025.111306}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111306}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A systematic review on advancement and challenges in multi-fault diagnosis of rotating machines}, volume = {156}, year = {2025}, }
• (2025). Bearing life prediction via a novel temporal convolutional network with embedded uncertainty quantification. EAAI, 156, 111303. (www)

  @article{ , abstract = {This paper presents a novel Modernized Temporal Convolutional Network (MTCN) framework with embedded uncertainty quantification for high-accuracy bearing remaining useful life (RUL) prediction. The proposed architecture introduces three key innovations: (1) A redesigned temporal convolution module leveraging grouped depthwise/pointwise convolutions to enable cross-dimensional feature interaction, capturing comprehensive degradation patterns from raw and derived signal dimensions; (2) A gated attention unit (GAU) that efficiently models temporal dependencies with 50 % lower computational cost than standard self-attention mechanisms; (3) A dual uncertainty quantification system combining epistemic uncertainty and aleatory uncertainty to generate reliable prediction intervals. Validated on two widely recognized public bearing degradation datasets, our framework achieves state-of-the-art performance, demonstrating 32.3 % and 36.5 % reductions in Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), respectively, compared to existing methods, while providing quantifiable confidence bounds critical for industrial decision-making. The integration of cross-dimensional learning, lightweight attention, and probabilistic modeling establishes a new paradigm for trustworthy predictive maintenance in rotating machinery.}, archive = {J_EAAI}, author = {Yang Ge and Guodong Sun and Benlian Xu}, doi = {10.1016/j.engappai.2025.111303}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111303}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bearing life prediction via a novel temporal convolutional network with embedded uncertainty quantification}, volume = {156}, year = {2025}, }
• (2025). Research on grinding force prediction of spiral bevel gear based on generalized regression neural network and undeformed grinding chips. EAAI, 156, 111301. (www)

  @article{ , abstract = {This study is rooted in a generalized regression neural network (GRNN) and undeformed grinding chips to predict the grinding force of spiral bevel gears. Firstly, establish a grinding force model for an individual abrasive particle, and integrate it with the dynamic number of abrasive particles in the grinding wheel to obtain a spiral bevel gear grinding force model based on undeformed grinding chips. Secondly, using the duplex spread blade design technique, the spiral conical gear blank geometric dimensions, machine tool adjustment parameters, and tool specifications are calculated, and the simulation of milling and grinding is carried out in three-dimensional solid modeling software. The undeformed chips and related parameters are obtained through Boolean intersection. Due to the complexity of calculation and simulation in the development procedure of spiral bevel gears, the GRNN is established to predict the undeformed grinding chip parameters of non-simulated gears. Finally, each group's comprehensive coefficient of cutting and friction forces were calculated through grinding experiments at different generating motion speeds. Predict these coefficients through experimental data. The results showed that the maximum standard deviation between the expected grinding force and the experimental value was 3.3260, which verifies the method's effectiveness. The GRNN algorithm can help select process parameters more reasonably, avoid tooth surface burns, and improve machining efficiency.}, archive = {J_EAAI}, author = {Nan Liu and Jiang Han and Xiaoqing Tian and Lian Xia and Minglei Li and Rui Xue}, doi = {10.1016/j.engappai.2025.111301}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111301}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on grinding force prediction of spiral bevel gear based on generalized regression neural network and undeformed grinding chips}, volume = {156}, year = {2025}, }
• (2025). A physical information guided method for bridge underwater crack detection based on two-stage pre-training learning with scarce samples. EAAI, 156, 111293. (www)

  @article{ , abstract = {This paper proposes an innovative two-stage pre-training learning method guided by physical information for automatic detection of bridge underwater cracks under scarce sample conditions. Addressing the critical challenges of limited training data and complex underwater environmental interference, this paper uniquely integrates transfer learning with domain-specific physical insights to enhance detection accuracy. The key contributions include: (1) A novel two-stage pre-training framework that sequentially learns underwater environmental features from a large-scale dataset and crack-specific features from terrestrial crack images, effectively bridging domain gaps while preserving discriminative characteristics; (2) A physics-guided DeepLabv3+ architecture enhanced with skip connections, Convolutional block attention module (CBAM), and a hybrid Tversky-Dice loss function to address class imbalance and boundary ambiguity in turbid underwater conditions; (3) First-time integration of local standard deviation maps as physical guidance to amplify crack-related intensity variations, enabling precise segmentation even with minimal annotated samples. Experimental results demonstrate state-of-the-art performance, achieving an accuracy of 96.57 %, a mean IoU (Intersection over Union) of 0.95, and an F1-Score of 0.96, outperforming baseline methods by 10.8 % in accuracy and 0.43 in IoU. The method's robustness is further validated against some mainstream segmentation models, showing significant advantages in both precision and computational efficiency. This work provides a paradigm for infrastructure health monitoring in data-scarce, environmentally challenging scenarios.}, archive = {J_EAAI}, author = {Shuai Teng and Airong Liu and Junchao Yang and Zuxiang Situ and Bingcong Chen and Jialin Wang and Zhihua Wu and Jiyang Fu}, doi = {10.1016/j.engappai.2025.111293}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111293}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A physical information guided method for bridge underwater crack detection based on two-stage pre-training learning with scarce samples}, volume = {156}, year = {2025}, }
• (2025). Artificial intelligence-driven models for predicting mechanical properties of low-emission microwave-cured geopolymer mortar. EAAI, 156, 111291. (www)

  @article{ , abstract = {This study reveals the unprecedented potential of artificial intelligence (AI) models in accurately predicting the mechanical properties of microwave-cured geopolymer mortars, thereby addressing a critical gap in the integration of AI and materials science. Furthermore, applying advanced algorithmic structure and machine learning is an unexplored area in extant literature. Four parameters: conventional curing period (time- NH), conventional curing temperature (temp-NH), microwave power (W), and microwave curing period (time-MW) are considered to generate a comprehensive dataset to predict compressive strength (CS) and flexural strength (FS). Four AI models have been adopted and rigorously compared: deep learning neural network (DL-NN), probabilistic neural network (PNN), radial basis function neural network (RBF-NN), and support vector machine (SVM). The performance was evaluated using various statistical matrices and visualization graphs. The findings showed that the DL-NN model performs exceptionally well in predicting compressive and flexural strengths, with correlation coefficient (R) values of 0.966 and 0.931, mean absolute error (MAE) of 3.544 MPa and 0.990 MPa, and root mean square error (RMSE) of 5.856 MPa and 1.442 MPa, respectively. These results demonstrate the model's ability to handle the complex, non-linear relationships inherent in the data. Meanwhile, the PNN model ranked second, with R values of 0.930 and 0.833, MAE values of 5.151 MPa and 1.566 MPa, and RMSE values of 7.947 MPa and 2.089 MPa, respectively. Furthermore, the carbon dioxide (CO 2 ) emissions and embodied energy were investigated. Finally, a sensitivity analysis was conducted to assess the relative importance of each parameter on the mechanical properties.}, archive = {J_EAAI}, author = {Faidhalrahman Khaleel and Haitham Abdulmohsin Afan and Alaa H. AbdUlameer and Abdulrahman S. Abdullah and Gökhan Kaplan and Cengiz Duran Atiş}, doi = {10.1016/j.engappai.2025.111291}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111291}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence-driven models for predicting mechanical properties of low-emission microwave-cured geopolymer mortar}, volume = {156}, year = {2025}, }
• (2025). Hybrid aero-engine performance modeling to enable real-time capability using physics-based analysis and machine learning. EAAI, 156, 111288. (www)

  @article{ , abstract = {The ability to achieve rapid and efficient computation is critical for real-time analysis or onboard implementation of aero engine performance models. This study presents a hybrid aero engine performance modeling approach that combines the accuracy of component map-based zero-dimensional cycle analysis with the speed of machine learning. Traditional models using iterative solvers like Newton-Raphson are computationally intensive and prone to convergence issues. To address this, a feedforward neural network was trained on data from diverse steady-state and dynamic conditions to replace the iterative solver. The machine learning -based solver reduced execution time by approximately 90 % while maintaining predictive accuracy. Applied to the F404 engine, the proposed method showed high agreement with conventional models under steady-state conditions and acceptable performance under dynamic scenarios. The distinct advantages of this integrated approach include significant computational savings, enhanced adaptability to operational changes, and improved stability, supporting real-time, onboard performance analysis and digital twin applications.}, archive = {J_EAAI}, author = {Sangjo Kim}, doi = {10.1016/j.engappai.2025.111288}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111288}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid aero-engine performance modeling to enable real-time capability using physics-based analysis and machine learning}, volume = {156}, year = {2025}, }
• (2025). Application of latent variable models for hidden pattern identification and machine learning prediction improvement in structural engineering. EAAI, 156, 111282. (www)

  @article{ , abstract = {Although machine learning models have been widely used in structural engineering, they require predetermined features as input. The input features are usually determined based on observations from previous studies and engineer's experience; therefore, they cannot always guarantee an optimal prediction outcome. To alleviate this critical issue, latent variable models were used in this study. The latent variable models allow for a better machine learning model prediction owing to the incorporation of less correlated latent variables that consider hidden patterns in unobserved data. The superiority of the proposed latent variable modeling approach was explored for two classification problems: (i) failure mode identification of reinforced concrete columns and (ii) seismic damage state identification of bridges. The best latent variable and machine learning model combination for each classification problem was identified based on 10-fold cross-validation. Sparse principal component analysis proved to be the most effective approach for dimensionality reduction. It was found that the use of latent variables can improve the machine learning-based classification accuracy by up to 8 % compared with conventional machine learning models while reducing input dimension by up to ten times significantly improving machine learning training time efficiency.}, archive = {J_EAAI}, author = {Sergei Shturmin and Sujith Mangalathu and Jong-Su Jeon}, doi = {10.1016/j.engappai.2025.111282}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111282}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Application of latent variable models for hidden pattern identification and machine learning prediction improvement in structural engineering}, volume = {156}, year = {2025}, }
• (2025). A data-driven sustainable scheduling model for dispatch-steerable last-mile delivery systems with negotiable time windows. EAAI, 156, 111280. (www)

  @article{ , abstract = {With the surge in e-commerce, last-mile delivery (LMD) faces challenges in meeting the dynamic arrival of large-scale customer demand. Existing research considers traditional, static, single or two-echelon, and inflexible vehicle routing-based dispatch strategies, yielding inefficiencies while the sustainability impacts remain unclear. This paper proposes a novel tri-objective mathematical model for dispatch-steerability in sustainable LMD systems by integrating insourced and crowdsourced transport modes with heterogeneous capacities. Deliveries originate from main dispatch points (MDPs) or dynamically generated alternative dispatch points (ADPs), adapting to stakeholder, driver, and customer preferences via negotiable timing through temporary rejections or incentives. The model optimizes economic efficiency (minimizing routing, rejection, and negotiation costs), environmental impact (reducing underutilization of transport modes), and social responsibility (balancing driver workload and just-in-time deliveries). A data-driven density-based spatial clustering of applications with noise (DBSCAN) algorithm identifies ADP locations, enhancing the adaptability of the decisions. Scalability is achieved via a rolling horizon strategy, an improved augmented epsilon-constraint method, and multi-objective meta-heuristics, including a genetic algorithm (GA) and simulated annealing (SA) featuring an archive and grid mechanism with a tailored representation scheme. Results show a 200 % increase in Pareto-optimal solutions as demand rises, and that a dispatch steerable LMD system with ADPs offers 150 % more diverse delivery options than MDP-only systems. Pareto surface analysis underscores LMD adaptation trends and the need for multi-criteria decision-making in sustainable LMD systems.}, archive = {J_EAAI}, author = {Keivan Tafakkori and Reza Tavakkoli-Moghaddam and Ali Siadat}, doi = {10.1016/j.engappai.2025.111280}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111280}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A data-driven sustainable scheduling model for dispatch-steerable last-mile delivery systems with negotiable time windows}, volume = {156}, year = {2025}, }
• (2025). SAFE-net: Multi-head attention enhanced framework for defect detection in anti-corrosion coatings on steel structures. EAAI, 156, 111278. (www)

  @article{ , abstract = {Steel structures exhibit poor corrosion resistance and require anti-corrosion coatings for protection. Coating aging can lead to reduced or even ineffective anti-corrosion effects, so maintenance of their anti-corrosion coatings is crucial. Computer vision offers efficient and accurate analysis of visual data, but detecting defects in anti-corrosion coatings remains challenging due to low contrast, subtle background differences, varied defect types, image noise, and environmental interference. Existing networks often struggle to address these complexities effectively. In this context, this study proposes a You Only Look Once version 8 (YOLOv8)-driven SAFE-Net (Steel Anti Corruption Fault Evaluation Network) based on multiple attention mechanisms for detecting defects in anti-corrosion coatings on steel structures, and constructs a dataset for defect detection in in-service steel structure anti-corrosion coatings for model validation. The proposed framework integrates a multi-scale deformable attention mechanism into the Concatenate-to-Fuse module of the Backbone, aiming to improve the detection performance for small-scale targets. The Head adopts a Dynamic Head that integrates the self-attention mechanism to enhance scale perception, spatial perception, and task perception. The classification loss function is optimized by introducing the Slide Loss to address the challenge of imbalanced samples in the dataset, while Shape Intersection over Union refines the bounding box regression for better multi-scale detection. SAFE-Net supports multi-defect detection and demonstrates strong generalization. Numerical case studies indicate that, compared with the YOLOv8 baseline model, the improved algorithm improves the accuracy by 5.82 %, recall by 5.01 %, and mean Average Precision index by 4.78 %, respectively.}, archive = {J_EAAI}, author = {Yue Yu and Shouchao Jiang and Yijun Wang and Shaojun Zhu}, doi = {10.1016/j.engappai.2025.111278}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111278}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SAFE-net: Multi-head attention enhanced framework for defect detection in anti-corrosion coatings on steel structures}, volume = {156}, year = {2025}, }
• (2025). Online diagnosis of weak welds in busbar laser welding based on photoelectric signals and machine learning. EAAI, 156, 111277. (www)

  @article{ , abstract = {Busbars are essential conductive components in power systems, laser welding is widely used for their electrical connections due to its advantages of high precision, high efficiency, and a low heat-affected zone. However, weak weld defects are prone to occur in laser welding of busbars due to factors such as laser power fluctuations, focal shift, surface contamination, and overlap gap instability. Weak weld defects can reduce joint strength, increase contact resistance, and affect the performance and safety of the equipment. This study presents a weak weld diagnosis method for laser welding of busbars driven by machine learning based on photoelectric signals. The proposed method introduces a Reference-Enhanced Signal Feature Definition method, which improves diagnostic accuracy by utilizing multi-band data segmentation and the calculation of the Reference Deviation Coefficient Feature Matrix. This feature definition method has advantages in adapting to different process conditions and improving the recognition rate of minor defects. On this basis, the random forest-based weak weld defect diagnosis model was constructed. The method not only detects the presence of weak weld defects in busbar laser welding but also diagnoses the factors inducing these defects, providing effective support for rapid identification of problem sources and targeted interventions in industrial production. Mechanistic analysis reveals that different types of weak weld defect-inducing factors exhibit significant differences in their photoelectric signal characteristics. Experimental results show that the model achieved an average detection accuracy of 99.20 % for weak weld defects across 1427 welding joints, with an average accuracy of 98.75 % in defect-inducing factors diagnosis.}, archive = {J_EAAI}, author = {Leshi Shu and Yifan Zhou and Wendi Wu and Ping Jiang and Yilin Wang}, doi = {10.1016/j.engappai.2025.111277}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111277}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Online diagnosis of weak welds in busbar laser welding based on photoelectric signals and machine learning}, volume = {156}, year = {2025}, }
• (2025). Maintenance decision-making model for gas turbine engine component considering the inspection threshold and partial repair. EAAI, 156, 111269. (www)

  @article{ , abstract = {Traditional inspection and maintenance (I&M) plans for gas turbine blades are expensive and over-conservative, as they involve replacing all blades at predetermined operation times. Developing an I&M strategy that accounts for the state of components is crucial to ensure system reliability while minimizing cost. The state of a component can be partially observed based on the inspection capability, which is defined by the detection threshold. However, accurately modeling repair methods used in the field is challenging due to the extensive experimental data. Therefore, perfect repair models based on component replacements have been widely implemented. This study proposes a maintenance decision-making model that considers nondestructive inspection thresholds and a partial repair method using blending to minimize maintenance costs. The maintenance decision-making model reduces costs by using improved inspection capabilities compared to traditional time-based maintenance strategies. Additionally, integrating blending repairs into the maintenance policy enables additional cost savings over models relying primarily on replacement.}, archive = {J_EAAI}, author = {Hongseok Kim and Do-Nyun Kim and Dooyoul Lee}, doi = {10.1016/j.engappai.2025.111269}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111269}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Maintenance decision-making model for gas turbine engine component considering the inspection threshold and partial repair}, volume = {156}, year = {2025}, }
• (2025). Large model-embedded deep reinforcement learning for stability control of multiband power system stabilizer with high penetration distributed energy sources. EAAI, 156, 111268. (www)

  @article{ , abstract = {With the “Net-Zero” target of China in full swing, various fields have made corresponding measures, and the top priority of the electrical industry is to build renewable systems based on renewable energy. Currently, the distributed grid integration of new energy sources with high penetration rates and the diversification of renewable energy sources bring great challenges to the stability control of multiband renewable system stabilizers. To improve the stability control and control accuracy of multiband power system stabilizers (MB-PSS), a large model-embedded deep reinforcement learning is proposed in this work, named attention-embedded Transformer deep deterministic policy gradient (ATDDPG). The ATDDPG incorporates attention and Transformer mechanisms into the two neural networks of deep deterministic policy gradient (DDPG); the large model can quickly find important information in complex environments of attention mechanism is utilized to ameliorate the problem of slow learning speed and sometimes poor control accuracy. The active rotor angle deviation is reduced by ATDDPG at least 10.94 % in three-phase ground fault, three-phase inter-phase fault, and two-phase ground fault under the two-area four-machine system; the voltage stabilization time of multiband renewable system stabilizer is faster by at least 12.54 %; and integral of time multiplied by the error metrics of ATDDPG is 26.30 % smaller than those of DDPG.}, archive = {J_EAAI}, author = {Linfei Yin and Yiran Zhao}, doi = {10.1016/j.engappai.2025.111268}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111268}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Large model-embedded deep reinforcement learning for stability control of multiband power system stabilizer with high penetration distributed energy sources}, volume = {156}, year = {2025}, }
• (2025). Artificial neural networks for mass transfer and bioconvection analysis in radiative eyring-powell flow over a convective cylinder surface: Application to microbial fuel cells. EAAI, 156, 111256. (www)

  @article{ , abstract = {Aim and significance of the present study This study explores bioconvection in fluid flows in circular cylinders; it adds new features and broadens the scope of applications, such as microbial interactions in cylinder flows. Assumptions considered in the present study The governing nonlinear partial differential equations were framed considering thermal radiative, Eyring Powell, mixed convection, chemical processes, viscous dissipation, and motile microbes, along with no-slip, convective boundary conditions, and these governing equations were transformed into dimensionless forms using similarity transformations. Methodology The partial differential equations solver function in the Maple version 2024 program is used to derive numerical solutions, and artificial neural networks are used to forecast true values using the matrix laboratory software. Conclusion This research explores several parameter features to illustrate potential technological uses. Artificial Neural Networks could predict the truth values accurately, with coefficient of determination almost approaching 1, representing the prediction accuracy and results shown in Table-1 were obtained using the Levenberg–Marquardt algorithm with backward propagation through matrix laboratory; for the majority of findings, the discrepancy between the truth values and anticipated values is less than 5 %, suggesting that both approaching very close. This demonstrates how accurately and closely the Artificial Neural Networks forecasts match the numerical results; Biotechnology, electrical engineering, cancer treatment, missile technology, and biomedicine are all significantly impacted by this research.}, archive = {J_EAAI}, author = {Maddina Dinesh Kumar and C.S.K. Raju and Kiran Sajjan and Gurram Dharmaiah and Nehad Ali Shah and Se-Jin Yook}, doi = {10.1016/j.engappai.2025.111256}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111256}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial neural networks for mass transfer and bioconvection analysis in radiative eyring-powell flow over a convective cylinder surface: Application to microbial fuel cells}, volume = {156}, year = {2025}, }
• (2025). Cloud classification and cloud cover estimation using hybrid deep kronecker network- ResNeXt. EAAI, 156, 111254. (www)

  @article{ , abstract = {Clouds have a huge effect on the weather, climate, and balancing energy on the Earth's surface. Accurate cloud classification is essential as different clouds have various radiation effects and estimating the cloud cover can assist in accurate weather prediction. In this paper, a Hybrid Deep Kronecker Network- Residual Networks with Aggregated Transformations is devised for cloud classification and cloud cover estimation. At first, the input image is attained from the dataset, and a Kalman filter is employed to preprocess the image. Later, cloud segmentation is performed through Swin-Unet for deriving the cloud cover. Thereafter, features, such as Entropy with Median Binary Pattern, and statistical features, like skewness, mean, standard deviation, and variance are mined. Subsequently, the generated features and input image are fed to the established model for cloud classification. The proposed model is formulated by merging Residual Networks with Aggregated Transformations and Deep Kronecker Network. Lastly, the classified output and preprocessed output are fed to the cloud cover estimation module, where cloud cover is estimated using the proposed model. Furthermore, the proposed model is examined using metrics, such as True Positive Rate, accuracy, and True Negative Rate. From the experiment, the proposed model verified the accuracy of 0.915. Also, the True Negative Rate of 0.923, and True Positive Rate of 0.896 are obtained higher.}, archive = {J_EAAI}, author = {Gujanatti Rudrappa and Nataraj Vijapur}, doi = {10.1016/j.engappai.2025.111254}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111254}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cloud classification and cloud cover estimation using hybrid deep kronecker network- ResNeXt}, volume = {156}, year = {2025}, }
• (2025). Coverage path planning of bridge inspection with unmanned aerial vehicle. EAAI, 156, 111253. (www)

  @article{ , abstract = {To address the low efficiency of bridge inspection using Unmanned Aerial Vehicle (UAV) due to redundant inspection viewpoints and frequent sharp maneuvers, this paper proposes a novel coverage path planning model that leverages artificial intelligence methods. Firstly, to minimize viewpoint redundancy, a Collaborative Greedy Neighborhood (CGN) algorithm is introduced to optimize the distribution of viewpoints. This method preprocesses the bridge model using Voronoi diagrams and generates optimal viewpoints by fully considering constraints such as camera field of view and gimbal angles. Secondly, in solving the UAV path planning problem, factors including collision risks between the UAV and the bridge, path length, frequency of steep ascents, and sharp turns are comprehensively considered. An improved Particle Swarm Optimization (PSO) algorithm based on a Greedy Strategy (Greedy Strategy-Improved PSO, GS-IPSO) is proposed to achieve global optimization of the path. The simulation results show that compared with the viewpoint offset method, Q-learning algorithm and clustering algorithm, the proposed CGN algorithm has significant advantages in significantly reducing the number of required viewpoints. In addition, in the path planning task, the GS-IPSO algorithm can not only effectively shorten the path length, but also significantly reduce the number of sharp turns and rapid climbs and descents compared to the traditional PSO algorithm and enhanced discrete particle swarm optimization (Enhanced-DPSO) algorithm, thereby significantly improving the inspection efficiency and reliability. Finally, through experimental verification in a real-world environment, the practicality and effectiveness of the method proposed in this paper are further confirmed.}, archive = {J_EAAI}, author = {Xudong Li and Yang Chen and Zhihuan Chen and Zixin Huang}, doi = {10.1016/j.engappai.2025.111253}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111253}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Coverage path planning of bridge inspection with unmanned aerial vehicle}, volume = {156}, year = {2025}, }
• (2025). A fruit fly-inspired path planning algorithm for unmanned aerial vehicle in underground environments based on low-discrepancy sequences. EAAI, 156, 111250. (www)

  @article{ , abstract = {Sampling planning algorithms are crucial in high-dimensional path planning for unmanned aerial vehicle (UAV), particularly in underground environments where Global Positioning System (GPS) signals are absent. The Rapidly-exploring Random Tree (RRT) algorithm, however, faces challenges due to the use of pseudo-random sequences, resulting in issues such as under-sampling, over-sampling, and they also suffer from high computational costs and redundant paths. To address these limitations, the (Halton-based Clustering) HBC-RRT algorithm is proposed. This algorithm utilizes the Halton sequence, replacing the pseudo-random sequence, which fundamentally resolves the problems of under-sampling and over-sampling in the RRT method. Additionally, a novel sampler is introduced, inspired by the fruit fly guidance mechanism, which optimizes the dual-tree sampling process by selecting either the optimal sampling candidate strategy or virtual sub-goal points. The optimal sampling candidate strategy conducts global sampling in three-dimensional space, effectively reducing sampling time. The virtual sub-goal points accelerate the dual-tree connection process, enhancing the directional guidance. The fruit fly mechanism reconstructs the path, leading to a reduction in UAV path costs. The performance of the HBC-RRT algorithm is evaluated through simulation experiments, demonstrating superior results in comparison to four baseline algorithms, including reductions in path length, convergence time, and memory usage. The practical applicability of the algorithm is further validated through real-world experiments involving a UAV and radar in an underground environment.}, archive = {J_EAAI}, author = {Huageng Zhong and Yu Du and Dong Liu and Minghao Wang and Ming Cong and Xiaojing Tian}, doi = {10.1016/j.engappai.2025.111250}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111250}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A fruit fly-inspired path planning algorithm for unmanned aerial vehicle in underground environments based on low-discrepancy sequences}, volume = {156}, year = {2025}, }
• (2025). Automatic sleep staging based on single-channel ballistocardiogram signals and multiple scales temporal feature analysis. EAAI, 156, 111249. (www)

  @article{ , abstract = {Traditional sleep staging studies often rely on contact sensors for signal data acquisition, which may compromise the integrity of sleep data. Our research presents an automatic sleep staging method utilizing non-contact single-channel Ballistocardiogram (BCG) signals to address this limitation. This study proposes a multiple scale(multi-scale) time window feature extraction method based on heart rate variability (HRV) and respiratory rate variability (RRV) to establish a more precise correlation between BCG signals and sleep stages. Additionally, we present an advanced two-layer stacked ensemble model designed to enhance the accuracy and robustness of sleep staging. The innovative sleep staging model is subjected to a rigorous 5-fold cross-validation on 10 diverse recordings, encompassing 10,614 sleep segments. Experimental results indicate that the proposed feature extraction method constructs a Top-50 feature set with an average weight of 0.2082, representing a 195.8 % improvement compared to the 0.6158 of the traditional 30s feature set. Additionally, the proposed classification model achieves an accuracy of 89.15 %, outperforming traditional sleep staging models by 2 percentage points. It provides valuable references for research based on HRV, RRV, and other biological information features. This advancement enhances sleep monitoring, especially for home and mobile healthcare, offering a more user-friendly experience and practical medical tools. " The BCG signal sleep data source code is available at [ https://github.com/ZJSRU-ICLaboratory/BCG-Sleepstaging .]."}, archive = {J_EAAI}, author = {Ke Wang and Bingyang Zhu and Banteng Liu and Jingyao Liang and Tan Lv and Jianfeng Wu}, doi = {10.1016/j.engappai.2025.111249}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111249}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic sleep staging based on single-channel ballistocardiogram signals and multiple scales temporal feature analysis}, volume = {156}, year = {2025}, }
• (2025). A non-destructive automatic pavement damage detection scheme based on end-to-end neural networks with multi-level attention mechanism. EAAI, 156, 111246. (www)

  @article{ , abstract = {The accurate classification and statistics of road damage detection technology are crucial for road condition evaluation and maintenance decisions. However, the accuracy of complex road surface damage detection based on deep learning is still insufficient for real engineering, and even one of the damages may be repeatedly counted. This study develops a new non-destructive automatic road damage detection technology that includes detect road damage based on deep learning and redundant damage image de-duplication. This technology based on multi-level attention mechanism is designed from the perspectives of convolutional kernels and loss functions, improves the accuracy of real road surface damage detection. Compared to the original network, mAP@0.5 and F1 score increase by 5.1 % and 4 % for the public dataset RDD-2020, respectively. This technology achieves de-duplicate accuracy of 94.29 % in the duplicate road damage dataset (DRDD) by adding image processing algorithm, which will accelerate the engineering application of non-destructive automatic pavement damage detection.}, archive = {J_EAAI}, author = {Yipeng Liu and Chuan Wang and Yingchao Zhang and Xiteng Sun and Cong Du and Dongdong Xie and Yuan Tian}, doi = {10.1016/j.engappai.2025.111246}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111246}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A non-destructive automatic pavement damage detection scheme based on end-to-end neural networks with multi-level attention mechanism}, volume = {156}, year = {2025}, }
• (2025). Enhancing the sustainability and effectiveness of non-governmental organizations: An integrated neutrosophic decision-making model for determining the most effective corporate governance strategies. EAAI, 156, 111245. (www)

  @article{ , abstract = {This research explores the development of effective corporate governance strategies for non-governmental organizations (NGOs) by evaluating both critical decision criteria and innovative organizational models. Based on an extensive literature review, seven governance criteria were identified: credibility and non-profit orientation, regular information flow and transparency, level of institutionalization, frequency of leadership turnover, internal and external stakeholder relations, annual budget, and ideological or religious motivation. In addition, five innovative organizational models were selected as strategic alternatives: continuous learning organization, technology-oriented policy development, income diversification, pluralistic governance, and transformational leadership. The empirical data were gathered through interviews with 16 experts from academia, civil society, and public institutions in Türkiye. To address the complexity and uncertainty of decision-making processes, a hybrid type-2 neutrosophic number (T2NN) multi-criteria decision-making methodology is introduced. The T2NN-ranking comparison and T2NN-criteria importance through intercriteria correlation methods are used to determine the weight of criteria, while T2NN-stable preference ordering towards ideal solution is utilized to rank the alternative models. Results show that the level of institutionalization is the most significant criterion, and income diversification is the most effective organizational model for improving NGO sustainability. The study not only provides practical guidance for NGO leaders but also introduces a robust methodological approach applicable to governance challenges under uncertain conditions.}, archive = {J_EAAI}, author = {Emrah Ayhan and Hakan Gökhan Gündoğdu and Ahmet Aytekin and Ahmet Furkan Karakaya and Vladimir Simic and Dragan Pamucar}, doi = {10.1016/j.engappai.2025.111245}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111245}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing the sustainability and effectiveness of non-governmental organizations: An integrated neutrosophic decision-making model for determining the most effective corporate governance strategies}, volume = {156}, year = {2025}, }
• (2025). Data-driven analysis of the process, organization and properties of large-size complex thin-walled die-casting aluminium alloys. EAAI, 156, 111244. (www)

  @article{ , abstract = {High pressure die casting technology, as an advanced precision forming process, is widely used in the manufacture of different size parts. Whereas in the production of large-size parts, the spatial heterogeneity of mechanical properties caused by local thermal field inhomogeneity becomes a key technical bottleneck. This paper presents a data-driven framework that effectively leverages the thermal history throughout the casting process, enabling precise predictions of part mechanical properties at various spatial locations. A physical simulation model is used to simulate the casting and molding process of front nacelle, with sensors embedded to provide temperature data at different locations. Subsequently, the effects of process and location on solidification rate, microstructure, and mechanical properties are discussed. Relying solely on the solidification rate to predict mechanical properties is inadequate, necessitating the consideration of additional factors and material characteristics. Finally, the thermal history data of the whole casting process is provided to the one-dimensional convolutional neural network model, which is used to extract global features, not just local information, to predict the mechanical properties at unknown locations. To gain deeper insight into the intricate relationship between the variables, each convolutional layer is visualized to elucidate the correlation between thermal history and mechanical properties. This research is an extension of die-casting technology in the field of artificial intelligence and a new application in the field of automotive lightweighting.}, archive = {J_EAAI}, author = {Jian Yang and Bo Liu and Dongwei Shu and Qin Yang and Yunbo Zeng and Jun Huang}, doi = {10.1016/j.engappai.2025.111244}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111244}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven analysis of the process, organization and properties of large-size complex thin-walled die-casting aluminium alloys}, volume = {156}, year = {2025}, }
• (2025). Gated transformer-based proximal policy optimization for multiple marine autonomous surface ships collision avoidance decision-making strategy. EAAI, 156, 111242. (www)

  @article{ , abstract = {—To address the challenges posed by inaccurate state space prediction at sea and uncertainty in multi-ship departure from International Regulations for Preventing Collisions at Sea (COLREGs), this study proposes a Gated Transformer-Based Proximal Policy Optimization (GTr-PPO) algorithm for autonomous collision avoidance decision-making for Multiple Marine Autonomous Surface Ships (MASS). The novelty of this method resides the following aspects: (1) The separate input of continuous and discrete states, leveraging the influence of transformer characteristics on different state space information to predict the importance of current information and enhance prediction accuracy. (2) The novel gating mechanism effectively captures the output from the multi-head self-attention, thereby improving computational efficiency and learning stability. (3) A high-precision observation model is adopted, which incorporates considerations in COLREGs and emergency situations based on an improved reward function. The simulation demonstrates that, compared to the Transformer architecture, the proposed GTR-PPO enhances training success rate by 11.19 % and accelerates the convergence rate of the average reward by 35.29 %. It is important to highlight that the proposed method achieves a 95 % generalization success rate involving complex eight-ship encounter situations, which is significantly superior to other classical collision avoidance method. This achievement provides a feasible potential solution for the development of ship intelligent assistant autonomous navigation.}, archive = {J_EAAI}, author = {Zhewen Cui and Wei Guan and Xianku Zhang}, doi = {10.1016/j.engappai.2025.111242}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111242}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Gated transformer-based proximal policy optimization for multiple marine autonomous surface ships collision avoidance decision-making strategy}, volume = {156}, year = {2025}, }
• (2025). Representing-behavior-based online driving style recognition and its road verification. EAAI, 156, 111241. (www)

  @article{ , abstract = {—Driving style recognition (DSR) is a critical task for intelligent vehicles, providing a means to analyze drivers’ behavioral characteristics and understand their preferences. In this study, a representing-behavior-based DSR method is proposed. It is not only capable of distinguishing various driving styles but also adept at depicting their dynamic changes during a trip. Initially, three driving behavior datasets are extracted from natural driving data. Subsequently, a clustering metric called driving style distinction is introduced to assist the k-means algorithm in searching for driving style representing behaviors (DSRBs) within these datasets. Considering the impact of input features on clustering, a genetic-algorithm-based feature selection method is implemented to optimize the DSRB search. Upon completing this task, DSRB recognition models are built through supervised learning, enabling the calculation of a novel metric called the driving style index, which describes driving styles in real time. A road test demonstrates the feasibility of the proposed method, establishing it as an effective solution for DSR problems.}, archive = {J_EAAI}, author = {Bingxian Li and Yanfang Liu and Junwei Zhao and Xiangyang Xu and Peng Dong and Songbo Chen and Xiaojun Wu and Xuewu Liu and Hongzhong Qi}, doi = {10.1016/j.engappai.2025.111241}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111241}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Representing-behavior-based online driving style recognition and its road verification}, volume = {156}, year = {2025}, }
• (2025). Collaborative optimization of vessel scheduling and tugboat allocation in seaports using a multi-action deep reinforcement learning framework. EAAI, 156, 111239. (www)

  @article{ , abstract = {The allocation of tugboat groups is critical for ensuring the timely and safe entry and departure of vessels in ports, while vessel scheduling determines the demand and timing for tugboat group deployment. This study introduces the collaborative optimization problem of vessel scheduling and tugboat group allocation (VT-COP) in seaports. An end-to-end deep reinforcement learning framework, integrating a graph neural network, is proposed to autonomously learn and optimize policies for solving the VT-COP. In this framework, the reinforcement learning agent simultaneously controls multiple actions, specifically assigning each vessel operation to a suitable tugboat group from a set of available options at each timestep. The multi-action nature of this problem is modeled as a multiple Markov decision process (MMDP). To address this, we introduce a novel multi-pointer graph network (MPGN) architecture and a bi-Proximal Policy Optimization (bi-PPO) algorithm. The MPGN architecture consists of two encoder-decoder modules that independently define action policies for vessel operations and tugboat groups, predicting probability distributions for different vessel operations and tugboat groups, respectively. The VT-COP is modeled as a disjunctive graph and utilizes a graph neural network to dynamically embed local states during the scheduling process. Computational experiments show that our approach significantly outperforms heuristic algorithms in terms of solution quality, with learned policies exhibiting strong generalization across various instances. Furthermore, tests on real-world scenarios confirm that the collaborative optimization of vessel scheduling and tugboat group allocation meets the operational efficiency requirements of modern seaports. By enabling more accurate and dynamic scheduling decisions, this study could reduce delays, optimize resource utilization, and enhance the overall throughput of port activities, offering a significant advantage in modern, high-demand port environments.}, archive = {J_EAAI}, author = {Wenqiang Guo and Xinyu Zhang and Yuankui Li and Lihua Zhang}, doi = {10.1016/j.engappai.2025.111239}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111239}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Collaborative optimization of vessel scheduling and tugboat allocation in seaports using a multi-action deep reinforcement learning framework}, volume = {156}, year = {2025}, }
• (2025). Prescriptive tool wear control: Tool wear early prediction and cutting parameter regulation for discrete turning operations. EAAI, 156, 111238. (www)

  @article{ , abstract = {This study presents the first prescriptive tool wear control system specifically designed for turning operations, addressing the challenge of variability in discrete manufacturing processes. The proposed system integrates early tool wear prediction using spindle current and three-axis accelerometer signals as input parameters, enabling real-time regulation of cutting parameters to precisely control tool wear to a predefined set limit. Distinct from existing approaches that focus mainly on diagnostics and predictive analytics, this research introduces prescriptive analytics by combining Long Short-Term Memory (LSTM) predictive models with dynamic real-time parameter adjustments. The effectiveness of the proposed system was validated through three distinct case studies covering both straight and profile turning operations, demonstrating significant reductions in tool wear variability, with interquartile ranges (IQR) reduced by approximately 20 %–50 %. The novelty of this research lies in the first-ever integration of predictive analytics with prescriptive control to actively manage tool wear and reduce process variation, underscoring its practical value for industrial applications.}, archive = {J_EAAI}, author = {Chunhui Chung and Bing-Xian Cai and Benvolence Chinomona}, doi = {10.1016/j.engappai.2025.111238}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111238}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prescriptive tool wear control: Tool wear early prediction and cutting parameter regulation for discrete turning operations}, volume = {156}, year = {2025}, }
• (2025). Intelligent decision-making of shield tunneling machine using deep reinforcement learning in complex geological environment. EAAI, 156, 111237. (www)

  @article{ , abstract = {To reduce constructional hazards of manual operation, the implementation an intelligent decision-making of the shield machine will be an inevitable development trend in the future. Hence, this paper suggests an intelligent decision-making approach using the Twin Delayed Deep Deterministic Policy Gradient (TD3) agent, which is constructed under the Actor-Critic framework. A robust nonlinear correlation exists between sealed cabin pressure and tunneling parameters. Therefore, a deep reinforcement learning (DRL) training environment is established using the sealed cabin pressure mechanism model. TD3 agent is trained by exploring empirical data obtained from continuous interactions between TD3 agent and the environment. Critic networks calculate the evaluation value of the action. TD3 selects the minimum evaluation value to avoid overestimating the parameters. Actor network adopts delayed updating parameters to make control process more stable and accurate. Based on earth theory model, target pressure values under different earth conditions are calculated. An appropriate reward function is established based on the target pressure for current soil conditions. TD3 agent dynamically adjusts the propulsion speed and screw conveyor speed in real-time according to geological environment to achieve intelligent control of the sealed cabin earth pressure. The experimental results show that the decision-making error of this method is 0 Megapascal (MPa), and it better adapts to random soil variations and can complete nearly a hundred different soil transition tasks in complex environments. This study promotes technological advancements in tunnel engineering by introducing artificial intelligence (AI) technology into tunnel construction field, realizing deep integration of AI technology and civil engineering.}, archive = {J_EAAI}, author = {Wenshuai Zhang and Xuanyu Liu and Jiliang Zhang}, doi = {10.1016/j.engappai.2025.111237}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111237}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent decision-making of shield tunneling machine using deep reinforcement learning in complex geological environment}, volume = {156}, year = {2025}, }
• (2025). Ensemble machine learning models for estimating mechanical curves of concrete-timber-filled steel tubes. EAAI, 156, 111234. (www)

  @article{ , abstract = {The use of concrete-timber-filled steel tubes (CTFSTs) as composite structural elements in buildings is gaining attraction among researchers due to their positive structural behavior and high load-bearing capacity. The combination of steel, concrete and timber materials improves energy absorption and ductility making CTFSTs a promising choice for modern construction. However, finding the mechanical properties of CTFSTs is a challenge during the design process due to the complexity of predicting behavior. Reliable modeling of these interactions is essential for an optimal design, which requires extensive experimental data and advanced computational methodologies. Therefore, this study proposed ensemble machine learning (ML) models for estimating load-displacement and stress-strain curves as well as the maximum axial capacity and elastic stiffness of CTFSTs. The results confirm the reliability of ensemble ML models for predicting the elastic stiffness and the maximum axial capacity of CTFST specimens with error percentages of 0.57 and 0.72, respectively. In addition, proposed ensemble ML models were used to estimate axial load-displacement and stress-strain curves of CTFSTs having different shapes of timber, which their curve fitting ability were superior compared to other ML models (i.e., accuracy of 97.6 %). Having ensemble ML models validated by experimental tests, a graphical user interface (GUI) tool is prepared for the preliminary evaluation of CTFST specimens, which can ease the way for reducing the experimental costs.}, archive = {J_EAAI}, author = {F. Kazemi and N. Asgarkhani and T. Ghanbari-Ghazijahani and R. Jankowski}, doi = {10.1016/j.engappai.2025.111234}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111234}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ensemble machine learning models for estimating mechanical curves of concrete-timber-filled steel tubes}, volume = {156}, year = {2025}, }
• (2025). Graph regularized one-hot-constrained nonnegative matrix factorization for data representation. EAAI, 156, 111224. (www)

  @article{ , abstract = {Non-negative matrix factorization (NMF) effectively reduces the dimensionality of high-dimensional data. The low-dimensional representations obtained by NMF are widely used in computer vision, information retrieval, and pattern recognition. Semi-supervised NMF can obtain more discriminative low-dimensional representation by incorporating partial label information. In this paper, we propose a semi-supervised NMF method called Graph regularized One-hot Constrained NMF(GOCNMF), by setting the low-dimensional representations of the labeled data points as One-hot vectors in the decomposition. The setting guarantees the clustering assignments of labeled data points are consistent with their ground truth classes. In clustering tasks, the low-dimensional representations of the unlabeled data points are guided by the graph regularization to approximate the One-hot vectors in new space, so that the low-dimensional representations generated by GOCNMF can directly be used as a clustering assignment matrix. In addition, the setting can also be considered as an initialization strategy for the optimization algorithm of GOCNMF, which can accelerate the convergence speed of the algorithm and reduce its computational complexity. The clustering experiments on six real datasets show that our proposed GOCNMF outperforms the comparison methods overall, which validates the effectiveness of our method. The code for reproducing our results can be obtained at: https://github.com/ljisxz/GOCNMF .}, archive = {J_EAAI}, author = {Jie Li and Chaoqian Li}, doi = {10.1016/j.engappai.2025.111224}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111224}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph regularized one-hot-constrained nonnegative matrix factorization for data representation}, volume = {156}, year = {2025}, }
• (2025). Deep gate information bottleneck-based prediction model for complex disease-related micro-ribonucleic acids via heterogeneous biological networks. EAAI, 156, 111222. (www)

  @article{ , abstract = {Accurate prediction of potential association patterns is becoming a routine and essential method for detecting, analyzing, and controlling diseases. However, the complexity of biological networks poses significant challenges to computational methods between micro-ribonucleic acids and diseases. In this work, we propose a flexible multi-view gate information bottleneck-driven prediction model for complex disease-related micro-ribonucleic acid prediction. Our proposed model can reduce noise and redundant information from complex biological networks between different scale representations via the information bottleneck mechanism, and then enhances the robustness and generalization performance. Unlike other computational methods, we design the gate variational information bottleneck via a shrinking and an enlarging gate mechanism for multi-view different-order feature learning. The gate variational information bottleneck fuses the shared similarity and the view-specific embedding to obtain discriminative representation, and then eliminates redundant information and enhances task-relevant patterns. Next, the information bottleneck-based model is parameterized by a gate variational autoencoder and the reparameterization trick. Extensive experiments on different genomic datasets show the superior performance of our model compared to baselines, and the proposed model could effectively support the validation of complex disease-related micro-ribonucleic acids. It also shows that the model performance can be effectively improved by multi-view embedding learning and gate variational information bottleneck.}, archive = {J_EAAI}, author = {Yanbu Guo and Yiyang Xin and Jinde Cao and Yaoli Xu and Dongming Zhou}, doi = {10.1016/j.engappai.2025.111222}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111222}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep gate information bottleneck-based prediction model for complex disease-related micro-ribonucleic acids via heterogeneous biological networks}, volume = {156}, year = {2025}, }
• (2025). Differential recurrent autoencoder for early fault detection of nonstationary industrial processes. EAAI, 156, 111218. (www)

  @article{ , abstract = {Artificial Intelligence (AI) has emerged as a prominent technique for industrial fault detection, where deep learning demonstrates unique advantages in modeling nonlinear and dynamic process systems. Nevertheless, the presence of stochastic factors and permissible external disturbances frequently induce nonstationary characteristic during industrial process operation, which poses significant obstacles to early fault detection, particularly in capturing time-varying statistical patterns that precede critical failures. To bridge this gap, we propose a differential recurrent autoencoder (DRAE) for early fault detection of nonstationary industrial processes. DRAE based on differential operation, aiming to optimize the update mechanism of the forget gate to enhances its capacity to capture short-term nonstationary features. To facilitate the transition of time-variant nonstationary information in memory states, differential operations are embedded into the latent layer of Long Short-Term Memory (LSTM) neural network. The integration of the differential gradient layer with LSTM units directly models transient nonstationary dynamics while retaining original sequential dependencies, thereby overcoming the stationarity assumption inherent in conventional process monitoring approaches. Combined with an unsupervised encoder-decoder structure, this framework enables automatic feature extraction through fault-discriminative reconstruction to perform early fault detection. The proposed method is validated on a nonstationary numerical process, the Tennessee Eastman (TE) process, and a real-world chemical industrial process. The experimental results demonstrate significant improvement in fault detection rate and reduced detection delays compared to baseline methods, showcasing the potential of AI in enhancing process safety management within industrial systems.}, archive = {J_EAAI}, author = {Cheng Ji and Tingting Tao and Fangyuan Ma and Jingde Wang and Wei Sun}, doi = {10.1016/j.engappai.2025.111218}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111218}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Differential recurrent autoencoder for early fault detection of nonstationary industrial processes}, volume = {156}, year = {2025}, }
• (2025). Physics-informed neural ordinary differential equations and mass-spring system modeling framework for predicting the real-time soft-tissue deformations. EAAI, 156, 111217. (www)

  @article{ , abstract = {Background and objective Childbirth simulations have been investigated to prevent delivery complications. Traditional numerical methods incur large processing costs and unstable convergence issues. Our study aims to develop and evaluate a novel suitable predictive model of the soft tissue deformation based on physics-informed neural networks and neural ordinary differential equations coupled with the mass-spring system modeling. Methods Theoretical basis of the proposed approach was established using the physics-informed neural networks, neural ordinary differential equations and mass-spring system formulations. Physics-informed neural network combines fundamental physical principles with neural networks. Neural ordinary differential equation enables the use of ordinary differential equation solvers to train the neural networks. We evaluated our novel approach on different models (two-dimensional and three-dimensional mass-spring systems, cantilever beam and a uterus model). Results The best extrapolation was in the experiment of the cantilever beam with a root mean squared error of 1.057 × 10 −5 m and a relative error of 0.031 % (percentage) for the time from 2 to 5 s. The extrapolating aspect of soft tissue uterus deformation from 1 to 2 s returned a root mean squared error of 6.404 × 10 −3 m and a relative error about 6 % (percentage). The Pearson correlation coefficient for most cases was 0.99. A key strength of our models is their ability to quickly and accurately adapt to new initial conditions during extrapolation. Conclusions Our proposed outstanding modeling approach opens new avenues in the real-time soft tissue deformation during the childbirth simulation toward a next-generation decision support tool for childbirth training and complication simulation.}, archive = {J_EAAI}, author = {Hien-Duyen Nguyen-Le and Abbass Ballit and Tien-Tuan Dao}, doi = {10.1016/j.engappai.2025.111217}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111217}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed neural ordinary differential equations and mass-spring system modeling framework for predicting the real-time soft-tissue deformations}, volume = {156}, year = {2025}, }
• (2025). Linguistic pythagorean fuzzy Aczel–Alsina aggregation operators and their application in multi-attribute group decision-making: An approach for smart city development. EAAI, 156, 111215. (www)

  @article{ , abstract = {This paper aims to present a new multi-attribute group decision-making (MAGDM) approach for solving problems in an uncertain and complex environment. In any MAGDM problem, the key challenge is how to quantify and aggregate the objective uncertainty information contained in the data. To answer this, in this paper, we utilize the idea of linguistic Pythagorean fuzzy set (LP y FS) to describe the information in terms of linguistic membership degree (LMD) and linguistic non-membership degree (LNMD). We first present some new operations of LP y FS, including Aczel–Alsina (AA) sum, AA product, and AA scalar multiplication. At that point, we develop various LP y F aggregation operators which are based on these defined operations, including the LP y F Aczel–Alsina weighted averaging (LP y FAAWA) operator, the LP y F Aczel–Alsina ordered weighted averaging (LP y FAAOWA) operator, LP y F Aczel–Alsina weighted geometric (LP y FAAWG) operator and LP y F Aczel–Alsina ordered weighted geometric (LP y FAAOWG) operator. We configure these operators with different characteristics. It can be seen that the proposed operators have commutativity, monotonicity, boundary, and idempotency properties. Then, we design a new technique dependent on these operators to solve MAGDM problems. The applicability of the established approach is demonstrated through a case study related to smart city development. The result demonstrates the new technique's feasibility and applicability. In the end, sensitivity analysis and a systematic comparison between the prevailing and suggested techniques are given.}, archive = {J_EAAI}, author = {Shahid Hussain Gurmani and Zunaira Rasool and Rifaqat Ali and Shehr Bano and Huayou Chen and Yuming Feng and Rana Muhammad Zulqarnain}, doi = {10.1016/j.engappai.2025.111215}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111215}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Linguistic pythagorean fuzzy Aczel–Alsina aggregation operators and their application in multi-attribute group decision-making: An approach for smart city development}, volume = {156}, year = {2025}, }
• (2025). Elements discriminative non-negative matrix factorization for data clustering. EAAI, 156, 111210. (www)

  @article{ , abstract = {Semi-supervised non-negative matrix factorization (NMF) is widely used in data clustering because it can generate more discriminative representations for high-dimensional data by leveraging partial label information. To advance research in this field, we propose a novel method, Element Discriminative NMF (EDNMF), which incorporates discrimination constraints based on the element ratio and element difference of the new representations of labeled data points. EDNMF is implemented in two variants, each with an iterative algorithm for solving the optimization problem. We further analyze the computational complexity and convergence of these algorithms. A key advantage of EDNMF is that its learned representations can serve directly as a clustering assignment matrix, thereby simplifying the clustering process. Extensive experiments on eight real-world datasets demonstrate that EDNMF consistently outperforms baseline methods, confirming its effectiveness in improving clustering performance. The code is available at https://github.com/ljisxz/EDNMF .}, archive = {J_EAAI}, author = {Jie Li and Xuzhu Shen and Chaoqian Li and Yaotang Li}, doi = {10.1016/j.engappai.2025.111210}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111210}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Elements discriminative non-negative matrix factorization for data clustering}, volume = {156}, year = {2025}, }
• (2025). Adaptive lightweight temporal convolutional network with context-aware downsampling strategy for traffic flow prediction. EAAI, 156, 111208. (www)

  @article{ , abstract = {Accurate traffic flow prediction is crucial for intelligent transportation systems, which remains challenging due to intricate spatiotemporal features. Despite the promising performance of recent advances in spatiotemporal models, expensive computational costs make them difficult to be used with limited hardware resources. Moreover, existing models tend to cope with multiple temporal patterns of traffic flow in a coarse-grained manner, making it difficult to deeply extract temporal features. In this study, a novel deep learning model is proposed to address the aforementioned issues and achieve accurate traffic flow prediction. First, a new context-aware downsampling strategy is proposed for reducing the computational cost of the model, which provides contextual information for the traffic flow at each time step and then performs downsampling to reduce the sequence length, thus making the model more lightweight. Second, a new adaptive lightweight temporal convolutional module is proposed to extract temporal features deeply, which can adaptively update model parameters in a lightweight and fine-grained manner to deal with multiple temporal patterns of traffic flow. Third, the proposed model employs spatiotemporal embedding to efficiently learn the underlying spatiotemporal patterns of traffic flow. Extensive experiments on multiple real-world datasets validate the effectiveness and robustness of the proposed model.}, archive = {J_EAAI}, author = {Shuai Zhang and Xiang Yin and Wenyu Zhang and Jiyuan Xu and Xin Jing}, doi = {10.1016/j.engappai.2025.111208}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111208}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive lightweight temporal convolutional network with context-aware downsampling strategy for traffic flow prediction}, volume = {156}, year = {2025}, }
• (2025). A comprehensive review on computer vision analysis of aerial data. EAAI, 156, 111206. (www)

  @article{ , abstract = {With the emergence of new technologies in the field of airborne platforms and imaging sensors, aerial data analysis is becoming very popular, capitalizing on its advantages over land data. This paper presents a comprehensive review of the computer vision tasks within the domain of aerial data analysis. While addressing fundamental aspects such as object detection and tracking, the primary focus is on pivotal tasks like change detection, object segmentation, and scene-level analysis. The paper provides the comparison of various hyper parameters employed across diverse architectures and tasks. A substantial section is dedicated to an in-depth discussion on libraries, their categorization, and their relevance to different domain expertise. The paper encompasses aerial datasets, the architectural nuances adopted, and the evaluation metrics associated with all the tasks in aerial data analysis. Applications of computer vision tasks in aerial data across different domains are explored, with case studies providing further insights. The paper thoroughly examines the challenges inherent in aerial data analysis, offering practical solutions. Additionally, unresolved issues of significance are identified, paving the way for future research directions in the field of aerial data analysis.}, archive = {J_EAAI}, author = {Vivek Tetarwal and Manpreet Kaur and Sandeep Kumar}, doi = {10.1016/j.engappai.2025.111206}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111206}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A comprehensive review on computer vision analysis of aerial data}, volume = {156}, year = {2025}, }
• (2025). Adaptive extreme learning framework for nitrogen oxide conversion prediction: A multi-scale approach. EAAI, 156, 111204. (www)

  @article{ , abstract = {Efficient selective catalytic reduction (SCR) systems for nitrogen oxide (NOx) emission control face significant challenges in catalyst design optimization due to data scarcity and complex parameter relationships. We present an Adaptive Extreme Learning Machine (AELM) framework designed to address these engineering challenges. The framework innovatively integrates automated text mining to significantly expand the dataset from literature, utilizes adaptive mechanisms (e.g., feature fusion, temperature bias) within a multi-output regression structure to model complex parameter interactions and temperature dependencies, enabling accurate catalyst performance prediction even under data-scarce conditions. This artificial intelligence approach demonstrates exceptional engineering applicability, achieving up to a 93.3 % improvement in prediction accuracy (measured by Fréchet distance for curve similarity on challenging cases) while reducing computational overhead by approximately 87 % compared to conventional machine learning models like Artificial Neural Networks. Through comprehensive validation across multiple zeolite catalysts, our framework successfully captures the complex relationships between key structural parameters (Copper content: 2–6 wt percent, Silicon/Aluminum ratio: 5–12) and conversion efficiency, even for diverse zeolite catalysts and challenging test conditions. This development provides engineers with a powerful tool for rapid catalyst screening and optimization in real-world emission control applications, offering practical solutions for sustainable environmental protection technologies.}, archive = {J_EAAI}, author = {Chengzheng Tong and Yanhua Wang and Liehao Wei and Hu Wang and Qingling Liu and Caixia Liu and Mingfa Yao}, doi = {10.1016/j.engappai.2025.111204}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111204}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive extreme learning framework for nitrogen oxide conversion prediction: A multi-scale approach}, volume = {156}, year = {2025}, }
• (2025). Multisource-domain regression transfer learning framework for predicting student academic performance considering balanced similarity. EAAI, 156, 111202. (www)

  @article{ , abstract = {The increasing integration of information technology and artificial intelligence has extensively implemented computer-aided intelligent education systems in higher education. A critical task within these systems is student performance prediction, which forecasts future academic outcomes by analyzing data such as historical grades, learning behaviors, and classroom participation. This enables early intervention and personalized teaching based on scientific evidence. However, most existing methods rely on traditional machine learning techniques, which can hardly address issues such as domain distribution discrepancies and data imbalance effectively. To overcome these challenges, we propose a multisource-domain transfer learning regression framework that integrates domain selection, hybrid feature extraction, and dynamic joint distribution adaptation techniques. Specifically, the framework first selects appropriate source domains on the basis of preset thresholds via cross-validation. Thereafter, a hybrid feature extractor is used to derive (i) common features from the target and selected source domains and (ii) domain-specific features from the target domain. Finally, a dynamic adaptive factor is introduced to balance differences between the marginal and conditional distributions. Experimental results indicate that the proposed framework significantly reduces the root mean square error with an average prediction improvement of 21.05 %, compared with baseline methods and other advanced approaches.}, archive = {J_EAAI}, author = {Li Wang and Lucong Zhang and Haotian Wu and Teng Zhang and Ke Qiu and Tianyu Chen and Hongwu Qin}, doi = {10.1016/j.engappai.2025.111202}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111202}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multisource-domain regression transfer learning framework for predicting student academic performance considering balanced similarity}, volume = {156}, year = {2025}, }
• (2025). Advanced multi-model prediction of aircraft engine remaining useful life with random sampling-based class balancing and voting-based features selection. EAAI, 156, 111201. (www)

  @article{ , abstract = {In Industry 4.0, predictive maintenance for critical systems like aircraft engines relies on accurate Remaining Useful Life (RUL) estimation to prevent unexpected failures and optimize maintenance schedules. However, existing models face several limitations that hinder their effectiveness in real-world applications. Common challenges include data imbalance, which can lead to biased predictions; suboptimal feature selection, which may overlook important predictive variables; and limitations in model accuracy, particularly when handling complex, high-dimensional datasets. These issues often reduce the reliability and generalizability of RUL predictions across varied operational contexts. To address these challenges, this study proposes a novel data-driven approach that integrates random-sampling class balancing, a voting-based feature selection method, and advanced machine learning techniques, including ensemble and deep learning models. Applied to the Commercial Modular Aero-Propulsion System Simulation dataset, our model demonstrates significant improvements in RUL prediction, achieving Root Mean Square Error values of 11.10% for FD001, 12.09% for FD002, 11.80% for FD003, and 12.88% for FD004. These results highlight the model’s robustness and its potential to enhance predictive maintenance in aeronautical engineering.}, archive = {J_EAAI}, author = {Ibrahima Barry and Meriem Hafsi}, doi = {10.1016/j.engappai.2025.111201}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111201}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advanced multi-model prediction of aircraft engine remaining useful life with random sampling-based class balancing and voting-based features selection}, volume = {156}, year = {2025}, }
• (2025). Optimizing aircraft engine longevity: A comparative framework for dynamically adaptive predictive maintenance using autoencoders, LSTMs, and gaussian processes. EAAI, 156, 111199. (www)

  @article{ , abstract = {The aviation industry's reliance on fixed-interval maintenance schedules for aircraft engines poses significant inefficiencies, balancing precariously between excessive resource expenditure and catastrophic safety risks. Traditional frameworks, such as the 50-h inspection cycle, fail to account for real-time engine degradation, often leading to unnecessary downtime or undetected failures. Recent machine learning (ML) advances offer transformative solutions through predictive maintenance (PdM), which dynamically adapts schedules based on engine health data. This study evaluates three ML methodologies: Autoencoders (AEs), Long Short-Term Memory networks (LSTMs), and Gaussian Process Regression (GPR) to construct degradation indicators, predict Remaining Useful Life (RUL), and optimize maintenance timing. This work addresses a critical gap in adaptive maintenance strategies by comparing their efficacy in anomaly detection, temporal modeling, and uncertainty quantification. It offers a data-driven pathway to enhance safety, reduce costs, and extend engine operational life.}, archive = {J_EAAI}, author = {Slawomir Szrama}, doi = {10.1016/j.engappai.2025.111199}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111199}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing aircraft engine longevity: A comparative framework for dynamically adaptive predictive maintenance using autoencoders, LSTMs, and gaussian processes}, volume = {156}, year = {2025}, }
• (2025). Heterogeneous ensemble short-term integrated load forecasting considering differentiated cycle-trend properties. EAAI, 156, 111198. (www)

  @article{ , abstract = {Aiming at the Integrated Energy System (IES), there are key problems such as the potential change pattern of seasonal differentiation of multiple loads, the lack of mining complex coupling characteristics, the limited generalization ability of forecasting models and the weak interpretability. A heterogeneous integrated short-term multiple load forecasting method considering differentiated cycle-trend characteristics is proposed. Firstly, Multiple Seasonal Trend decomposition using Loess (MSTL) is used to decompose the multivariate loads into multi-timescale cycle and trend series, and to construct multi-timescale cycle-trend feature sets, which breaks through the limitations of the traditional single-cycle decomposition in characterizing inter-seasonal coupling characteristics. Secondly, the evolutionary multi-objective optimization mechanism based on the dual indexes of diversity and accuracy is introduced into the Stacking integration framework for the first time, and the seasonal adaptive optimal model combination selection of electric/cooling/heating loads is realized. Finally, a two-layer interpretable analysis paradigm based on Shapley Additive Explanations (SHAP) values is established to improve the interpretability of the proposed multivariate load forecasting method by performing input feature attribution analysis from both global and individual dimensions. The experimental analysis is carried out on the example of Tempe Campus IES of Arizona State University in the U.S.A. The results show that the average percentage errors of electricity, cooling, and heating forecasts of the proposed method reach 1.783 %, 1.012 %, and 2.225 %, respectively, demonstrating superior forecasting accuracy compared to existing methods.}, archive = {J_EAAI}, author = {Nantian Huang and Yaoyao Wang and Bingling Li and Guowei Cai and Liang Zhang}, doi = {10.1016/j.engappai.2025.111198}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111198}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Heterogeneous ensemble short-term integrated load forecasting considering differentiated cycle-trend properties}, volume = {156}, year = {2025}, }
• (2025). A problem-specific knowledge-based multi-objective algorithm for sustainable scheduling of distributed heterogeneous welding permutation flow shop. EAAI, 156, 111197. (www)

  @article{ , abstract = {The distributed heterogeneous welding permutation flow shop scheduling problem (DHWPFSP) involves the diversity, heterogeneity, and coordination of welding processes, making it more complex than traditional distributed flow shop scheduling problems. Additionally, welding is a high-energy-consuming process, and reducing energy consumption to achieve sustainable green manufacturing has always been a primary goal in the industry. Currently, research on this problem is minimal, and existing studies often overlook the impact of transportation on production efficiency and energy consumption. This paper proposes an efficient and energy-saving scheduling model, the goal is to minimize the maximum completion time and total energy consumption. By combining the features of the multi-objective evolutionary algorithm based on decomposition (MOEA/D) algorithm and the non-dominated sorting genetic algorithm-II (NSGA-II) algorithm, we designed the MONS-II algorithm (MOea/d and NSga II). Initial solutions are generated using the distributed neighborhood exchange heuristic (DNEH) method and a random generation strategy, with improvements made to the encoding method. The partially matched crossover (PMX) and precedence operation crossover (POX) strategies are applied, along with local neighborhood search, and enhancements to external archive management and adaptive adjustment strategies. Experimental results demonstrate that the MONS-II algorithm performs excellently in terms of both total time and energy consumption, providing more uniform and reasonable solutions. Using a crane manufacturing enterprise as an example, the effectiveness of the model and algorithm in the distributed welding shop scheduling problem is verified, providing theoretical support for the sustainable development of enterprises.}, archive = {J_EAAI}, author = {Jianguo Duan and Zixuan Liu and Mengting Wang and Yulin Du and Mengpei Yang}, doi = {10.1016/j.engappai.2025.111197}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111197}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A problem-specific knowledge-based multi-objective algorithm for sustainable scheduling of distributed heterogeneous welding permutation flow shop}, volume = {156}, year = {2025}, }
• (2025). Navigating sentiment analysis through fusion, learning, utterance, and attention methods: An extensive four-fold perspective survey. EAAI, 156, 111195. (www)

  @article{ , abstract = {The growing need to automate user sentiment evaluation for goods and services, especially given the rising popularity of video content for online opinion expression, highlights the importance of sentiment analysis in artificial intelligence. This survey introduces a novel taxonomy and analyzes sentiment analysis approaches from four perspectives: Fusion-based, Learning-based, Utterance-based, and Attention-based methods. The study is organized around fundamental elements, significance, and applications of sentiment analysis. The proposed literature provides details about fusion techniques, which may be categorized based on levels, modalities, and procedures, including rule-based and classification-based approaches. This study seeks to provide a concise overview of several learning approaches, accompanied by a brief discussion of a benchmark dataset specific to each model. This work offers a concise analysis of utterance acquisition, focusing on the classification of sentiments according to language specificity and bilingual or multilingual proficiency, as well as the factors of linguistic basis, granularity, and multimodality. Additionally, a benchmark set derived from publicly accessible datasets is presented, accompanied by comprehensive insights. Critical analysis of sentiment analysis limitations provides valuable insights into challenges and potential solutions. The survey concludes by highlighting prospects in the dynamic landscape of sentiment analysis.}, archive = {J_EAAI}, author = {Bhavana Verma and Priyanka Meel and Dinesh Kumar Vishwakarma}, doi = {10.1016/j.engappai.2025.111195}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111195}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Navigating sentiment analysis through fusion, learning, utterance, and attention methods: An extensive four-fold perspective survey}, volume = {156}, year = {2025}, }
• (2025). Towards improved human arm movement analysis: Advanced feature engineering and model optimization. EAAI, 156, 111194. (www)

  @article{ , abstract = {Human arm movement (HAM) analysis plays a crucial role in rehabilitation, sports science, and human-computer interaction (HCI). This has led to extensive research, introducing various systems and algorithms to capture and analyze human movement data. This study presents a novel integrated pipeline for analyzing human arm movement time series data collected from an Inertial Measurement Unit (IMU) embedded in an Apple Watch. The pipeline combines time and frequency domain feature extraction, dimensionality reduction using Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE), and embedded feature selection methods including Least Absolute Shrinkage and Selection Operator (Lasso), Random Forest, and Elastic Net, to optimize classification accuracy while reducing computational complexity. Targeting key movements such as elbow extension/flexion, shoulder extension/flexion, and wrist protonation/supination, data were captured using accelerometers and gyroscopes. The initial dataset, consisting of 720 features from both time-domain and frequency-domain analyses, was efficiently reduced to 46 features, mitigating the curse of dimensionality while preserving essential information. As a result, the proposed method achieved a classification accuracy of over 96 %, with precision, recall, and F1-score of 0.95.}, archive = {J_EAAI}, author = {Yassine Benachour and Moez Rehman and Farid Flitti}, doi = {10.1016/j.engappai.2025.111194}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111194}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards improved human arm movement analysis: Advanced feature engineering and model optimization}, volume = {156}, year = {2025}, }
• (2025). Critical success factors for implementing artificial intelligence in construction projects: A systematic review and social network analysis. EAAI, 156, 111192. (www)

  @article{ , abstract = {Artificial intelligence (AI) is increasingly deployed to automate routine tasks, generate accurate insights from big data, and build predictive models to inform better decision-making in construction projects. However, AI deployment in construction projects constitutes a sociotechnical process, such that adopting solely a technical approach becomes inadequate. This study investigated the critical success factors for implementing AI in construction projects. It combined a systematic literature review, meta-analysis, and social network analysis to evaluate the scientific evidence on the critical success factors, and quantitatively reveal the underrepresented factors. The meta-analysis identified 38 critical success factors, ranked according to normalized scores and degree centralities. The study derived four dimensions of the critical success factors, including organizational, technological, stakeholder, and data success factors. The social network analysis quantitatively revealed the strengths and existing gaps in the reviewed studies and provide insights into factors that need further investigation.}, archive = {J_EAAI}, author = {Ibrahim Yahaya Wuni}, doi = {10.1016/j.engappai.2025.111192}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111192}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Critical success factors for implementing artificial intelligence in construction projects: A systematic review and social network analysis}, volume = {156}, year = {2025}, }
• (2025). Two-stage vision system: Application of multi-perspective object detection network and character recognition network in industrial product classification. EAAI, 156, 111190. (www)

  @article{ , abstract = {In industrial automation, precise identification of incoming goods using vision systems significantly enhances production efficiency and reduces shipment errors. To address issues of size discrepancies in target detection and the need for further category code identification, a two-stage visual classification system has been introduced. The first stage presents a novel You Only Look Once-DualSight Fusion Network (YOLO-DSF) model that adopts multi-perspective backbones: one enhances fine-detail extraction via a lightweight space-to-depth downsampling module (LSPDown), and the other captures global features through a lightweight GhostNet structure. A newly developed FocusFusion module (FFM) merges the outputs from these perspectives, reinforcing small-target detection and maintaining the detection capabilities for multi-scale objects. In the detection head, a low-level feature layer is introduced for improved performance on tiny objects, and a lightweight design is applied to control computational overhead, thereby address staircase convergence issues while preserving efficiency. Compared to the baseline model, the YOLO-DSF achieves a Mean Average Precision at Intersection over Union of 0.50 (mAP@0.5) of 98.6 %, an improvement of 4.4 %. Moreover, tests on the Northeastern University Surface Defect (NEU-DET) public dataset have also demonstrated the superiority of this model. The second stage involves a U-shaped denoising network (UDNet) that acts as a preprocessor in character recognition, effectively reducing background noise and boosting character visibility, resulting in an accuracy improvement of 1.5 %. Field tests have demonstrated that the system achieves an object detection accuracy of 96.3 % and a goods classification accuracy of 98.9 %, thereby verifying its practicality and value in industrial applications.}, archive = {J_EAAI}, author = {Shengjie Jin and Zhen Cao and Chaogang Yu}, doi = {10.1016/j.engappai.2025.111190}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111190}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Two-stage vision system: Application of multi-perspective object detection network and character recognition network in industrial product classification}, volume = {156}, year = {2025}, }
• (2025). Few-shot tire tracks recognition based on metric learning. EAAI, 156, 111188. (www)

  @article{ , abstract = {Tire tracks are crucial evidence for investigating and identifying traffic accident scenes. Intelligent recognition of tire track images enables rapid and precise identification of suspicious vehicles, assisting investigators in determining accident causes and liabilities. This enhances both the efficiency and accuracy of case resolution. However, tire track images are complex and diverse, significantly influenced by environmental factors such as lighting and road conditions. Moreover, the limited availability of tire track image samples poses a challenge for training deep learning models in practical applications. To address these challenges, particularly the limited generalization ability of few-shot metric learning models on unseen categories and their poor recognition rates in open-world environments, we propose a novel framework called TireNet for few-shot tire track recognition. First, we adopt a residual network integrated with coordinate attention as the backbone for feature extraction. Secondly, the context-aware features of the support set and the query set are extracted separately through attention-based bi-directional long short-term memory model. Subsequently, the cosine similarity between the features of the support set and the query set is calculated to determine the category of the query image. Finally, to address the class imbalance issue in the tire track image dataset, we utilize an improved Focal Loss for gradient updates. This enables the model to focus more on difficult samples during training, thereby enhancing the training efficiency, particularly in scenarios involving few-shot learning and open-world environments. To validate the proposed method, we constructed a tire track image dataset of 1700 samples, covering various environmental conditions, providing a rich resource for tire track recognition research. Experimental results demonstrate that the proposed method significantly outperforms few-shot image classification and classic machine learning methods in recognition accuracy, achieving optimal performance. Within a 95 % confidence interval, the average classification accuracy reaches 78.32 % for 5-way 1-shot and 93.48 % for 5-way 5-shot, respectively. Ablation experiments have confirmed the effectiveness of each module, indicating that the proposed model can be effectively used for few-shot tire track image recognition. The code and dataset are available at: https://github.com/hejinrong/tire_recognition_CAMN.git .}, archive = {J_EAAI}, author = {Longlong Zhai and Jinrong He and Zhaokui Li and Yingzhou Bi}, doi = {10.1016/j.engappai.2025.111188}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111188}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Few-shot tire tracks recognition based on metric learning}, volume = {156}, year = {2025}, }
• (2025). A transfer learning approach to image-based intelligent detection of street lighting lamp types and wattages. EAAI, 156, 111186. (www)

  @article{ , abstract = {Energy distribution companies in Brazil face the critical challenge of maintaining an accurate database for public lighting networks to ensure precise billing of city councils. Changes in this network often go unreported, underscoring the need for vigilant monitoring and regular updates to minimize commercial losses. This paper introduces new machine learning techniques to enhance the database update process, offering a cost-effective alternative to human-intensive methods for lamp classification. In a departure from previous approaches which utilized radiometric and image data, the classification task exclusively utilizes image data, employing traditional and deep learning classifiers well-established in the literature. The implementation of nested cross-validation ensures unbiased and reliable experimental results. This work applies three different transfer learning methods — feature extraction, partial fine-tuning, and full fine-tuning — to a machine learning application in the field of public lighting engineering, achieving superior results compared to existing methodologies, and establishing a new benchmark result for the problem. Notably, the best results were achieved by incorporating features extracted from deep learning architectures. The reported results demonstrate state-of-the-art performance in public lighting lamp classification, the best approach yielded a mean accuracy of 0.89, superior to the 0.86 reported previously by literature, showcasing the efficacy of machine learning approaches in optimizing database management processes for energy distribution companies.}, archive = {J_EAAI}, author = {Vitor Berger Bonella and Flávio Miguel Varejão}, doi = {10.1016/j.engappai.2025.111186}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111186}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A transfer learning approach to image-based intelligent detection of street lighting lamp types and wattages}, volume = {156}, year = {2025}, }
• (2025). Occluded skeleton-based multi-stream model using part-aware Spatial–Temporal graph convolutional network for human activity recognition. EAAI, 156, 111183. (www)

  @article{ , abstract = {Human activity recognition using skeleton data has engrossed significant research attention in pattern recognition due to its broad applications. However, occlusion remains a major challenge in activity recognition. In this paper, we propose a multi-stream part-aware occluded skeleton-based graph convolutional network designed to improve predictions in the presence of occlusions. The model consists of three key modules: the Input Inhibition Module for Skeleton Sequences, which handles incomplete or occluded skeleton data; the Part-Aware Spatial–Temporal Graph Convolutional Network, which captures spatial–temporal dependencies among human body key joints and the Predicted Score Inhibition, which refines the output by mitigating the effects of noisy data. By integrating these components, the model enhances robustness in occluded scenarios. The experiments demonstrate that the proposed method outperforms state-of-the-art models on several benchmark datasets, achieving a 6% improvement in recognition accuracy compared to previous approaches. Additionally, we extracted multi-modal features to construct more discriminative features, such as key-joint coordinates, relative coordinates, and temporal differences.}, archive = {J_EAAI}, author = {Roshni Singh and Abhilasha Sharma}, doi = {10.1016/j.engappai.2025.111183}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111183}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Occluded skeleton-based multi-stream model using part-aware Spatial–Temporal graph convolutional network for human activity recognition}, volume = {156}, year = {2025}, }
• (2025). Electroencephalography-driven three-dimensional object decoding with multi-view perception diffusion. EAAI, 156, 111180. (www)

  @article{ , abstract = {With the rapid development of artificial intelligence and neuroscience, existing research has shown that two-dimensional (2D) visual stimuli observed by subjects can be reconstructed from Electroencephalography (EEG) signals. However, few studies have attempted to decode three-dimensional (3D) visual objects from EEG signals evoked by 2D visual stimuli. Neuroscience research has confirmed that the brain can perceive 3D forms and cues from 2D visual stimuli, and this perceptual activity is reflected in EEG signals. To address this gap, this paper explores high-fidelity 3D object decoding from EEG signals by leveraging a multi-view perception diffusion model combined with Neural Radiance Field (NeRF) representations. The proposed EEG-to-3D method employs a two-stage learning process. The first stage captures latent EEG codes that imply 3D perceptual representations, by presenting a multi-task optimization strategy that combines EEG signal reconstruction with semantic classification tasks. In the second stage, a multi-view perception diffusion model, conditioned on latent EEG codes, is fine-tuned to constrain and optimize the parameters of the NeRF model, thereby generating multi-views of 3D objects with both semantic and viewpoint consistency. Experimental results demonstrate the effectiveness and superiority of the proposed method for 3D object decoding from EEG signals. The source codes are publicly available at: https://github.com/xiangxinhello/EEG_to_3D .}, archive = {J_EAAI}, author = {Xin Xiang and Wenhui Zhou and Guojun Dai}, doi = {10.1016/j.engappai.2025.111180}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111180}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Electroencephalography-driven three-dimensional object decoding with multi-view perception diffusion}, volume = {156}, year = {2025}, }
• (2025). Multi-regularized tensor-based framework for identifying hard landings. EAAI, 156, 111178. (www)

  @article{ , abstract = {Hard landings are a significant concern in civil aviation, often resulting in aircraft structural damage, financial losses, and compromised passenger safety. Automating the detection of such incidents faces challenges due to the complexities of Quick Access Recorder (QAR) data, which exhibit multi-channel interdependencies and temporal dynamics. Furthermore, environmental factors tied to flight-variant data, such as the geographic attributes of landing airports, can influence the occurrence of hard landings, yet these factors are often neglected in existing methodologies. This omission limits the practical utility of current approaches for enhancing safety in civil aviation. To address these challenges, we propose a multi-regularized tensor-based framework that models QAR data as a high-order tensor and applies tensor decomposition to extract latent patterns that characterize hard landing scenarios. The model incorporates tailored regularization terms to address both temporal correlations and inter-channel couplings across aircraft systems. To enable efficient computation, we develop a customized Block Coordinate Descent (BCD) algorithm, designed for efficient processing with high-dimensional factor matrices. The effectiveness of the proposed framework is validated using real-world civil aviation data from China, demonstrating superior performance in identifying hard landings.}, archive = {J_EAAI}, author = {Chenyang Chang and Yu An and Xi Zhang}, doi = {10.1016/j.engappai.2025.111178}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111178}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-regularized tensor-based framework for identifying hard landings}, volume = {156}, year = {2025}, }
• (2025). Application of an electronic tongue combined with meta-learning for rice origin detection. EAAI, 156, 111174. (www)

  @article{ , abstract = {Rice origin identification can offer brand protection for high-quality rice. The traditional deep learning methods for identifying the origins have the disadvantages, such as relying on a large number of labeled samples, insufficient generalization performance, and difficulty in learning and adapting to unknown tasks. A meta-learning strategy is introduced to solve these problems, this paper proposes a novel method for traceability detection of different rice origins based on an electronic tongue (ET) combined with an improved prototypical net work (IPN) model. Specifically, the taste fingerprint information of various rice samples is collected separately by using a homemade low-cost ET system. Then, a prototypical network equipped with an encoder structure composed of a convolutional neural network and a convolutional block attention module is proposed as a method of effectively processing the ET signals. A novel adaptive distance metric function and a prototype correction method are further introduced into the prototypical network, providing excellent classification and generalization performance. The proposed IPN model is trained with a semi-supervised strategy under few sample conditions, and its effectiveness and practicality are demonstrated by multiple extensive experiments. Compared with classical deep learning models, the IPN not only achieves better classification accuracy (99.2 %) but also requires fewer training samples and incurs a lower computational cost. This research found that the ET combined with meta-learning method could be a rapid, inexpensive and straightforward detection method for identifying the origins of rice, which is a novel application of artificial intelligence method and has great potential for rapid and on-site application.}, archive = {J_EAAI}, author = {Litong Chen and Zihan Wang and Zhiqiang Wang and Xuerui Bai and Jingyu Ma and Chuanzheng Liu and Tao Sun and Yubin Lan}, doi = {10.1016/j.engappai.2025.111174}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111174}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Application of an electronic tongue combined with meta-learning for rice origin detection}, volume = {156}, year = {2025}, }
• (2025). Utilizing convolutional neural networks for calculating full-field stress components and directions in photoelasticity. EAAI, 156, 111173. (www)

  @article{ , abstract = {Photoelasticity is a crucial experimental technique extensively used in various engineering and scientific domains. The integration of convolutional neural networks can substantially enhance the efficiency and capability of photoelasticity in resolving full-field stress. Nevertheless, existing neural network–based methods in photoelasticity are limited to computing only the difference between the principal stresses rather than determining the individual principal stress components (i.e., the absolute values of the first and second principal stresses) and the principal stress direction (isoclinic parameter). Directly solving for the principal stress components is of greater significance in many practical problems, and the principal stress direction is essential for evaluating material failure and optimizing designs. In this paper, a convolutional neural network is proposed for the first time that can directly and simultaneously determine the full-field first principal stress, second principal stress, and the principal stress direction. A dataset generation method was developed to train this network, producing a novel high-quality dataset containing 41,000 raw samples, without data augmentation. The proposed network exhibits high accuracy and strong generalization across synthetic and experimental validation sets. On the synthesized dataset, the structural similarity exceeds 0.98, and the mean squared error is below 0.45, with similarly satisfactory results on the experimental validation sets. This network establishes a direct mapping between photoelastic images and full-field stress components and directions, thereby enhancing the efficiency and potential applications of photoelasticity. The proposed dataset generation method may also offer valuable insights for advancing deep learning in photoelasticity.}, archive = {J_EAAI}, author = {Huajian Zhang and Shuhai Jia and Bo Wen and Xing Zhou and Zihan Lin and Longning Wang and Mengyu Han}, doi = {10.1016/j.engappai.2025.111173}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111173}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Utilizing convolutional neural networks for calculating full-field stress components and directions in photoelasticity}, volume = {156}, year = {2025}, }
• (2025). Deep learning-accelerated multiple design generation for sound-absorbing metaporous materials. EAAI, 156, 111172. (www)

  @article{ , abstract = {The design of sound-absorbing metaporous materials presents significant challenges due to the design complexity and the non-uniqueness of solutions regarding the targeted acoustic property. To address these challenges, we propose a deep learning-based approach capable of generating multiple metaporous designs that meet desired acoustic performance. Specifically, our proposed method integrates a generative adversarial network (GAN) framework into a surrogate network, guiding the generation process to ensure geometric diversity and physical consistency with the desired sound absorption coefficients. Both qualitative and quantitative evaluations confirm that the proposed model successfully generates a diverse range of unit-cell configurations that achieve the specified sound absorption behavior. Experimental results show that the proposed method outperforms existing approaches, increasing design diversity by approximately 85.8% and improving physical consistency by 64.8%. Additionally, we assess the advantages of our approach in terms of computational efficiency and design interactivity, demonstrating its capability to facilitate the exploration of metaporous designs with the desired characteristics. This study holds the potential to enhance and expedite the design process for advancing metamaterial discovery across various engineering disciplines.}, archive = {J_EAAI}, author = {Sooyoung Lee and Jihun Lee and Joong Seok Lee and Seungchul Lee}, doi = {10.1016/j.engappai.2025.111172}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111172}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning-accelerated multiple design generation for sound-absorbing metaporous materials}, volume = {156}, year = {2025}, }
• (2025). Machine learning approaches to predict oxidative potential of fine particulate matter based on chemical constituents. EAAI, 156, 111170. (www)

  @article{ , abstract = {Exposure to fine particulate matter (PM 2.5 ) poses significant health risks, primarily due to its oxidative potential (OP), which induces oxidative stress and related diseases. This study aimed to predict the OP of PM 2.5 based on its chemical constituents using machine learning (ML) models. We collected 119 p.m. 2.5 samples from Seoul, Korea, between 2019 and 2021, and analyzed their chemical composition and OP using the dithiothreitol (DTT) assay. Three ML models—k-Nearest Neighbors (kNN), Random Forest (RF), and Fully Connected Deep Neural Network (FCDNN)—were developed to predict OP. Among them, the RF model demonstrated the highest prediction accuracy, with coefficient of determination (R 2 ) values ranging from 0.88 to 0.89 for training data and 0.36 to 0.62 for test data, followed by Extreme Gradient Boosting (XGBoost) and FCDNN with test R 2 values up to 0.53 and 0.39, respectively. Explainable Artificial Intelligence (AI) techniques, specifically feature importance and SHapley Additive exPlanations (SHAP), were employed to enhance the interpretability of the model, revealing the significant contributions of various chemical constituents. The study underscores the mixed effects of multiple factors on OP and highlights the potential of AI in providing robust predictive tools for environmental health. As OP measurement automation progresses, the availability of large datasets will further improve the accuracy and applicability of AI models, facilitating better health risk assessments and policy-making.}, archive = {J_EAAI}, author = {Jongkeun Lee and Young Su Lee and Joo-Ae Kim and Seulki Jeong}, doi = {10.1016/j.engappai.2025.111170}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111170}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning approaches to predict oxidative potential of fine particulate matter based on chemical constituents}, volume = {156}, year = {2025}, }
• (2025). Domain generalization plant leaf disease recognition: Toward from laboratory to field. EAAI, 156, 111168. (www)

  @article{ , abstract = {Existing plant leaf disease recognition methods have achieved remarkable progress based on the independent and identically distributed assumption. However, two challenges remain in plant leaf disease recognition. First, the distribution between the training data and the real-world scene differs due to the diversity of illumination, style, and background, which results in models that do not generalize well to unseen domains. Second, plant leaf disease recognition lacks unified and public benchmarks to evaluate the generalization ability of models in unseen domains. Although a few methods apply the model trained on the laboratory datasets to the field data, the field data are always private. Based on these challenges, first, multiple benchmarks are proposed for domain generalization plant leaf disease recognition, where the public datasets collected in the laboratory are set to the source domain, and the public datasets collected in the field are set to the unseen domain. Second, these benchmarks are evaluated on different models, including convolutional neural networks, Transformer-based models, and self-supervised vision foundation models, as baselines. Finally, we propose a novel approach for domain generalization plant leaf disease recognition, which mainly contains a style-sensitive distillation module and a logit-prototype matching strategy. The former aims to eliminate domain-specific information about features extracted by the encoder. The latter aims to learn domain-invariant prototypical logits of plant leaf diseases to avoid the disturbance of domain-specific information, such as irrelevant background, styles, etc. Extensive experiments demonstrate that the proposed approach achieves superior performance to the baseline on multiple benchmarks. The benchmarks and related code are released at the GitHub repository: https://github.com/seabearlmx/DG-PLDR .}, archive = {J_EAAI}, author = {Kun Zhan and Yingqiong Peng and Muxin Liao and Yinglong Wang}, doi = {10.1016/j.engappai.2025.111168}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111168}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Domain generalization plant leaf disease recognition: Toward from laboratory to field}, volume = {156}, year = {2025}, }
• (2025). Scene adaptation network for visual-thermal urban scene semantic segmentation. EAAI, 156, 111166. (www)

  @article{ , abstract = {As one aspect of an intelligent transportation system, urban scene segmentation is important for understanding and analyzing a scene as part of the operation of an autonomous driving system. Over recent decades, research into semantic segmentation has made great breakthroughs due to the development of deep learning. However, in complex urban scenes, the performance of deep learning models is usually seriously affected, and the issue of how to effectively combine information from different spectra, such as visual images with red, green, and blue (RGB) colors and thermal infrared (TIR) images for urban scene segmentation has become a hot topic. To achieve effective feature extraction and fusion from different spectra while simultaneously considering the characteristics of urban scenes, a scene adaptation network (SAN) is proposed here to dynamically adjust and incorporate the features from RGB and TIR images based on the scene conditions. We employ multi-level operations and a self-attention module to fuse and enhance the semantic features. In addition, to promote the learning of the correlated and complementary information between different spectra, we design a self-supervised learning scheme for the training of our SAN. Through comprehensive experiments on three visual-thermal urban scene semantic segmentation datasets, we demonstrate the effectiveness and superiority of the proposed SAN against other state-of-the-art approaches.}, archive = {J_EAAI}, author = {Houwang Zhang and Yong-Jie Li and Leanne Lai-Hang Chan}, doi = {10.1016/j.engappai.2025.111166}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111166}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Scene adaptation network for visual-thermal urban scene semantic segmentation}, volume = {156}, year = {2025}, }
• (2025). A general framework for chromosomal anomaly detection based on dual constraints of nearest-neighbor and regionality. EAAI, 156, 111165. (www)

  @article{ , abstract = {The precise identification of structural chromosomal abnormalities (SCA) is essential for the diagnosis of genetic disorders and malignancies. Traditional karyotype analysis is labor-intensive and necessitates the expertise of cytogeneticists. We propose a dual-constraint enhanced framework that combines nearest-neighbor contrastive learning with one-class classification, facilitating automated abnormality detection without the need for anomalous data. Initially, positive sample pairs are constructed utilizing a Chromosomal Query Library (CQL). This process involves the dynamic selection of nearest neighbors, employing soft nearest neighbor selection and cosine similarity to improve feature consistency. Gaussian noise injection enhances generalization by diversifying representations, whereas a momentum update refines CQL embeddings. The Chromosome Banding module (CB module) extracts chromosomal features at multiple scales, whereas the Chromosome Batch Perception module (CBP module) emphasizes challenging samples through spatial and channel attention mechanisms. In the second stage, we present ChromosomeCutMix to create synthetic chromosomal anomalies, enhancing inter-class separation and improving anomaly detection. The proposed framework attains a classification accuracy of 97.32% and an F1-score of 96.69%, surpassing current methodologies in terms of sensitivity and robustness. Validated on public and clinical datasets, it offers dependable localization of biological anomalies and automated cytogenetic diagnostics, thereby enhancing the analysis of genetic disorders.}, archive = {J_EAAI}, author = {Yue Hao and Xin Wang and Ge Song and Zhiyuan Li and Lei Wang and Lingwei Li and Yongqi Nie and Peng Wang}, doi = {10.1016/j.engappai.2025.111165}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111165}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A general framework for chromosomal anomaly detection based on dual constraints of nearest-neighbor and regionality}, volume = {156}, year = {2025}, }
• (2025). Semi-supervised dual-constraint centroid contrastive prototypical network for flip chip defect detection under limited labeled data. EAAI, 156, 111163. (www)

  @article{ , abstract = {Flip chips are widely used in electronic systems for defense, aerospace, and other applications where packaging reliability is critical. However, flip chip defect samples present a variety of defect types and few samples with labels in actual industrial applications. The paucity of labeled defect samples indicates that the existing data volume cannot be matched with deep learning detection models. Therefore, flip chip intelligent defect detection faces the problems of poor model adaptability and weak generalization performance. As a solution to these problems, a semi-supervised dual-constraint centroid contrastive prototypical network (SSDCPN) for flip chip defect detection under limited labeled data is proposed in this paper. First, a prototype-based supervised contrastive learning strategy is developed to construct the contrastive prototypical network, which increases the inter-class sparsity and intra-class compactness of features to acquire more discriminative features. Then, to address the susceptibility of the support set prototypes to outliers, dual constraints are imposed on the support set prototypes to calibrate and refine the prototypes. Finally, a pseudo-labeled sample selection mechanism based on epistemic uncertainty and entropy is proposed to obtain rich semi-supervised information to guide the model training. The mechanism can select high-confidence pseudo-labeled samples that can complement the training samples to further strengthen the generalization performance of the model. Defect detection experiments on flip chip vibration signals indicate that the present method is superior to other methods in the case of limited labeled samples.}, archive = {J_EAAI}, author = {Yunxia Lou and Lei Su and Jiefei Gu and Xinwei Zhao and Ke Li and Michael Pecht}, doi = {10.1016/j.engappai.2025.111163}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111163}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised dual-constraint centroid contrastive prototypical network for flip chip defect detection under limited labeled data}, volume = {156}, year = {2025}, }
• (2025). Towards mechanized harvesting of pineapples: A masked self-attention instance segmentation network and pineapple detection dataset. EAAI, 156, 111162. (www)

  @article{ , abstract = {Automated pineapple harvesting using robotic and computer systems has been a pressing issue that researchers need to tackle urgently. However, varying light conditions in orchards, complex environments, and leaf shading present significant challenges for accurate, real-time pineapple identification and localization. Furthermore, the damage to fruits caused by robotic harvesting further limits the development of automated pineapple harvesting. In this study, we propose a real-time instance segmentation scheme to address the above two issues simultaneously. Correspondingly, we design a masked self-attention instance segmentation network based on mixed supervised learning (MAISNet) to quickly extract pineapples’ positional and geometric information and reduce the damage rate during robotic arm grasping. First, to meet the real-time needs of the robotic arm, a one-stage detection neural network is employed as the baseline model, achieving fast instance segmentation. Second, we incorporate a masked self-attention module to efficiently identify pineapple regions, reducing interference from irrelevant information. Third, we design a mixed supervised learning approach that allows the model to have some degree of uncertainty, enhancing the model’s ability to recognize occluded regions while reducing the over-reliance on labels. At the same time, to promote the pineapple detection field’s development and train the above algorithm, we present a public pineapple dataset. It is collected from real orchards, involves diverse complex scenarios, and is carefully labeled by hand. Many ablation experiments and comparison experiments demonstrate the validity and superiority of the method proposed in this study. Notably, experiments on edge devices verified the practical applicability of our approach in mechanized pineapple harvesting. This research changes the original paradigm of detecting only the location by accurately delineating fruit contours for direct robotic grasping. Consequently, it significantly reduces fruit damage during mechanical harvesting and advances the feasibility of large-scale automation in agriculture.}, archive = {J_EAAI}, author = {Zhe Shan and Songtao Ye and Cong Lin and Zhong Xue}, doi = {10.1016/j.engappai.2025.111162}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111162}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards mechanized harvesting of pineapples: A masked self-attention instance segmentation network and pineapple detection dataset}, volume = {156}, year = {2025}, }
• (2025). Predicting mechanical properties of magnetorheological elastomers during the manufacturing process using a new machine learning method. EAAI, 156, 111160. (www)

  @article{ , abstract = {Precisely predicting the shear storage modulus of magnetorheological elastomer (MRE) is crucial for effective vibration control. Traditional methods, however, are time-consuming and resource-intensive. This study introduces a novel hybrid deep learning model (RICBM) to efficiently predict this modulus. RICBM combines Random Forest (RF), an improved football team training algorithm, convolutional neural networks (CNN), bidirectional long short-term memory networks (BiLSTM), and multi-head attention (MAT) mechanisms. Initially, the significance of chosen input features is assessed through the RF. Subsequently, spatial pyramid matching (SPM) chaotic mapping and nonlinear weighting factors are employed to enhance the performance of the Football team training algorithm (FTTA). Next, a CNN-BiLSTM model is developed and the improved FTTA (IFTTA) is utilized to refine its parameters. Ultimately, the multi-head attention mechanism is utilized to highlight crucial input features, thereby further enhancing the predictive capabilities of the model. MRE samples with varied preparation parameters are prepared and tested by a rheometer in this study, resulting in a database of MRE shear storage modulus with 6 input and 1 output feature. The proposed model is applied to this database, alongside various comparative models. The experimental results demonstrate that RF effectively processes data and enhances prediction accuracy. The IFTTA enhances the CNN-BiLSTM model's predictive accuracy for the shear storage modulus of MRE. The resulting model shows significant effectiveness in making these predictions.}, archive = {J_EAAI}, author = {Qiyu Wang and Lai Peng and Yurui Shen and Hua Dezheng and Xinhua Liu and Zhixiong Li and Sumika Chauhan and Govind Vashishtha}, doi = {10.1016/j.engappai.2025.111160}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111160}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predicting mechanical properties of magnetorheological elastomers during the manufacturing process using a new machine learning method}, volume = {156}, year = {2025}, }
• (2025). Machine learning study for three-dimensional magnetohydrodynamics casson fluid flow with cattaneo-christov heat flux using linear regression technique: Application in engineering science and technology. EAAI, 156, 111159. (www)

  @article{ , abstract = {This paper explores Casson fluid three-dimensional laminar steady flow across an expanding layer, including Cattaneo-Christov and magnetohydrodynamic impact. A similarity transformation is carried out and leads to a set of coupled ordinary differential equations representing the governing equations that are numerically solved using bvp4c in MATLAB. The paper uses artificial neural networking-based Levenberg-Marquardt scheme and machine learning-based linear regression model defined by Y = p 1 ∗ x + p 2 , which examine the interactions between the important parameters and fluid behavior. For scenario 1, the greatest coefficient of determination value is 0.9315, indicating a solid fit and effective analysis of the underlying pattern. The velocity profile depicts a reduction with an increase in the Casson fluid parameter, representing higher resistance to deformation. In contrast, in the concentration profile, a reduction with an increase in Lewis number is noted, showing lower mass transfer rates because of dominant thermal diffusion. Isothermal and contour plots of the Nusselt number show significant heat transfer variations, helping to analyze the thermal management aspects. The zero-error line indicates the least amount of error when the Casson fluid parameter is used, and the mean square error value ranges from 10 ‗ − 6 to 10 ‗ − 10 . The Lewis number and heat flux indicate greater sensitivity to the Sherwood number and Nusselt number, respectively. Applications range from advanced energy-efficient technology, such as heat exchangers and cooling electronics, to healthcare and medical fields, such as medical bills, insurance, the simulation of blood flow in specific medical equipment, industries or food manufacturing systems, etc. To provide techniques for forecasting or a deeper knowledge of complex fluid dynamics for a variety of practical engineering and research objectives, the study integrates a numerical approach with machine learning.}, archive = {J_EAAI}, author = {Sohaib Abdal and Talal Taha and Nehad Ali Shah and Se-Jin Yook}, doi = {10.1016/j.engappai.2025.111159}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111159}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning study for three-dimensional magnetohydrodynamics casson fluid flow with cattaneo-christov heat flux using linear regression technique: Application in engineering science and technology}, volume = {156}, year = {2025}, }
• (2025). Deep learning-based real-time estimation of transcranial focused ultrasound acoustic field. EAAI, 156, 111157. (www)

  @article{ , abstract = {Transcranial focused ultrasound (tFUS) techniques have garnered considerable attention as a novel noninvasive brain stimulation modality due to their high spatial specificity and depth penetration. However, estimating the intensity, location, and shape of the ultrasound focus is challenging due to wave distortion through the inhomogeneous skull. Because conventional imaging methods cannot capture low-intensity acoustic foci, numerical simulations are required to estimate intracranial pressure fields. However, such simulations are computationally intensive, limiting real-time use. In this study, we introduce a deep learning-based surrogate model to enable real-time estimation of the intracranial acoustic field distribution of tFUS. The proposed model effectively captures skull computed tomography (CT) features via a pre-trained deep neural network and includes two modules: one predicts acoustic field distributions, and the other estimates peak pressure values to enhance overall accuracy. The model was trained using data from 13 cranial CT scans and validated against direct field measurements from three ex vivo calvaria. The proposed model demonstrated high accuracy in focal point estimation, achieving a peak pressure ratio error of 3.94 % and a focal position error of 2.46 mm, indicating precise localization of the ultrasound focus. For focal volume prediction, the model exhibited a maximum boundary error of 5.90 mm while maintaining a focal volume conformity of 81 %. Notably, the inference time was 16 ms, which is significantly faster than conventional numerical simulations, ensuring feasibility for real-time applications. This method facilitates precise intracranial targeting, significantly enhancing clinical viability of tFUS for therapeutic applications, including functional neuromodulation.}, archive = {J_EAAI}, author = {Minyeong Jang and Minwook Choi and Insu Jeong and Seung-Schik Yoo and Kyungho Yoon and Gunwoo Noh}, doi = {10.1016/j.engappai.2025.111157}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111157}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning-based real-time estimation of transcranial focused ultrasound acoustic field}, volume = {156}, year = {2025}, }
• (2025). Robust non-integer predictive control for wind turbine pitch angle regulation in full load regions using deep on-policy learning. EAAI, 156, 111156. (www)

  @article{ , abstract = {The pitch angle regulation of wind turbine (WT) plants faces serious challenges due to the nonlinear behavior of these systems and remarkable moment of inertia, along with stochastic changes in wind speed. In this paper, an adaptive pitch angle control based on fractional order model predictive control (FOMPC) is developed for variable speed of WT under full load region to regulate the rotor speed and to limit the output power. In particular, the deep state-action-reward-state-action (SARSA) is utilized to tune critical parameters of FOMPC, which is useful to stabilize rotor speed by adapting to the changes of turbulent wind conditions. By defining a reward function based on the specific goals of WT systems, the deep neural networks (DNNs) of the SARSA agent learn an effective policy for stabilizing the wind turbine, adjusting the FOMPC-based pitch angle control in response to varying wind conditions to capture the maximum output. By utilizing the training capability of DNNs, the FOMPC controller can dynamically respond to the changes in the WT system. The deployment of deep SARSA makes the intelligent controller to automatically learn and implement optimal control strategies in complex, dynamic environments, enhancing overall system performance and resilience. For comparison purposes, three other pitch angle controllers including the basic model predictive control (MPC), gain scheduling proportional-integral (GSPI) controller, and proportional-integral (PI) controller were also designed for the case-study. The feasibility of designed pitch angle controllers for WT stabilization was verified under two different operation conditions: normal and extreme stochastic wind circumstances. Dynamic responses and numerical analysis confirmed the suggested FOMPC-based deep SARSA agent can provide superior transient behavior and a higher level of robustness than the MPC, GSPI, and PI controller when the WT system is working at full load region.}, archive = {J_EAAI}, author = {Mai The Vu and Van-Truong Nguyen and Quang Thang Do and Wonkeun Youn and Tuan Hai Nguyen}, doi = {10.1016/j.engappai.2025.111156}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111156}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust non-integer predictive control for wind turbine pitch angle regulation in full load regions using deep on-policy learning}, volume = {156}, year = {2025}, }
• (2025). Convolutional self-attention with adaptive channel-attention network for obstructive sleep apnea detection using limited training data. EAAI, 156, 111154. (www)

  @article{ , abstract = {Obstructive sleep apnea (OSA) is a chronic sleep disorder caused by blockage of the upper airway for at least 10 s due to the collapsing of the tongue and soft palate. OSA can cause serious health problems including hypertension and coronary heart. Polysomnography is a technique to simultaneously record physiological signals such as electroencephalograms, electrooculograms, electrocardiograms (ECGs) etc., to diagnose various diseases including OSA. However, the process is time-consuming and tedious. Therefore, detecting OSA from ECGs (electrical signals recording heart variability using electrodes) is an alternative that can be extended to wearable devices. However, two challenges hinder their real-world applications: 1) Performance is directly proportional to the data size, and 2) algorithms are not robust for cross-dataset evaluation. We propose a novel deep-learning model called convolutional self-attention with adaptive channel-attention network (CSAC-Net) to address these issues. Specifically, the first issue is addressed by using the proposed Convolutional self-attention module in a multi-scale projection approach and fusing the features at the end. This enables the exploitation of long-range dependencies with diverse feature vectors. The second issue is addressed by leveraging invariant mapping through the proposed adaptive channel-attention (ACA) and inter-feature attention (IFA) modules. ACA module fuses multi-level features to embed adaptive characteristics while the IFA module exploits features from different stage to preserve the originality of features. To the best of our knowledge, this is the first study to address the underlying issues. Extensive experiments validate the effectiveness of CSAC-Net using two open databases: physiologic signal network apnea electrocardiogram (PhysioNet Apnea-ECG) and national sleep research resource best apnea interventions in research (NSRR-BestAIR). Their respective accuracies are respectively 93.4 % and 76.1 %, outperforming the state-of-the-art methods. Furthermore, the robustness of the CSAC-Net is validated through cross-database evaluation using various open databases.}, archive = {J_EAAI}, author = {Nadeem Ullah and Haseeb Sultan and Jin Seong Hong and Seung Gu Kim and Rehan Akram and Kang Ryoung Park}, doi = {10.1016/j.engappai.2025.111154}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111154}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Convolutional self-attention with adaptive channel-attention network for obstructive sleep apnea detection using limited training data}, volume = {156}, year = {2025}, }
• (2025). Multi-step control method of traffic flow data quality based on spatiotemporal similarity at video frame rate. EAAI, 156, 111153. (www)

  @article{ , abstract = {Traffic flow data is a quantitative description of traffic operation status. It lays a data foundation for optimizing traffic management, improving travel efficiency and ensuring traffic safety. Traffic flow data quality is crucial to the construction and operation of intelligent transportation system. At present, traffic flow data mostly come from vehicle detectors, but its long sampling period and sparse sampling points affect the quality control effect of traffic flow data. To solve these problems, we propose a multi-step automatic control method of traffic flow data quality based on spatiotemporal similarity. In this framework, we divide the data quality control process into preliminary repair and combinatorial optimization by combining the spatiotemporal advantages of the cross-sectional traffic flow data at video frame rate. Firstly, according to the different characteristics of the missing data, the preliminary repair is carried out automatically. For the missing data at the section level, the combined repair method based on the preprocessing of continuous missing data and the segmented low-order interpolation is proposed. For the missing data at road network level, the adaptive weight exponential smoothing method based on the time similarity is proposed. Then, a multi-sectional optimization model of based on spatiotemporal similarity was constructed to further optimize the preliminary repair results. The experimental results show that the preliminary repair method proposed in this paper is superior to other baseline models, and the combined optimization model based on the preliminary repair data greatly improves the data repair effect, which is suitable for different missing rates and types, and has certain competitiveness in traffic flow data quality control.}, archive = {J_EAAI}, author = {Yue Chen and Jian Lu}, doi = {10.1016/j.engappai.2025.111153}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111153}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-step control method of traffic flow data quality based on spatiotemporal similarity at video frame rate}, volume = {156}, year = {2025}, }
• (2025). Planning scheme of artificial assembly posture and arm movement path in narrow space. EAAI, 156, 111152. (www)

  @article{ , abstract = {Manual assembly in a narrow space involves problems of low efficiency and difficult assembly. In view of the lack of assembly process planning and assisted manual assembly in this kind of scenario, a hybrid modeling simulation method of human posture was proposed. This method combined the characteristics of manual assembly in narrow space. The assembly planning process was divided into two parts: trunk and lower limb posture planning and human arm movement planning, to reduce the complexity of planning and the difficulty of manual assembly. In the posture planning part, this study solved for the human trunk and lower limbs by establishing a multi-objective optimization model and achieved automatic screening of assembly posture according to the weight of each target element. Arm movement planning involved a neural network of assembly spaces to guide the sampling process of the path planner combined with the inverse solution of arm kinematics for environmental collision detection to quickly obtain a feasible collision-free arm movement path from the initial position to the assembly target. Finally, the feasibility of the method in a narrow space was verified by building a scene and carrying out the corresponding manual assembly operation experiments.}, archive = {J_EAAI}, author = {Yizhen Zheng and Yuefeng Li and Xudong Pan and Fanwei Meng and Changyu Chen}, doi = {10.1016/j.engappai.2025.111152}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111152}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Planning scheme of artificial assembly posture and arm movement path in narrow space}, volume = {156}, year = {2025}, }
• (2025). Progressive self-supervised framework for anomaly detection in hyperspectral images. EAAI, 156, 111151. (www)

  @article{ , abstract = {Hyperspectral image anomaly detection focuses on identifying anomalous targets within hyperspectral images. Many existing techniques rely on explicit priors from hyperspectral datasets and external data, such as sparsity, low rank, and pretrained neural networks. However, these explicit priors can limit the ability to capture subtle complexities in hyperspectral data, and a domain gap problem often exists between hyperspectral and external data. To address these issues, this paper presents a progressive self-supervised framework that eliminates the need for extensive training. The framework learns implicit information from the anomalous hyperspectral image itself, iteratively updating this information to progressively guide the reconstruction of the background hyperspectral image. We introduce an implicit neural prior, termed the progressive prior, through an untrained over-parameterized neural network. This prior enhances the distinction between background and anomaly targets during the iterative process. Since the progressive prior is derived directly from the hyperspectral image data, it does not require external datasets, thereby eliminating the domain gap problem. Extensive qualitative and quantitative evaluations across six hyperspectral image datasets demonstrate that our method achieves an increase of 0.0129 in the area under the curve compared to state-of-the-art anomaly detection methods.}, archive = {J_EAAI}, author = {Pan Liu and Yuanyang Bu and Yongqiang Zhao and Seong G. Kong}, doi = {10.1016/j.engappai.2025.111151}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111151}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Progressive self-supervised framework for anomaly detection in hyperspectral images}, volume = {156}, year = {2025}, }
• (2025). Exploiting non-likelihood adversarial training for chinese counterfactual data augmentation. EAAI, 156, 111149. (www)

  @article{ , abstract = {Chinese natural language processing (NLP) models often struggle with out-of-distribution generalization due to data biases and shortcut learning. Existing counterfactual data augmentation (CDA) methods heavily rely on manual intervention, limiting their scalability. To address this issue, we propose CAT-CDA ( C hinese Non-likelihood A dversarial T raining for C ounterfactual D ata A ugmentation), an automated approach that generates counterfactual data with opposite labels while maintaining semantic consistency. CAT-CDA employs a classifier to identify causal features and optimizes a generator through non-likelihood adversarial training, ensuring both diversity and fluency. Experimental results demonstrate that CAT-CDA significantly improves model robustness, enhances out-of-distribution performance, and mitigates shortcut learning while requiring lower computational resources than existing methods, making it highly applicable to Chinese NLP tasks.}, archive = {J_EAAI}, author = {Dezhi An and Fang Wang and Shengcai Zhang and Yan Li}, doi = {10.1016/j.engappai.2025.111149}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111149}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploiting non-likelihood adversarial training for chinese counterfactual data augmentation}, volume = {156}, year = {2025}, }
• (2025). A camouflage target classification method based on spectral difference enhancement and pixel-pair features in land-based hyperspectral images. EAAI, 156, 111141. (www)

  @article{ , abstract = {Hyperspectral images are capable of capturing rich spatial and spectral information of targets, rendering them particularly valuable for camouflage target detection and classification applications. However, as camouflage technologies continue to advance, the spectral similarity between camouflaged targets and their backgrounds has become increasingly pronounced, presenting significant challenges for camouflaged target classification in hyperspectral imagery. To overcome this challenge, this paper introduces a novel land-based hyperspectral image classification approach for camouflaged targets, termed Spectral Difference Enhancement and Pixel-Pair Features (SDE-PPF). The proposed methodology initially conducts spectral rearrangement of the hyperspectral image based on target spectral characteristics, which induces oscillatory patterns in the background spectrum. Subsequently, first-order spectral differentiation coupled with nonlinear processing is applied to the rearranged hyperspectral image to effectively amplify subtle spectral differences between targets and backgrounds, thereby improving their discriminability. Following spectral enhancement, the method constructs pixel pairs from the processed hyperspectral image and employs a convolutional neural network to extract pixel-pair features. Network parameters are optimized through comprehensive analysis of pixel-pair sample relationships. During testing, the central pixel is systematically paired with its neighboring pixels, and classification is performed using the trained model. Ultimately, the final classification of each central pixel is determined through a voting mechanism that consolidates all classification results. Comprehensive experiments were performed on four distinct land-based hyperspectral image datasets containing camouflage targets. The experimental results demonstrate that the proposed SDE-PPF method outperforms conventional hyperspectral image classification approaches, achieving remarkable average classification accuracies of 98.46 %, 99.05 %, 98.94 %, and 99.21 % for detecting camouflage targets against grassland, barren Grassland, withered leaf, and shrubbery backgrounds, respectively. This innovative approach establishes an effective and robust technical solution for camouflage target classification and detection, exhibiting considerable potential for diverse practical applications.}, archive = {J_EAAI}, author = {Jiale Zhao and Dan Fang and Jiaju Ying and Yudan Chen and Qi Chen and Qianghui Wang and Guanglong Wang and Bing Zhou}, doi = {10.1016/j.engappai.2025.111141}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111141}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A camouflage target classification method based on spectral difference enhancement and pixel-pair features in land-based hyperspectral images}, volume = {156}, year = {2025}, }
• (2025). Artificial intelligence based semantic segmentation on aerial images with variational mode decomposition. EAAI, 156, 111140. (www)

  @article{ , abstract = {Segmentation of objects in an image is a significant task in the domain of computer vision that entails algorithms customized to the nature of images and the type of their constituent objects. Although modern deep learning models apropos aerial images deliver state-of-the-art performance, major challenges in their segmentation of aerial images are fewer training samples, and segmentation of images with small and multiple objects. This paper proposes Variational Mode Decomposition (VMD) based semantic segmentation on aerial images for six different classes of objects. VMD extracts the predominant frequency components in an image, offering a better representation of the input image before segmentation. These frequency components are coupled with basic deep-learning models, ensuing in better segmentation results. This paper suggests a novel mode selection method for extracting coarser information aiding in efficient segmentation. Also, the paper proposes a solution for the problem of fewer training samples through VMD. The proposed methodology was experimented on images of three publicly available datasets: Dubai, Northwestern Polytechnical University-Very High Resolution (NWPU-VHR10), and Aerial Image Datasets (AID). Based on the quantitative and qualitative analysis of the results on the Dubai dataset, it is inferred that the performance of basic deep neural networks has improved when coupled with the VMD technique. Among the deep neural networks experimented on in this paper, VMD-based UNet with Vision Transformer outperformed the other models in terms of visual comparison, mean Intersection over Union (mIoU), and mean F1 score.}, archive = {J_EAAI}, author = {Anupa Vijai and S. Padmavathi and D. Venkataraman}, doi = {10.1016/j.engappai.2025.111140}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111140}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence based semantic segmentation on aerial images with variational mode decomposition}, volume = {156}, year = {2025}, }
• (2025). Corrigendum to “Loss of coolant accident monitoring and pipe break diagnosis in pressurized water reactors using bayesian-optimized long short-term memory models” [Eng. appl. artif. intellig. 149 (2025) 110531]. EAAI, 156, 111139. (www)

  @article{ , archive = {J_EAAI}, author = {Johndel Obra and Shuichiro Miwa}, doi = {10.1016/j.engappai.2025.111139}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111139}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Corrigendum to “Loss of coolant accident monitoring and pipe break diagnosis in pressurized water reactors using bayesian-optimized long short-term memory models” [Eng. appl. artif. intellig. 149 (2025) 110531]}, volume = {156}, year = {2025}, }
• (2025). A hybrid detection method for YouTube fake news using related video data. EAAI, 156, 111130. (www)

  @article{ , abstract = {Fake news has rapidly evolved from simple textual forms to complex multimedia content, with YouTube emerging as a major platform for its dissemination. However, most existing detection techniques focus solely on content-based features, often overlooking the contextual signals embedded in related video data. We propose a novel hybrid detection framework that integrates content- and context-based approaches to address this limitation. Our method combines multimodal features extracted from textual and visual components of the original video, along with contextual information from related videos. Text data is processed using word embedding techniques, while image and visual elements are analyzed through Convolutional Neural Networks. To evaluate our framework's robustness and generalizability, we conducted experiments on two complementary datasets: a publicly available multilingual dataset and a newly constructed Korean dataset. The results show that the proposed method achieves a 0.1–9.7 % improvement in detection accuracy compared to traditional content-only approaches. These findings underscore the value of leveraging related video information for more reliable fake news detection, and they confirm that our hybrid method can be effectively applied across diverse language and regional contexts. Our work contributes to mitigating the societal harm caused by misinformation on video-sharing platforms and offers practical insights into developing more robust multimedia fake news detection systems.}, archive = {J_EAAI}, author = {Junho Kim and Yongjun Shin and Gyeongho Jung and Hyunchul Ahn}, doi = {10.1016/j.engappai.2025.111130}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111130}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid detection method for YouTube fake news using related video data}, volume = {156}, year = {2025}, }
• (2025). One for all: Geometric structural-guided single domain generalization for vibration damper defect detection. EAAI, 156, 111129. (www)

  @article{ , abstract = {Improving the generalization capabilities of deep learning-based vibration damper defect detection models is crucial for their applicability in complex power scenarios. Existing models struggle to cope with scenario data not encountered during training, including diverse geographical regions, weather conditions, and lighting scenarios. To address these challenges, this paper introduces a novel single domain generalization(DG) method for vibration damper defect detection, aimed at enhancing the generalization capabilities of current models. Under analyzing the structural and morphological characteristics of vibration damper defects, this study selects their geometric structural features as domain invariant features(DIF), guiding model training with artificial samples generated via three dimensional(3D) modeling. To ensure comprehensive learning of the DIF of vibration damper defects, this study employs multi-scale contrastive learning(MCL) loss and multi-morphology similarity metric learning(MSM) loss during the feature extraction and candidate box processing stages, respectively. These approaches guide the model in extracting geometric features and achieving consistent expression of defect characteristics. The experimental results demonstrate that the proposed method outperforms advanced object detection models in DG for vibration damper defect detection, achieving an average DG accuracy of 68.6% across multiple target domains and a sensitivity of only 0.219, thereby realizing the “one-time training, everywhere application” concept in vibration damper defect detection models.}, archive = {J_EAAI}, author = {Ke Yang and Zhedong Hu and Xiaoyu Zhao and Qianming Wang and Dongyang Hu and Yongjie Zhai}, doi = {10.1016/j.engappai.2025.111129}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111129}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {One for all: Geometric structural-guided single domain generalization for vibration damper defect detection}, volume = {156}, year = {2025}, }
• (2025). Fire resistance rating prediction of timber-to-steel connections and design optimization informed by explainable machine learning. EAAI, 156, 111127. (www)

  @article{ , abstract = {Timber as a construction material is experiencing its renaissance, while fire safety is a critical factor for timber-based building design. Currently, the fire resistance rating of wood-steel-wood (WSW) connections is evaluated using empirical equations derived from experimental results. However, these equations consider a limited set of parameters and lack interpretability. This paper developed an explainable machine learning (ML) model considering comprehensive parameters related to connection’s configuration, based on 140 experimental and experimental-validated numerical data. The performances of various machine learning models are evaluated in terms of predicting the fire resistance rating of connections after hyperparameter tuning. The eXtreme Gradient Boosting (XGBoost) model outperforms other ML models ( R 2 = 0 . 93 ) and empirical equations. The local sensitivity analysis (LSA), global sensitivity analysis (GSA), and SHapley Additive exPlanations (SHAP) analysis are conducted based on the XGBoost model to investigate the contributions of nine parameters to the fire resistance rating. Both sensitivity analysis and SHAP analysis identify timber thickness and load ratio as the primary factors influencing fire resistance. Finally, the calibrated XGBoost model is incorporated into a non-dominated sorting genetic algorithm (NSGA-II) to optimize the design, aiming to minimize the self-weight of the connection while maximizing the fire resistance rating and load-carrying capacity of the connection subjected to constraints on limited dimensions.}, archive = {J_EAAI}, author = {Tongchen Han and Zhidong Zhang and Weiwei Wu}, doi = {10.1016/j.engappai.2025.111127}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111127}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fire resistance rating prediction of timber-to-steel connections and design optimization informed by explainable machine learning}, volume = {156}, year = {2025}, }
• (2025). Evolutionary multi-objective multi-agent deep reinforcement learning for sustainable maintenance scheduling. EAAI, 156, 111126. (www)

  @article{ , abstract = {In recent years, sustainability has emerged as a major priority for businesses across various industries, and the manufacturing sector is no exception. Production and maintenance processes now need to be economically profitable while also adopting practices that adhere to the principles of environmental integrity and social responsibility. This article explores an innovative approach aimed at optimizing maintenance scheduling from an economic perspective (considering maintenance, breakdown, downtime costs), an environmental perspective (considering the carbon footprint produced during production) and a social perspective (considering the fatigue experienced by technicians during maintenance activities). To the best of our knowledge, this is the first study to propose a manufacturing scheduling approach that considers all three pillars of sustainability. Another significant contribution of this research is the innovative way in which the optimization problem is addressed. We propose an evolutionary multi-objective multi-agent Deep Q-network-based approach, where multiple agents explore the preference space to maximize the hypervolume of these sustainable objectives. Our methodology uses industrially representative data that incorporate realistic machine degradation signals, carbon intensity indicators, and technician constraints. The results demonstrate the trade-offs between these objectives when compared to traditional maintenance policies such as corrective and condition-based maintenance, as well as different Deep Q-network policies trained with various preferences. Our approach demonstrates superior performance compared to both baselines. Specifically, we observe an 11.6% improvement in hypervolume over Deep Q-network and an 18.9% improvement over Proximal Policy Optimization, resulting in significantly increased profitability within the system.}, archive = {J_EAAI}, author = {Marcelo Luis Ruiz-Rodríguez and Sylvain Kubler and Jérémy Robert and Alexandre Voisin and Yves Le Traon}, doi = {10.1016/j.engappai.2025.111126}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111126}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evolutionary multi-objective multi-agent deep reinforcement learning for sustainable maintenance scheduling}, volume = {156}, year = {2025}, }
• (2025). Strategies for energy-efficient flow control leveraging deep reinforcement learning. EAAI, 156, 111125. (www)

  @article{ , abstract = {This study investigates active flow control in two-dimensional flows at a Reynolds number of 100 using Deep Reinforcement Learning (DRL). We utilize DRL to develop flow control strategies that enhance energy efficiency and minimize energy consumption, thereby addressing the limitations of traditional methods. We find that the optimal jet placement for both square and circular cylinders is at the main flow separation point, achieving the best balance between energy efficiency and control effectiveness. For the circular cylinder, positioning the jet at approximately 105°from the stagnation point requires only 1% of the inlet flow rate and achieves an 8% reduction in drag, with energy consumption one-third of that at other positions. For the square cylinder, placing the jet near the rear corner requires only 2% of the inlet flow rate, achieving a maximum drag reduction of 14.4%, whereas energy consumption near the front corner is 27 times higher, resulting in only 12% drag reduction. In multi-action control, the convergence speed and stability are lower compared to single-action control, but activating multiple jets significantly reduces initial energy consumption and improves energy efficiency. Physically, the interaction of the synthetic jet with the flow generates new vortices that modify the local flow structure, significantly enhancing the cylinder’s aerodynamic performance. Our control strategy achieves a superior balance between energy efficiency and control performance compared to previous studies, underscoring its significant potential to advance sustainable and effective flow control.}, archive = {J_EAAI}, author = {Wang Jia and Hang Xu}, doi = {10.1016/j.engappai.2025.111125}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111125}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Strategies for energy-efficient flow control leveraging deep reinforcement learning}, volume = {156}, year = {2025}, }
• (2025). CTPR: Contrastive transition predictive representation for reinforcement learning. EAAI, 156, 111124. (www)

  @article{ , abstract = {Learning policies from high-dimensional observations is a challenging problem for pixel-based reinforcement learning. Most existing pixel-based reinforcement learning methods struggle with the inefficiency of extracting meaningful state representations from raw pixel data, lacking temporal correlation and resulting in suboptimal performance. To this end, we propose a innovative method named contrastive transition predictive representation for reinforcement learning (CTPR), which utilizes contrastive learning and a transition model to efficiently extract high-level state representations from raw pixels for sample-efficient reinforcement learning. In the reinforcement learning component, we perform policy control based on the learned contrastive representations. We have evaluated the effectiveness of the proposed method by conducting numerous experiments on DeepMind Control, and the results show that our method has achieve significant improvements over the state-of-the-art methods.}, archive = {J_EAAI}, author = {Hao Sun and Changpeng Wang}, doi = {10.1016/j.engappai.2025.111124}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111124}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {CTPR: Contrastive transition predictive representation for reinforcement learning}, volume = {156}, year = {2025}, }
• (2025). LIVFusion: Luminance-optimized fusion of infrared and visible images with wavelet transformer. EAAI, 156, 111123. (www)

  @article{ , abstract = {In the field of image processing, effective fusion of infrared and visible images remains a significant challenge, particularly in terms of illumination adjustment and detail preservation. Current methods often struggle to adequately balance these aspects, resulting in degraded visual perception and texture detail. This compromises the ability to accurately infer realistic details and restore true colors, which are crucial for subsequent applications. To address these limitations, this study introduces an innovative image fusion framework called luminance-optimized fusion of infrared and visible images with wavelet transformer (LIVFusion). Specifically, considering that the wavelet transform can effectively capture local and global features of images, we propose a wavelet transformer fusion network (WTFNet) to enhance the texture details of the fused features. Furthermore, to further improve the fusion quality of images under low-light conditions, we integrate the scene-illumination disentangled network (SIDNet) with the WTFNet. This combination achieves precise brightness adjustment and superior feature integration, significantly improving the quality of image processing. Additionally, we propose a contrast equilibration loss to ensure harmonious integration of details and clarity across different lighting scenarios. Extensive evaluations demonstrate that our framework outperforms existing state-of-the-art technologies in rendering natural appearances and fine textures, thereby enhancing night image analysis for critical visual applications.}, archive = {J_EAAI}, author = {Dingli Hua and Qingmao Chen and Wenying Wen and Haonan Luo}, doi = {10.1016/j.engappai.2025.111123}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111123}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {LIVFusion: Luminance-optimized fusion of infrared and visible images with wavelet transformer}, volume = {156}, year = {2025}, }
• (2025). Enhancing spatiotemporal wind power forecasting with meta-learning in data-scarce environments. EAAI, 156, 111121. (www)

  @article{ , abstract = {Accurate wind power forecasting is critical for maintaining stable power grids, yet the inherent variability of wind and limited data availability for new wind farms present significant challenges. To address these issues, we present a novel artificial intelligence framework that integrates a self-attention enhanced Spatiotemporal Long Short-Term Memory (ST-LSTM) network with Model-Agnostic Meta-Learning (MAML), termed as the Meta-Learning Spatiotemporal Attention Long Short-Term Memory framework (MAML-STALSTM). This deep learning combination enables the model to effectively capture long-range spatiotemporal dependencies while rapidly adapting to new wind farm configurations or changing wind conditions with minimal training data. By employing rigorous data preprocessing techniques and ensuring temporal separation in data splitting, we mitigate potential data leakage and enhance the model’s generalizability. Extensive experiments conducted on both onshore and offshore wind farm datasets demonstrate that our artificial intelligence approach outperforms established baseline models, particularly excelling in data-scarce environments. Ablation studies highlight the crucial roles of the self-attention mechanism and meta-learning in improving forecasting accuracy, adaptation speed, and model robustness. These results emphasize the practical benefits of our approach in enhancing grid stability and supporting the seamless integration of wind energy, thereby contributing significantly to the advancement of sustainable energy solutions.}, archive = {J_EAAI}, author = {Renfang Wang and Jingtong Wu and Xu Cheng and Xiufeng Liu and Hong Qiu}, doi = {10.1016/j.engappai.2025.111121}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111121}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing spatiotemporal wind power forecasting with meta-learning in data-scarce environments}, volume = {156}, year = {2025}, }
• (2025). Fabrication and characterization of biological biosensors in sports injury treatment: High sensitivity of silver oxide using artificial neural network modeling. EAAI, 156, 111120. (www)

  @article{ , abstract = {This study focuses on the potential of biosensor technology to revolutionize sports injury management. Conventional methods for sports injury management, such as physical examinations and imaging techniques, often lack the sensitivity and real-time monitoring capabilities required to track the healing process effectively. These methods are also invasive, relying on blood sampling, which can be uncomfortable for athletes. In contrast, the proposed wearable biosensor offers a non-invasive, painless alternative by measuring biomarkers like myoglobin and creatine kinase in sweat. This study introduces a novel graphene field-effect transistor biosensor integrated into a wristband, combined with an artificial neural network (ANN) model to predict material properties and optimize biomarker detection. The results show the potential of this technology to revolutionize sports injury management by providing real-time, accurate, and non-invasive monitoring of injury progression and recovery. The results indicate that changes in compressive strength and porosity have an impact on dissolution rate, pore size growth, and chemical stability. Lower compressive strength leads to an increase in dissolution rate, while higher compressive strength promotes pore size growth and chemical stability. The accuracy of the ANN model's predictions was evaluated using linear regression and demonstrated acceptable error levels compared to experimental testing. Among the nanocomposite hydrogel scaffolds containing silver oxide nanoparticles, a specific sample showed noteworthy characteristics, including a compressive strength of 2.4 Mega Pascal, 55 % porosity, 22 % dissolution rate, 27 % pore size growth, and 65 % chemical stability.}, archive = {J_EAAI}, author = {Youliang Wu and B. Kamyab Moghadas and Dongqiang Gu and S. Salahshour and M. Hashemi}, doi = {10.1016/j.engappai.2025.111120}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111120}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fabrication and characterization of biological biosensors in sports injury treatment: High sensitivity of silver oxide using artificial neural network modeling}, volume = {156}, year = {2025}, }
• (2025). Multi-source perception data fusion of vessels in visual occlusion scenarios: Leveraging prior knowledge of vessel motion. EAAI, 156, 111118. (www)

  @article{ , abstract = {Automatic Identification System (AIS) and cameras are widely used in harbor and coastal supervision to detect ship movement. Integrating both can enhance the perception and monitoring of the navigation status of surrounding vessels. However, their integration faces several challenges. First, the two data sources have different coordinate systems and sampling frequencies. Additionally, in video data, visual occlusion during vessel encounters may lead to the loss or displacement of ship detection targets, significantly affecting the fusion of ship data. To address these issues, we incorporate prior knowledge of vessel motion to strengthen the model’s ability to identify and track occluded targets. Firstly, when detecting and tracking ships in video, we use occlusion prior knowledge and tracking results to evaluate and manage the occluded detection box areas. Secondly, when the occluded detection box disappears or undergoes severe deformation, we use the prior knowledge of ship motion characteristics from AIS and images to predict the detection box. Finally, we validated our improved method on the FVessel_v1.0 dataset, confirming its accuracy in data fusion under occlusion conditions. Compared with the state-of-the-art ship data fusion algorithm, our method improved the Multiple Object Fusion Accuracy (MOFA), Identification Precision (IDP), Identification Recall (IDR), and Identification F1 score (IDPF1) metrics by 3.01%, 1.33%, 1.57%, and 1.33%, and reduced MOFA by 1.85%. Additionally, by only changing the method of utilizing prior knowledge without altering the detection and tracking algorithms of the state-of-the-art ship data fusion algorithm, we improved the MOFA, IDP, IDR, and IDPF1 metrics by 2.65%, 1.94%, 0.62%, and 0.87%, respectively, and reduced MOFA by 1.16%.}, archive = {J_EAAI}, author = {Wei He and Wenbo He and Jinyu Lei and Sitong Wan and Zhiyuan Wang}, doi = {10.1016/j.engappai.2025.111118}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111118}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-source perception data fusion of vessels in visual occlusion scenarios: Leveraging prior knowledge of vessel motion}, volume = {156}, year = {2025}, }
• (2025). An ensemble velocity learning strategy for particle swarm optimization integrating multiple local search mechanisms. EAAI, 156, 111117. (www)

  @article{ , abstract = {Particle Swarm Optimization (PSO) is an effective and efficient metaheuristic algorithm that has been widely applied to valuable real-world applications. Since PSO was proposed nearly three decades, many PSO-based variants have been developed. In recent years, research on ensemble complementary operators and variants aimed at enhancing the performance of PSO has maintained a certain level of enthusiasm. In this work, an ensemble velocity learning strategy is proposed and combined with two local search methods: the Improved Sine Cosine Algorithm (ISCA) and Sequential Quadratic Programming (SQP). Meanwhile, to improve the effectiveness of the ensemble velocity learning strategy, elite learning, modified inertia weight, modified acceleration coefficients, and mutation strategy are incorporated. The proposed metaheuristic algorithm, named the Ensemble Velocity Learning Strategy for Particle Swarm Optimization Integrating Multiple Local Search Mechanisms (EVLS-PSOIMLSM), is tested on the Institute of Electrical and Electronics Engineers (IEEE) Congress on Evolutionary Computation 2017 (CEC2017) benchmark functions and applied to real-world problems, including the Gear Train Design (GTD) and the Planetary Gear Train Design (PGTD) problem. Compared to 14 recently proposed PSO-based variants and 11 non-PSO algorithms, the experimental results demonstrate that EVLS-PSOIMLSM achieves superior performance. The source code of EVLS-PSOIMLSM is provided at https://github.com/microhard1999/CODES .}, archive = {J_EAAI}, author = {Libin Hong and Zhantao Gu and Tianxiang Cui}, doi = {10.1016/j.engappai.2025.111117}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111117}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An ensemble velocity learning strategy for particle swarm optimization integrating multiple local search mechanisms}, volume = {156}, year = {2025}, }
• (2025). Role of noise elimination algorithms in speech processing applications: A comprehensive research and some experimental results. EAAI, 156, 111116. (www)

  @article{ , abstract = {The performance of speech-based systems is severely degraded due to the presence of background noise in real-world environments. Effective noise elimination algorithms are essential for enhancing speech quality and improving the performance of speech processing applications, such as voice activity detection (VAD) and speech encoding. Various speech enhancement techniques have been proposed to tackle this, and in this context, choosing an appropriate enhancement technique for improving speech quality and intelligibility is an important consideration. This paper presents a concise experimental review of different noise elimination techniques using objective and subjective metrics. The experiments are conducted on the noisy speech corpus (NOIZEUS) across different noise types and signal-to-noise ratio (SNR) levels. Comparative results indicate that the soft mask estimator with a priori SNR uncertainty (SMPR) is considerably more useful in enhancing speech quality. Furthermore, we analyze the SMPR performance in enhancing speech quality under various noise conditions, specifically focusing on their impact on speech encoding and VAD applications. Our results reveal that integrating the SMPR enhancement module into linear predictive coding (LPC)-based speech encoding system significantly improves speech quality. Additionally, the application of SMPR in VAD systems demonstrates notable improvements, enhancing the accuracy of speech detection.}, archive = {J_EAAI}, author = {Nagaraja B.G. and Thimmaraja Yadava G. and Raghudathesh G.P. and Jayanna H.S.}, doi = {10.1016/j.engappai.2025.111116}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111116}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Role of noise elimination algorithms in speech processing applications: A comprehensive research and some experimental results}, volume = {156}, year = {2025}, }
• (2025). A codebook-driven approach for low-light image enhancement. EAAI, 156, 111115. (www)

  @article{ , abstract = {Low-light image enhancement (LLIE) aims to improve low-illumination images. However, existing methods face two challenges: (1) uncertainty in restoration from diverse brightness degradations; (2) loss of texture and color information caused by noise suppression and light enhancement. In this paper, we propose a novel enhancement approach, CodeEnhance, by leveraging discrete codebook priors and image refinement to address these challenges. In particular, we reframe LLIE as learning an image-to-code mapping from low-light images to discrete codebook, which has been learned from high-quality images. To enhance this process, a Semantic Embedding Module (SEM) is introduced to integrate semantic information with low-level features, and a Codebook Shift (CS) mechanism, designed to adapt the pre-learned codebook to better suit the distinct characteristics of our low-light dataset. Additionally, we present an Interactive Feature Transformation (IFT) module to refine texture and color information during image reconstruction, allowing for interactive enhancement based on user preferences. Extensive experiments on both real-world and synthetic benchmarks demonstrate that the incorporation of prior knowledge and controllable information transfer significantly enhances LLIE performance in terms of quality and fidelity. The proposed CodeEnhance exhibits superior robustness to various degradations, including uneven illumination, noise, and color distortion. The code can be obtained from https://github.com/csxuwu/CodeEnhance or https://www.scholat.com/laizhihui.cn .}, archive = {J_EAAI}, author = {Xu Wu and Xianxu Hou and Zhihui Lai and Jie Zhou and Ya-nan Zhang and Witold Pedrycz and Linlin Shen}, doi = {10.1016/j.engappai.2025.111115}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111115}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A codebook-driven approach for low-light image enhancement}, volume = {156}, year = {2025}, }
• (2025). Transformers in pathological image analysis: A survey. EAAI, 156, 111114. (www)

  @article{ , abstract = {With the advancements of artificial intelligence, deep learning has emerged as the predominant approach in computational pathology. It is dedicated to automatically analyzing the intricate phenotype information embedded in various pathological images, with the goal of delivering more precise diagnoses, prognoses, and treatment recommendations for cancer patients. As the latest breakthrough in deep learning technology, Transformers are gaining traction in the realm of pathological image analysis by harnessing self-attention mechanisms to capture global information. Consequently, this study presents a comprehensive review of state-of-the-art models leveraging Transformers, applied across tasks such as classification, segmentation, and survival analysis in pathological image analysis. Initially, we delineate the concept and key components of Transformers, followed by a survey of their recent applications in pathology. These applications encompass pathological image classification, segmentation, lesion detection and localization, as well as the utilization of specific Transformer architectures for patient survival analysis. Subsequently, we delve into the challenges encountered in employing Transformers for pathological image analysis and speculate on future developmental trajectories. Our aim is to furnish readers with an exhaustive roadmap to deepen their comprehension of Transformer applications in pathology, thereby fostering the advancement of more sophisticated technologies and enabling more precise diagnoses and treatment strategies for cancer patients.}, archive = {J_EAAI}, author = {Liangliang Liu and Zhihong Liu and Jinpu Xie and Hongbo Qiao and Jing Chang}, doi = {10.1016/j.engappai.2025.111114}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111114}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformers in pathological image analysis: A survey}, volume = {156}, year = {2025}, }
• (2025). A self-feedback knowledge elicitation approach for chemical reaction predictions. EAAI, 156, 111112. (www)

  @article{ , abstract = {The task of chemical reaction predictions (CRPs) plays a pivotal role in advancing drug discovery and material science. However, while artificial intelligence (AI) has significantly advanced in capturing chemical reaction patterns, its effectiveness is still constrained by the vast and uncertain chemical reaction space and challenges in achieving high reaction selectivity, particularly due to existing methods’ limitations in fully leveraging the inherent knowledge within the data. To address these challenges, we introduce a data-curated self-feedback knowledge elicitation approach. This method starts from iterative optimization of molecular representations and facilitates the extraction of knowledge on chemical reaction types (RTs). Then, we employ adaptive prompt learning to infuse the prior knowledge into the large language model (LLM). As a result, we achieve significant enhancements: a 14.2% increase in retrosynthesis prediction accuracy, a 74.2% rise in reagent prediction accuracy, and an expansion in the model’s capability for handling multi-task chemical reactions. This research offers a novel paradigm for knowledge elicitation in scientific research and showcases the untapped potential of LLMs in CRPs.}, archive = {J_EAAI}, author = {Pengfei Liu and Jun Tao and Zhixiang Ren}, doi = {10.1016/j.engappai.2025.111112}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111112}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A self-feedback knowledge elicitation approach for chemical reaction predictions}, volume = {156}, year = {2025}, }
• (2025). Ensemble learning framework for detecting electricity theft in smart grids using weighted average method. EAAI, 156, 111111. (www)

  @article{ , abstract = {Electricity theft is an essential issue within non-technical losses of energy. Its complexity intensifies with the emergence of novel network attacks. Previous studies have highlighted the efficacy of machine learning (ML) techniques in detecting individual instances of electricity theft using raw data. However, insufficient depth of data utilization, inadequate detection of compound attacks, and poor interpretability of models limit their reliability. To overcome these predicament, electrical and environmental features are obtained through manual curation, and their relationship and temporal lag effects are evaluated using Pearson correlation coefficients. Moreover, a weighted average electricity theft detection algorithm based on Deep and Cross Network Version 2 (DCN_V2)-Transformer is proposed. The independent training and configuration optimization of the DCN_V2 and Transformer are implemented by grid search methodology, respectively. During the model integration, the optimized predictions yielded by the two models are adeptly combined according to appropriate weights. Furthermore, the decision-making process of Transformer is analyzed using visual attention weight methodology. The experimental results demonstrate a significant leap in the performance of the proposed model by considering environmental factors. The evaluation metrics perform remarkably, with an accuracy of 95.9%, precision of 94.8%, recall of 96.8%, F1 score of 95.8%, and false positive rate of 5%. Additionally, the comparative analysis verifies the advantages of this approach over convolutional neural networks (CNN), long short-term memory network (LSTM), CNN+LSTM, Wide+DeepCNN, support vector machine (SVM), and decision tree+SVM algorithms. This study provides a promising technique for electricity theft detection and assists in understanding the operating mechanism of detection model.}, archive = {J_EAAI}, author = {Kaiyang Zhang and Jun Wang and Yonghai Zhu and Yifei Si and Shanshan Yin and Hang Zhang}, doi = {10.1016/j.engappai.2025.111111}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111111}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ensemble learning framework for detecting electricity theft in smart grids using weighted average method}, volume = {156}, year = {2025}, }
• (2025). Automatic requirements elicitation from user-generated content: A review of data, methods, and representations. EAAI, 156, 111110. (www)

  @article{ , abstract = {Requirements elicitation (RE) is the first and most important phase in requirements engineering. In the era of big data, as online platforms have become one of the primary channels for users to provide product feedback, extensive research has been conducted to explore how customer requirements can be elicited from user-generated content (UGC) through artificial intelligence technologies such as text clustering, topic modeling, and machine learning. We systematically review the existing literature in the area of UGC-based RE by categorizing research according to three dimensions: UGC data sources for RE, RE methods, and requirement representations. Furthermore, we propose a comprehensive research framework for UGC-based RE and identify key directions for future research. The findings of this study can contribute to the development of more effective and intelligent RE processes, while helping to establish a systematic framework for UGC-based RE research.}, archive = {J_EAAI}, author = {Mengsi Cai and Wenchuan Yang and Yonghao Du and Yuejin Tan and Xin Lu}, doi = {10.1016/j.engappai.2025.111110}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111110}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic requirements elicitation from user-generated content: A review of data, methods, and representations}, volume = {156}, year = {2025}, }
• (2025). Emotion-aware dual cross-attentive neural network with label fusion for stance detection in misinformative social media content. EAAI, 156, 111109. (www)

  @article{ , abstract = {The rapid evolution of social media has generated an overwhelming volume of user-generated content, conveying implicit opinions and contributing to the spread of misinformation. The method aims to enhance the detection of stance where misinformation can polarize user opinions. Stance detection has emerged as a crucial approach to effectively analyze underlying biases in shared information and combating misinformation. This paper proposes a novel method for S tance P rediction through a L abel-fused dual cross- A ttentive E motion-aware neural Net work (SPLAENet) in misinformative social media user-generated content. The proposed method employs a dual cross-attention mechanism and a hierarchical attention network to capture inter and intra-relationships by focusing on the relevant parts of source text in the context of reply text and vice versa. We incorporate emotions to effectively distinguish between different stance categories by leveraging the emotional alignment or divergence between texts. We also employ label fusion that uses distance-metric learning to align extracted features with stance labels, improving the method’s ability to accurately distinguish between stances. Extensive experiments demonstrate the significant improvements achieved by SPLAENet over existing state-of-the-art methods. SPLAENet demonstrates an average gain of 8.92% in accuracy and 17.36% in F1-score on the RumourEval dataset. On the SemEval dataset, it achieves average gains of 7.02% in accuracy and 10.92% in F1-score. On the P-stance dataset, it demonstrates average gains of 10.03% in accuracy and 11.18% in F1-score. These results validate the effectiveness of the proposed method for stance detection in the context of misinformative social media content. Our code is publicly available at: https://github.com/lata04/SPLAENet .}, archive = {J_EAAI}, author = {Lata Pangtey and Mohammad Zia Ur Rehman and Prasad Chaudhari and Shubhi Bansal and Nagendra Kumar}, doi = {10.1016/j.engappai.2025.111109}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111109}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Emotion-aware dual cross-attentive neural network with label fusion for stance detection in misinformative social media content}, volume = {156}, year = {2025}, }
• (2025). Vehicle repacking strategy and enhanced asynchronous advantage actor-critic for multi-objective task scheduling and orchestration in cloud–edge vehicular networks. EAAI, 156, 111108. (www)

  @article{ , abstract = {Vehicular cloud networks (VCNs) enhance task scheduling and improve driving experiences within vehicle networks. VCNs demonstrate greater resource efficiency; however, they experience significant challenges due to prolonged response times referred to as end-to-end delay (EED). Due to the extended duration required by VCNs, vehicular edge networks (VENs) have been introduced as a viable solution for applications necessitating quick responses. VENs face challenges, including vehicle disparities and limitations in computing and storage capacities. Existing heuristic solutions demonstrate inadequate performance and insufficient robustness in addressing these issues. This study presents a network system integrating edge and cloud computing with an enhanced asynchronous advantage actor-critic (A3C) deep reinforcement learning (DRL) methodology. The system organizes and assigns tasks based on available resources and possible task modifications. The architecture relies fundamentally on the vehicle repacking (VR) method, reorganizing vehicle locations within the network to mitigate compute and storage resource overload. The system optimizes resource utilization and equitable job distribution through appropriate car readjustment, reducing traffic and improving performance. Our neural network architecture classifies incoming applications as delay-sensitive or compute-sensitive based on their requirements, thereby improving scheduling efficiency. Simulation results indicate that our solution is effective and outperforms non-DRL and DRL-based approaches. The model demonstrates a reduction in end-to-end delay by 12.04%, an improvement in resource utilization by 49.58%, an enhancement in service quality by 58.76%, a decrease in packet losses by 46.33%, and a reduction in overhead by 71.94%. These results underscore the significant advantages of the VR strategy for VENs’ performance.}, archive = {J_EAAI}, author = {Mustafa Ibrahim Khaleel}, doi = {10.1016/j.engappai.2025.111108}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111108}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Vehicle repacking strategy and enhanced asynchronous advantage actor-critic for multi-objective task scheduling and orchestration in cloud–edge vehicular networks}, volume = {156}, year = {2025}, }
• (2025). A comprehensive review of non-latin natural scene text detection and recognition techniques. EAAI, 156, 111107. (www)

  @article{ , abstract = {This paper provides a comprehensive review of non-Latin scene text recognition (STR), with a focus on deep learning methods. It discusses the challenges presented by non-Latin scripts and how current methodologies address these challenges in both text detection and recognition. We explore traditional and deep learning-based approaches, with particular emphasis on transformer-based models, and analyze language-specific challenges across different scripts, such as Arabic, Chinese, and Indic. The review covers both text detection, where we examine the role of deep learning in identifying text regions, and scene text recognition, highlighting the integration of attention mechanisms and cross-lingual transfer learning. We also propose promising future research directions, including data augmentation, better multi-language adaptability, and the fusion of cross-modal information. Finally, we introduce our recent research on Uyghur scene text recognition as a case study. This paper offers valuable insights into advancing STR technologies and overcoming the unique challenges in non-Latin script recognition.}, archive = {J_EAAI}, author = {Elham Eli and Dong Wang and Wenting Xu and Hornisa Mamat and Alimjan Aysa and Kurban Ubul}, doi = {10.1016/j.engappai.2025.111107}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111107}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A comprehensive review of non-latin natural scene text detection and recognition techniques}, volume = {156}, year = {2025}, }
• (2025). Multilayer neural network analysis of penta-hybrid nanofluid flow with entropy optimization: Application to rocket engine nozzle cooling. EAAI, 156, 111106. (www)

  @article{ , abstract = {Effective thermal management of rocket engine nozzles is crucial for ensuring structural integrity and optimizing performance under extreme heat loads. Traditional coolants often exhibit limitations in heat dissipation and stability. To address this, the present study introduces a novel electro-magnetohydrodynamic Casson-Williamson penta-hybrid nanofluid flow over a porous conical nozzle as an advanced cooling strategy. Silicon Carbide ( S i C ) , magnetite ( F e 3 O 4 ) , Aluminium oxide ( A l 2 O 3 ) , single-walled carbon nanotubes ( S W C N T ) , and multi-walled carbon nanotubes ( M W C N T ) in a liquid sodium base, chosen for their superior thermal conductivity, heat dissipation, and stability, making it highly suitable for aerospace applications. Artificial Neural Networks employs as a predictive tool with the dataset split into 70 % for training, 15 % for testing, and 15 % for validation to ensure robust model evaluation. The governing equations are solved using MATLAB's BVP4C method after using similarity transformations. Artificial neural network, utilizing the scaled conjugate gradient (SCG) algorithm, predicts critical parameters such as skin friction and Nusselt number. Key results reveal that Penta hybrid nanofluid enhances heat transfer by 31.27 % and reduces thermal resistance by 21.38 % compared to traditional nanofluids. The artificial neural network model exhibits high accuracy, with R = 0.97617 for training and R = 0.98016 for testing (skin friction prediction), and R = 0.99823 for training and R = 0.98916 for validation (Nusselt number prediction), achieving a mean squared error (MSE) of 0.00007301 at epoch 61. These findings confirm that Penta hybrid nanofluid provides superior thermal performance and reliability, making it a promising candidate for next-generation aerospace propulsion systems.}, archive = {J_EAAI}, author = {Chandrakanta Parida and Satya Prakash Ratha and Ganeswar Mahanta and Smita Parija and Sachin Shaw}, doi = {10.1016/j.engappai.2025.111106}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111106}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multilayer neural network analysis of penta-hybrid nanofluid flow with entropy optimization: Application to rocket engine nozzle cooling}, volume = {156}, year = {2025}, }
• (2025). A physical-modulated framework for process optimization and shape inference of industrial metal tube. EAAI, 156, 111105. (www)

  @article{ , abstract = {Accurate forming shape prediction and process optimization are crucial for ensuring the quality of tubular components throughout both the design and iteration phases. However, the nonlinear multi-physics coupling between plastic deformation and process attributes presents significant complexity. Although the two tasks are inherently interdependent, they are often treated as separate paradigms in industrial applications. This lack of a synergistic approach impedes the establishment of an efficient closed-loop manufacturing process. We propose a physical-modulated dual-branch prediction framework, called Forming Process to Three-Dimension (FP-3D). It operates under a unified feature scale, which interactively maps from the process attributes to the three-dimensional (3D) tube shape. It bridges branches by extracting structurally embedded geometric latent features as a reliable intermediate representation. The process optimization branch contrasts shape features to learn latent disparity in pairs. It alleviates limitations posed by sample quantity and encourages the model to learn process attribute adjustments as a historically measured shape deformed to target one. A physical-increment-modulated (PIM) layer is proposed to facilitate the accelerated learning of physical increments that are sensitive to process attributes. In the shape inference branch, we propose radially transferring features toward the implicit skeleton, which enables physical information to intervene in the latent space for controllable shape generation. Under the sole condition of process parameters, FP-3D allows the conditional generation of point-wise features to decode refined 3D shapes. The extensive experiments conducted on diverse tube and benchmark datasets demonstrate that FP-3D exhibits state-of-the-art performance.}, archive = {J_EAAI}, author = {Le Wang and Zili Wang and Shuyou Zhang and Jianrong Tan and Yaochen Lin and Yongzhe Xiang}, doi = {10.1016/j.engappai.2025.111105}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111105}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A physical-modulated framework for process optimization and shape inference of industrial metal tube}, volume = {156}, year = {2025}, }
• (2025). Enhancing online industrial quality index prediction with a general deep temporal feature extraction and incremental ensemble modeling framework. EAAI, 156, 111104. (www)

  @article{ , abstract = {Data-driven modeling methods for industrial quality index prediction often face the challenge of limited data representation. Using process variable snapshots from a single time step is insufficient for building high-performance soft sensors. Moreover, during online prediction, the performance of soft sensors is affected by diverse operating conditions and concept drift in industrial data streams, leading to performance degradation. To address these challenges, this paper proposes a Temporal Feature Extraction and Incremental Variational Bayesian Regression Ensemble (TFE-IVBRE) framework, which provides a general solution for various online industrial quality index prediction tasks. The TFE-IVBRE framework combines the temporal feature extraction capabilities of deep neural networks, the ability of ensemble learning to handle diverse operating conditions, and the online learning features of variational Bayesian models. An incremental update strategy is also incorporated to maintain consistent prediction performance. Experiments in the Debutanizer Column and Sulfur Recovery Unit scenarios show that the prediction performance of TFE-IVBRE significantly outperforms other static and online comparison models, with the effectiveness of its components validated through ablation studies. Finally, the overall model also demonstrates good robustness. These results offer valuable insights for the advancement of industrial soft sensor development.}, archive = {J_EAAI}, author = {Yuchen Zhao and Rui Yang and Jie Ying and Song Wang and Zhenyu Zeng}, doi = {10.1016/j.engappai.2025.111104}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111104}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing online industrial quality index prediction with a general deep temporal feature extraction and incremental ensemble modeling framework}, volume = {156}, year = {2025}, }
• (2025). Fully cooperative domain adaptive neural network for defect classification in printed circuit boards. EAAI, 156, 111103. (www)

  @article{ , abstract = {The rapid development of deep learning algorithms has enabled effective identification of real and pseudo defects in Printed Circuit Boards (PCBs) when sufficient annotated data are available. However, many types of PCBs typically contain unique defect types and circuit specifications and often lack sufficient training datasets for defect detection. Using them to train models with high-precision classification capabilities remains a challenge. In response to these challenges, this study proposes Fully Cooperative Domain Adaptive Neural Network (FC-DANN), aimed at utilizing similar principles for defect recognition on circuit boards of different styles. The FC-DANN model integrates a cyclic feature extractor with a transformation unit and a collaborative discriminator with non-adversarial features. At the same time, a co-quotient classifier composed of multiple single-domain label classifiers is introduced. This method can effectively achieve feature alignment of data with different styles for high-precision, small sample target domain classification. This study use datasets for four PCB styles as well as the public dataset, the efficacy of the FC-DANN model was confirmed. The proposed method has been proven to be effective, with average classification accuracy exceeding 1.06% and 1.88% of existing technologies on both datasets, respectively.}, archive = {J_EAAI}, author = {Jiafu Wen and Yuebin Wu and Wenkai Huang and Kunbo Han and Bingjun Luo}, doi = {10.1016/j.engappai.2025.111103}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111103}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fully cooperative domain adaptive neural network for defect classification in printed circuit boards}, volume = {156}, year = {2025}, }
• (2025). Joint image synthesis and fusion with converted features for alzheimer’s disease diagnosis. EAAI, 156, 111102. (www)

  @article{ , abstract = {The effectiveness of complete multi-modal neuroimaging data in the diagnosis of Alzheimer’s disease has been extensively demonstrated and applied. Dealing with incomplete modalities poses a common challenge in multi-modal neuroimaging diagnosis. The mainstream approaches aim to synthesize missing neuroimaging data in order to make full use of all available samples. However, these methods treat image synthesis and disease diagnosis as two independent tasks, overlooking the potential feature of cross-modality image synthesis for downstream tasks. To this end, we propose the Joint Image Synthesis and Classification Learning method to jointly optimize image synthesis and disease diagnosis using incomplete neuroimaging modalities. Our approach comprises a submodule for synthesizing missing neuroimaging data and a decision fusion submodule that integrates features from different modalities and the high-level/converted features generated during synthesis. Experimental results demonstrate that our joint optimization approach outperforms conventional two-stage methods. Our method is capable of handling arbitrary neuroimaging modality missing scenarios and achieves state-of-the-art performance in both Alzheimer’s Disease identification and mild cognitive impairment conversion classification tasks. Finally, we further explored the importance of different converted features. This highlights the effectiveness of our approach in addressing the challenges of Alzheimer’s Disease diagnosis and provides insights for future research in multi-modal medical image analysis.}, archive = {J_EAAI}, author = {Zhaodong Chen and Mingxia Wang and Fengtao Nan and Yun Yang and Shunbao Li and Menghui Zhou and Jun Qi and Hanwen Wang and Po Yang}, doi = {10.1016/j.engappai.2025.111102}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111102}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Joint image synthesis and fusion with converted features for alzheimer’s disease diagnosis}, volume = {156}, year = {2025}, }
• (2025). Weakly supervised histopathology tissue semantic segmentation with multi-scale voting and online noise suppression. EAAI, 156, 111100. (www)

  @article{ , abstract = {The development of an Artificial Intelligence (AI) assisted tissue segmentation method of digital pathology images is critical for cancer diagnosis and prognosis. Excellent performance has been achieved with the current fully supervised segmentation approach, which relies on a huge number of annotated data. However, drawing dense pixel-level annotations on the giga-pixel whole slide image (WSI) is extremely time-consuming and labor-intensive. To this end, we propose a tissue segmentation method using only patch-level classification labels to reduce such annotation burden and significantly improve the quality of the pseudo-masks. We introduce a framework with two phases of classification and segmentation. In the classification phase, we propose a multi-scale voting method on the Class Activation Map (CAM) based model to obtain more stable pseudo masks. In the segmentation phase, an Online Noise Suppression Strategy (ONSS) is proposed to encourage the model to focus on more reliable signals in the pseudo mask rather than noisy signals. Extensive experiments on two weakly supervised pathology image tissue segmentation datasets Lung Adenocarcinoma (LUAD-HistoSeg) and Breast Cancer Semantic Segmentation (BCSS-WSSS) demonstrate our model outperforms state-of-the-art weakly-supervised semantic segmentation (WSSS) methods using patch-level labels. Furthermore, our method exhibits superior generalization ability compared to other models, and demonstrates promising adaptation performance on unseen domains with only small amounts of data.}, archive = {J_EAAI}, author = {Xipeng Pan and Hualong Zhang and Huahu Deng and Huadeng Wang and Lingqiao Li and Zhenbing Liu and Lin Wang and Yajun An and Cheng Lu and Zaiyi Liu and Chu Han and Rushi Lan}, doi = {10.1016/j.engappai.2025.111100}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111100}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weakly supervised histopathology tissue semantic segmentation with multi-scale voting and online noise suppression}, volume = {156}, year = {2025}, }
• (2025). A novel triplet loss architecture with visual explanation for detecting the unwanted rotation of bolts in safety-critical environments. EAAI, 156, 111097. (www)

  @article{ , abstract = {In the commonly used method of bolting to secure parts of equipment and structure, the bolts must be tightened to an adequate preload force. Failure to do so could affect the integrity of the structure, as well as the efficient running of the site and, crucially, employees’ safety. In this project, we consider the use of artificial intelligence (AI) techniques to analyse maintenance videos and identify the unwanted loosening of bolts over time in order that they might be used as additional tools in a continuous maintenance plan. We found that accuracy levels of up to 97% could be achieved in identifying bolt rotation with our proposed machine learning-based triplet loss architecture. The use of gradient-weighted class activation mapping (Grad-CAM) visualisations to identify areas of the image where change had occurred enabled us to test how robust our model was to noise in the data. This explanation may assist users in safety-critical environments guiding them to the problem, and helping mitigate the black-box nature of machine learning algorithms. Whilst the accuracy of the models varies depending on the rotational angle of the bolt, we clearly demonstrate that triplet loss is a good basis for performing change detection in industrial settings. Furthermore, Grad-CAM has shown to be a useful technique to help a user understand the decisions made by the network and allow them to see where unwanted rotation has occurred.}, archive = {J_EAAI}, author = {Tom Bolton and Julian Bass and Tarek Gaber and Taha Mansouri and Peter Adam and Hossein Ghavimi}, doi = {10.1016/j.engappai.2025.111097}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111097}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel triplet loss architecture with visual explanation for detecting the unwanted rotation of bolts in safety-critical environments}, volume = {156}, year = {2025}, }
• (2025). Enhancing the effectiveness of emergency department computed tomography scans using pre-trained language models. EAAI, 156, 111094. (www)

  @article{ , abstract = {Objective This study aims to develop a predictive model using a medical clinical assistive large language model to determine the necessity of computed tomography (CT) scans in emergency department settings based solely on data collected at triage. The model seeks to improve patient flow and more efficiently allocate limited medical resources while reducing unnecessary radiation exposure. Methods The model uses data collected from emergency department triage and includes patient symptoms, chief complaints, vital signs and medical history, without the need for physiological test data. Results This study analyzed 165,391 emergency department records from Shuang Ho Hospital of Taipei Medical University and used a large language model to develop a model for predicting whether a patient should undergo a CT scan. While initial results indicate that detailed symptom descriptions and severity of pain assessments can enhance prediction accuracy, our approach centers on data preprocessing, the integration of unstructured data, and external features. In our final performance comparison, the model developed using a large language model exhibited the best performance, achieving an area under the receiver operating characteristic curve (AUROC) of 0.88 and an area under the precision-recall curve (AUPRC) of 0.5414. This represents a 0.6 % improvement over existing language models and a 4.8 % improvement over traditional machine learning approaches. It is notable that the model achieved a high negative predictive value of 0.9261, indicating strong reliability in identifying patients who don't require CT scans. This model will allow physicians to better understand the overall health status of patients and provide earlier diagnostic and treatment recommendations based on comprehensive model information, which will ultimately lead to better patient care. Conclusion This research establishes foundational work for future studies that aim at optimizing emergency diagnostic processes and enhancing patient care through improved medical predictions. However, expanding the dataset’s diversity and pursuing external validations are essential to improve the predictive accuracy and applicability of the model in a variety of emergency department settings.}, archive = {J_EAAI}, author = {Heng-Yu Lin and Yung-Chun Chang and Pei-Ying Yang and Ting-Yun Huang}, doi = {10.1016/j.engappai.2025.111094}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111094}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing the effectiveness of emergency department computed tomography scans using pre-trained language models}, volume = {156}, year = {2025}, }
• (2025). Backtracing byzantine attacks in distributed average consensus networks: A gated graph neural network approach with graph reconstruction. EAAI, 156, 111093. (www)

  @article{ , abstract = {Byzantine attackers disguise themselves as normal nodes yet propagate incorrect information in the network to disrupt the function of distributed systems. Current Byzantine detecting approaches commonly rely on the delivered data among agents, where an agent is supposed to be a Byzantine attacker if its delivered data significantly differs with others. In this paper, for the first time we suppose that the transmission data within the network is not known due to privacy consideration and propose the problem of backtracing Byzantine attackers through observed states. To address this problem, we adopt the idea of backtracing patient in disease spreading networks and propose a gated graph neural network with graph reconstruction to effectively localize the Byzantine attackers in multi-agent networks. This approach first develops a data-driven method to learn the networking structure of the multi-agent systems and then builds a gated graph neural network based on the reconstructed graph to classify the nodes into normal ones and Byzantine ones. Compared with previous Byzantine detecting methods, the proposed approach is fully driven by the measured output state of each agent and localizes the attacker from external perspective. Extensive experimental results are conducted to confirm that the proposed Byzantine attack detector performs well across various testing parameters.}, archive = {J_EAAI}, author = {Xinliang Wang and Shaolin Tan and Ye Tao and Nian Liu and Bing Li and Suixiang Gao}, doi = {10.1016/j.engappai.2025.111093}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111093}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Backtracing byzantine attacks in distributed average consensus networks: A gated graph neural network approach with graph reconstruction}, volume = {156}, year = {2025}, }
• (2025). Machine learning-based failure prediction in united states of america lobbying firms: The role of director networks. EAAI, 156, 111091. (www)

  @article{ , abstract = {The role of directors and their associated social networks is critical in determining corporate failure, particularly within lobbying firms in the United States. However, predicting corporate failure in increasingly complex organizational structures remains an underexplored challenge. Using firm-level director data from the United States between 2005 and 2018, we apply both traditional machine learning models – Logistic Regression, Random Forest, and Support Vector Machines (SVM) – and more recent tabular deep learning approaches, including TabTransformer and Feature Tokenizer Transformer, to predict and uncover complex non-linear relationships between director network attributes and firm failure. Our results demonstrate that corporate failure can be accurately predicted based on director network characteristics, achieving a mean performance of 99.98% Area Under the Receiver Operating Characteristic Curve (AUC-ROC), 85.10% Precision, and 80.23% Recall using a stratified 5-fold cross-validation procedure with a weighted averaging strategy. Furthermore, network-derived attributes such as centrality, betweenness, and remuneration significantly influence failure risk in lobbying firms. These findings contribute to the management literature by highlighting the predictive value of director networks in corporate failure and offer practical insights for directors, engineering practitioners, and policymakers seeking to mitigate organizational risk.}, archive = {J_EAAI}, author = {Dustin van der Haar and Suman Lodh and Monomita Nandy}, doi = {10.1016/j.engappai.2025.111091}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111091}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-based failure prediction in united states of america lobbying firms: The role of director networks}, volume = {156}, year = {2025}, }
• (2025). Fast shallow multi-subnet detector for real-time object detection. EAAI, 156, 111090. (www)

  @article{ , abstract = {Real-time object detection algorithms, underpinned by Deep Neural Networks (DNNs), are extensively applied in fields like autonomous driving and security surveillance. However, current algorithms face issues of low hardware resource utilization and high synchronization delays between network layers when deployed on DNN hardware accelerators, adversely affecting overall performance and efficiency. To address these issues, we have proposed an innovative single-stage object detection framework, the Shallow Multi-Subnet Detector (SMS-Det). SMS-Det adopts a multi-parallel-shallow-subnet architecture, which reduces inter-layer synchronization latency by decreasing network depth. Furthermore, it fully utilizes DNN hardware accelerators by executing convolution operations in parallel, preventing resource underutilization and maximizing throughput. The proposed network is comprised of multiple parallel shallow subnets, each of which processes feature maps of different scales. The Feature Fusion Layer (FFL) ensures seamless information exchange across subnets, significantly improving the detection of small and occluded objects. Finally, we introduce the multi-scale channel attention projections to enhance the feature mapping between the teacher model and the student model in the training process. Experimental results on the Microsoft Common Objects in Context (MS COCO) dataset demonstrate that our model achieves a state-of-the-art mean Average Precision (mAP) of 42.6%, surpassing You Only Look Once Version 5 Small (YOLOv5-S 37.4%) with only 19.4 Giga Floating Point Operations (GFLOPs) and 11.0 million parameters. Our model obtains 156 Frames Per Second (FPS), achieving a real-time inference acceleration of 51.4% compared to YOLOv5-S (103 FPS).}, archive = {J_EAAI}, author = {Yuan Li and Mengdie Song and Ke Hu and Song Chen and Yi Kang}, doi = {10.1016/j.engappai.2025.111090}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111090}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast shallow multi-subnet detector for real-time object detection}, volume = {156}, year = {2025}, }
• (2025). Training-free multi-scale neural architecture search for high-incidence cancer prediction. EAAI, 156, 111089. (www)

  @article{ , abstract = {Deep neural networks excel in high-incidence cancer prediction; however, designing networks that predict specific cancers is time-consuming and requires expert. The neural architecture search method offers a way to automate network design and has shown success in natural image. However, the small and varying lesion sizes in cancer image pose challenges, and most neural architecture search methods are computationally expensive and exhibit low agent correlation. Therefore, we propose a training-free multi-scale neural architecture search method for high-incidence cancer prediction. We introduce a multi-scale search space to address varying lesion sizes; and identify optimal scale combinations for feature extraction. To reduce computational costs and improve agent correlation, we design a training-free agent that evaluates network performance based on convergence, expressiveness, trainability, and complexity, enabling efficient neural architecture search implementation. Our extensive experiments on the NAS-Bench-201, MedmnistV2, LC25000, BreakHis, and CRC-5000 datasets show that our method outperforms both manually designed networks and state-of-the-art neural architecture search methods. The results demonstrate average improvements of 4.2%, 1.88%, 79.45%, 34.31%, and 31.71% in accuracy, area under the curve, search time, and Kendall and Spearman correlation coefficients, respectively.}, archive = {J_EAAI}, author = {Jie Zheng and Chunlin He and Wenxing Man and Jing Wang}, doi = {10.1016/j.engappai.2025.111089}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111089}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Training-free multi-scale neural architecture search for high-incidence cancer prediction}, volume = {156}, year = {2025}, }
• (2025). A novel neural network method for solid mechanics problems of heterogeneous bars. EAAI, 156, 111084. (www)

  @article{ , abstract = {This paper introduces a novel neural network (NN) method using Physics-Informed Neural Networks (PINNs) to solve boundary value problems in solid mechanics for heterogeneous materials. Existing domain decomposition methods in PINNs require multiple NNs and indiscriminate data sampling, often with heuristically chosen NN architectures. In contrast, this work proposes a domain variation strategy that solves heterogeneous problems using a single-hidden-layered NN. The strategy uses boundary, interface, and collocation spatial coordinates from all sub-domains to generate variations of the heterogeneous solid, which facilitates the establishment of mechanical connections within the NN. From each variation, a unique dataset is generated and fed into an instance of NN, referred to as sub-NN, which output displacements and their gradients. Outputs from all sub-NNs are utilized to compute boundary losses (via boundary conditions), interface losses (via continuity conditions), and governing equation losses, which are minimized using Adam and Limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) optimization methods. The number of data samples and optimal neurons for convergence are determined through theoretical studies. The method’s effectiveness is demonstrated on a heterogeneous bar fixed at one end and subjected to a load at the other. Validation against Finite Element Method (FEM) simulations shows mean squared error (MSE) of 0.04, with a mean absolute percentage error (MAPE) of 4.64 % for the 20-material bar problem involving force variation. The NN model effectively captures displacements, stresses, and strain discontinuities at the material interfaces. This approach offers an efficient alternative to domain decomposition methods and a framework for solving multiple problems via surrogate modeling.}, archive = {J_EAAI}, author = {Mitra Ramakrishna Lanka and Gui-Rong Liu}, doi = {10.1016/j.engappai.2025.111084}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111084}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel neural network method for solid mechanics problems of heterogeneous bars}, volume = {156}, year = {2025}, }
• (2025). Graph neural network-based deep reinforcement learning algorithm for virtual network function forwarding graph embedding in Space–Air–Ground integrated network. EAAI, 156, 111083. (www)

  @article{ , abstract = {Space–Air–Ground Integrated Network (SAGIN) can provide seamless three-dimensional coverage and enhanced flexibility, and it has been recognized as the core architecture of future 6G networks. Software Defined Network (SDN) and Network Function Virtualization (NFV) are two enabling technologies of SAGIN, which can sequentially arrange multiple Virtual Network Functions (VNFs) into VNF Forwarding Graph (VNF-FG) to provide users with scalable and parallel network services. However, SAGIN exhibits significant dynamism and heterogeneity, and VNFs may be deployed in multiple different heterogeneous locations, which brings great challenges to the efficient embedding of VNF-FGs required for user service request flows. In this paper, we study the VNF-FG embedding problem by jointly considering the structural constraints, node and link resource constraints, and End-to-End (E2E) delay constraint of VNF-FG in SDN/NFV-enabled SAGIN. Specifically, we first design a three-layer SDN/NFV-enabled SAGIN architecture consisting of a global controller and distributed intra-domain SDN controllers. Then, we define the Delay and Cost Efficient Dynamic VNF-FG Embedding Problem (DCE-DVEP) and formulate it as an Integer Linear Programming (ILP) with the objective of minimizing the weighted sum of E2E delay and embedding cost of all VNF-FGs. Finally, a Graph Neural Network-based Deep Reinforcement learning Embedding (GNN-DRE) algorithm is proposed to solve the DCE-DVEP, which can more accurately capture the rich feature information from both SAGIN and VNF-FG by specifically integrating different GNN models and adopts Deep Reinforcement Learning (DRL) to make effective embedding decisions. The simulation results demonstrate that, compared with other baseline algorithms, the GNN-DRE can reduce E2E delay and embedding cost by about 8% and 13%, respectively.}, archive = {J_EAAI}, author = {Liang Liu and Tengxiang Jing and Siyuan Tan and Yujie Zhang and Yejun He and Chuan Xu}, doi = {10.1016/j.engappai.2025.111083}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111083}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph neural network-based deep reinforcement learning algorithm for virtual network function forwarding graph embedding in Space–Air–Ground integrated network}, volume = {156}, year = {2025}, }
• (2025). Self-similar spectral reasoning network for efficient anti-aliasing seismic data reconstruction. EAAI, 156, 111081. (www)

  @article{ , abstract = {The reconstruction of seismic data is an essential and challenging task in seismic data processing. To solve this problem, researchers have developed different deep learning-based algorithms and made great progress. However, most unsupervised methods based on deep learning cannot effectively solve the spectral aliasing problem of missing seismic data. In this work, we propose an unsupervised method for anti-aliasing seismic data reconstruction according to their self-similarity. The innovation of this method is to explore two self-similarity properties of seismic images, the similarity of spectral features between different localities of seismic images and the similarity of spectral features of the image at different resolutions. Based on these two self-similarity properties, we build a self-similar spectral reasoning network to reconstruct the regular missing seismic data. The self-similarity properties can be regarded as an expression of the high correlation of seismic data. Our method extracts the self-similar components from the observed data in the spectral domain and uses them to fill in the missing parts. The superior performance of the proposed method is verified in challenging realistic scenarios. Experiments show that the proposed method has better anti-aliasing capability than other previously proposed methods. At the end of the article, we discuss the robustness of the self-similar spectral reasoning network.}, archive = {J_EAAI}, author = {Xin Geng and Changpeng Wang and Aoqi Song and Chunjia Zhang and Jiangshe Zhang and Zhiliang Zhou}, doi = {10.1016/j.engappai.2025.111081}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111081}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-similar spectral reasoning network for efficient anti-aliasing seismic data reconstruction}, volume = {156}, year = {2025}, }
• (2025). A lightweight guided and shared feature fusion model for aircraft skin defect detection. EAAI, 156, 111078. (www)

  @article{ , abstract = {The detection of aircraft skin defects is critical for aviation safety. Traditional manual inspections are inefficient and error-prone, while existing automated models struggle to identify small defects and operate efficiently in resource-constrained environments. We propose the Lightweight Guided and Shared Feature Fusion Network (LGSF-Net), integrating StarNet-based feature extraction, Adaptive Guided Feature Module (AGFM), and Shared Feature Detection (SFD) head. LGSF-Net achieves superior accuracy for small and medium-sized defects at just 3.0 megabyte (MB) and 139 frames per second (FPS), enabling real-time deployment on mobile platforms including drones. Our experiments show LGSF-Net significantly outperforms state-of-the-art models including You Only Look Once version 10 (YOLOv10), Real-Time Detection Transformer (RT-DETR), and You Only Look Once version 11 (YOLO11) while requiring 42 % fewer parameters and 35 % lower computational resources. LGSF-Net achieves 93.7 % average precision (AP) on our Aircraft Skin Defect Dataset and demonstrates strong cross-domain generalization on Steel Surface Defect with 85.7 % AP, Printed Circuit Board with 95.2 % AP, and Vision Drone 2019 with 39.6 % AP. This versatility makes LGSF-Net a transformative solution for aviation maintenance, with applications for automated drone-based inspection systems and potential to significantly reduce aircraft downtime, maintenance costs, and human error. Despite minor limitations under extreme conditions, LGSF-Net bridges the critical gap between detection accuracy and resource efficiency in safety-critical inspection workflows.}, archive = {J_EAAI}, author = {Li Ji and ChaoHang Huang and Haiwei Li and Wenjie Han and JiaYu Fan}, doi = {10.1016/j.engappai.2025.111078}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111078}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight guided and shared feature fusion model for aircraft skin defect detection}, volume = {156}, year = {2025}, }
• (2025). CTRL-F: Pairing convolution with transformer for image classification via multi-level feature cross-attention and representation learning fusion. EAAI, 156, 111076. (www)

  @article{ , abstract = {Transformers have captured growing attention in computer vision, thanks to its large capacity and global processing capabilities. However, transformers are data hungry, and their ability to generalize is constrained compared to Convolutional Neural Networks (ConvNets), especially when trained with limited data due to the absence of the built-in spatial inductive biases present in ConvNets. In this paper, we strive to optimally combine the strengths of both convolution and transformers for image classification tasks. Towards this end, we present a novel lightweight hybrid network that pairs C onvolution with T ransformers via R epresentation L earning F usion and Multi-Level Feature Cross-Attention named CTRL-F. Our network comprises a convolution branch and a novel transformer module named multi-level feature cross-attention (MFCA). The MFCA module operates on multi-level feature representations obtained at different convolution stages. It processes small patch tokens and large patch tokens extracted from these multi-level feature representations via two separate transformer branches, where both branches exchange knowledge through cross-attention mechanism. We fuse the local responses acquired from the convolution path with the global responses acquired from the MFCA module using novel representation fusion techniques dubbed adaptive knowledge fusion (AKF) and collaborative knowledge fusion (CKF). Experiments demonstrate that our CTRL-F variants achieve state-of-the-art performance, whether trained from scratch on large data or even with low-data regime. For Instance, CTRL-F achieves top-1 accuracy of 84.01 %, 61.68 % and 99.91 % when trained from scratch on Oxford-102 Flowers, CUB-200 and PlantVillage datasets respectively, surpassing state-of-the-art models which showcase the robustness of our model on image classification tasks. Code at: https://github.com/hosamsherif/CTRL-F .}, archive = {J_EAAI}, author = {Hosam S. EL-Assiouti and Hadeer El-Saadawy and Maryam N. Al-Berry and Mohamed F. Tolba}, doi = {10.1016/j.engappai.2025.111076}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111076}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {CTRL-F: Pairing convolution with transformer for image classification via multi-level feature cross-attention and representation learning fusion}, volume = {156}, year = {2025}, }
• (2025). Confidence-aware iterative training for cross-lingual entity alignment. EAAI, 156, 111069. (www)

  @article{ , abstract = {Entity alignment (EA) aims to identify equivalent entities across knowledge graphs, serving as a critical step in integrating multi-source knowledge graphs. In recent years, EA methods relying on alignment seeds have achieved impressive performance. However, the high cost of manually labeled alignment seeds has posed a significant limitation to their practical applications in the real world. Most methods adopt iterative strategies to generate pseudo-alignment seeds automatically. However, unreliable iterative strategies introduce a large number of noisy pseudo-alignment seeds, leading to low-quality entity embeddings. In this paper, we propose a Confidence-Aware Iterative Training (CAIT) framework for unsupervised EA tasks. The framework initially extracts semantic and structural features of entities from knowledge graphs, then generates pseudo-alignment seeds by the confidence-aware iterative strategy, which limits the quantity of noisy pseudo-alignment seeds and progressively enhances the quality of entity embeddings. Extensive experiments conducted on widely used benchmark datasets demonstrate that CAIT outperforms existing state-of-the-art methods in cross-lingual EA tasks, both with and without prior alignment seeds.}, archive = {J_EAAI}, author = {Fan Ye and Yu Zeng and Zhangling Duan and Zhaolong Ling and Yun Yang}, doi = {10.1016/j.engappai.2025.111069}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111069}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Confidence-aware iterative training for cross-lingual entity alignment}, volume = {156}, year = {2025}, }
• (2025). Seismic denoising diffusion restoration model for seismic data processing. EAAI, 156, 111068. (www)

  @article{ , abstract = {Interpolation and denoising are critical preprocessing steps for seismic data. Particularly for diverse tasks such as random discrete missing, continuous missing, and noise, while deep learning-based methods have been extensively applied to the fields of seismic data interpolation and denoising, current research lacks effective multitasking approaches. In this paper, We have extended Denoising Diffusion Restoration Model (DDRM) into the domain of seismic data processing, introducing the Seismic Denoising Diffusion Restoration Model (SeisDDRM). Built upon a pre-trained diffusion model, the SeisDDRM constructs a joint distribution encompassing data, measurements, and linear degradation operators tailored for seismic data interpolation and denoising. Posterior sampling is employed to facilitate both interpolation and denoising. The method accommodates various missing scenarios, including random discrete missing and continuous missing, and performs seismic data denoising without the need for retraining. Furthermore, we have refined the nonlinear activation-free network (NAFNet) as the noise prediction network, integrating a time embedding module to introduce time information for guiding the training of the diffusion model. This design not only achieves performance comparable to Unet but also reduces intra-block complexity. Numerous experiments on synthetic and field datasets validate the effectiveness of our model, and ablation studies confirm the necessity of key components.}, archive = {J_EAAI}, author = {Kewen Li and Jian Ma and Yimin Dou and Yingzhi Zhao and Zhixuan Yang and Zhaohui Li}, doi = {10.1016/j.engappai.2025.111068}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111068}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Seismic denoising diffusion restoration model for seismic data processing}, volume = {156}, year = {2025}, }
• (2025). A hybrid architecture based on structured state space sequence model and convolutional neural network for real-time object detection. EAAI, 156, 111067. (www)

  @article{ , abstract = {Real-time performance is essential for practical deployment of object detection on edge devices, where high processing speed and low latency are paramount. This paper introduces a novel approach aimed at boosting real-time object detection while strictly adhering to computational constraints. A structured state space sequence model, Mamba, is strategically embedded in the early stages of the backbone network to capture long-range dependencies, thereby enhancing the model’s representation capability. Given the limitations of Mamba in directional perception, a lightweight spatial attention mechanism is introduced to integrate global context into each spatial location. Additionally, a computationally efficient module inspired by the Ghost module is developed to reduce resource demands. This dual-strategy approach optimizes both performance and efficiency in real-time object detection. Extensive experiments demonstrate the superiority of this proposed approach; on the Microsoft Common Objects in Context (MS COCO) dataset, it achieves a +1.6 AP (Average Precision) improvement over state-of-the-art methods, reaching 41.1 AP with minimal added model complexity on the nano scale. The effectiveness and efficiency of each component are further substantiated through ablation studies on the Pascal Visual Object Classes (Pascal VOC dataset). To verify the universality of the proposed method, this study selects underwater object detection, characterized by an extremely complex background environment, as the other validation scenario. Through the application of this proposed approach to underwater object detection, a state-of-the-art result of 69.5 AP was obtained on the Detecting Underwater Objects (DUO) dataset, exceeding that of You Only Look Once Detector version 11 (YOLO11) by +0.3 AP. Code: https://github.com/chenjie04/Hybrid-YOLO .}, archive = {J_EAAI}, author = {Jie Chen and Meng Joo Er}, doi = {10.1016/j.engappai.2025.111067}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111067}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid architecture based on structured state space sequence model and convolutional neural network for real-time object detection}, volume = {156}, year = {2025}, }
• (2025). Chemically-guided extreme gradient boosting models for predicting the elastic modulus of alkali-activated concrete: Insights into base learner variants. EAAI, 156, 111066. (www)

  @article{ , abstract = {The widespread use of ordinary Portland cement (OPC) significantly contributes to carbon dioxide (CO 2 ) emissions, posing a major threat to the environment and accelerating climate change. Alkali-activated concrete (AAC) has emerged as a sustainable alternative to conventional OPC-based concrete, offering significant environmental benefits while providing desirable mechanical properties for structural applications. This study investigates the application of extreme gradient boosting regression (XGBR) algorithms to model the static modulus of elasticity (SME) of AAC, addressing the complexity of predicting SME by exploring the impact of different base learners on XGBR performance. Advancing beyond conventional approaches, this work integrates alternative base learners-namely artificial neural networks (ANN) and support vector regressors (SVR)- within the XGBR framework and systematically compares their performance with the widely adopted decision tree (DT) base learner. The models were developed and validated on a comprehensive database of AAC mixes, encompassing diverse material compositions and reactivities, mixture designs, and curing conditions. Results of Wilcoxon signed-rank test and the paired t -test showed that while the DT-based XGBR achieved the lowest error metrics during development, the novel SVR-based XGBR exhibited comparable generalisation performance on unseen data, outperforming the ANN-based XGBR. Interpretability analyses, including SHapley Additive exPlanations (SHAP) and permutation feature importance, identified compressive strength, total water content, and sodium oxide (Na 2 O) content of the activator as key predictors of SME. Several interactions among factors influencing the system's chemistry and polymerisation reactions were revealed through partial dependence analysis to investigate the optimal chemical conditions. The findings underscore the efficacy of XGBR ensembles for modelling complex AAC properties, providing insights for optimising sustainable material design to meet the urgent need for environmentally responsible construction solutions.}, archive = {J_EAAI}, author = {Seyed Ali Eftekhar Afzali and Emad Golafshani and Tuan Ngo}, doi = {10.1016/j.engappai.2025.111066}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111066}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Chemically-guided extreme gradient boosting models for predicting the elastic modulus of alkali-activated concrete: Insights into base learner variants}, volume = {156}, year = {2025}, }
• (2025). Novel situation assessment method for amphibious aircraft maritime rescue using probabilistic linguistic hybrid cloud model and best-worst method. EAAI, 156, 111065. (www)

  @article{ , abstract = {Maritime rescue aims to safeguard individuals and assets amidst maritime emergencies, wherein amphibious aircraft play a proactive role by providing reactive and dependable support for rescue operations. The uncertain maritime emergency environment poses significant challenges to amphibious aircraft maritime rescue (AAMR), necessitating an urgent need for a systematic situation assessment to capture and address the threats and risks involved. Current assessment methods, however, suffer from critical deficiencies in addressing multiple uncertainties in situation information and measuring the impacts of interacting threat factors in the maritime emergency environment. To fill these gaps, this paper proposes a novel situation assessment method using probability linguistic hybrid cloud (PLHC) model and best-worst method (BWM) to delineate the optimal situation level of AAMR. Initially, the situation assessment model for AAMR is developed where the threatening factors are identified through literature reviews and empirical analysis. Then, to facilitate reasonable knowledge utilization, the PLHC model, which combines probability linguistic term sets (PLTSs) and hybrid normal and trapezium clouds, is introduced to address experts’ assessment with various uncertainties such as hesitation, and fuzziness. Moreover, the enhanced BWM method is extended to determine the weights of the threatening factors and their mutual interactions. Finally, validated through vessel-sinking emergency scenarios, the proposed method can offer significant situation information for pilots and maritime authorities to facilitate more effective maritime rescue strategies, highlighting its vital role in enhancing maritime emergency response capabilities.}, archive = {J_EAAI}, author = {Xudong Zhu and An Zhang and Wenhao Bi and Zhanjun Huang}, doi = {10.1016/j.engappai.2025.111065}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111065}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Novel situation assessment method for amphibious aircraft maritime rescue using probabilistic linguistic hybrid cloud model and best-worst method}, volume = {156}, year = {2025}, }
• (2025). Mamba-based hybrid network for multi-modality feature fusion. EAAI, 156, 111064. (www)

  @article{ , abstract = {Multi-modality image fusion aims to extract the private information of different modalities and the shared information between modalities in order to fuse them into information-rich images. In order to efficiently achieve remote cross-modality information interaction and modality-specific information fusion tasks, we propose a Mamba-based Hybrid Network. First, the model adopts a two-stage training method to reconstruct, extract and fuse features. Then, We refine the Mamba branch and Convolutional Neural Network branch, then losslessly transmit multi-source image features via Invertible Neural Network Module, and enhance the local feature representation using Convolutional Block Attention Module. For fine-grained spatial extraction, Edge Enhancement Module is utilized to enrich the fused image contour information. Finally, the fused image containing long-range modeling features and local texture features is output. Extensive experiments have shown that the model achieves good results on infrared and visible image fusion as well as medical image fusion tasks on a total of six test sets with nine evaluation metrics. On infrared and visible image datasets, the model outperforms the second-best values by 36.2%, 48.4%, and 33.2% on the Average Gradient metric, respectively. Similarly, on the medical image dataset, the model outperforms the second best method on the metric Spatial Frequency by values of 41.5%, 65.0%, and 46.8%, respectively.}, archive = {J_EAAI}, author = {Pan Wu and Jin Tang}, doi = {10.1016/j.engappai.2025.111064}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111064}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mamba-based hybrid network for multi-modality feature fusion}, volume = {156}, year = {2025}, }
• (2025). Improving perceptual quality in spatiotemporal timeseries forecasting. EAAI, 156, 111062. (www)

  @article{ , abstract = {Representing spatiotemporal time series data as image sequences and learning data-driven models that capture the dynamics of such data offers an efficient method for spatiotemporal forecasting. While pixel-wise metrics such as Mean Squared Error (MSE) are commonly used to train and evaluate these forecasting models, this study reveals that such distortion metrics struggle to capture localized variations, particularly when forecasting large spatial domains with high-frequency, irregular components leading to blurry predictions that average out temporal variations. This phenomenon is attributed to the uncertainty in the data and its relation to widely used loss functions, thus highlighting a tradeoff between distortion and perceptual quality metrics in spatiotemporal forecasting. We propose custom loss functions to control this tradeoff and evaluate it with perceptual metrics on irregular spatiotemporal timeseries forecasting tasks. We also show that this tradeoff can be implicitly affected through model design decisions. Based on forecasting applications in the transportation and weather domain, our experiments demonstrate that these approaches enhance the model’s ability to capture the spatiotemporal dynamics represented in image timeseries datasets, in terms of the proposed perceptual metrics while maintaining competitive performance on traditional distortion measures such as MSE.}, archive = {J_EAAI}, author = {Shehel Yoosuf and Hamza Baali and Abdesselam Bouzerdoum}, doi = {10.1016/j.engappai.2025.111062}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111062}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving perceptual quality in spatiotemporal timeseries forecasting}, volume = {156}, year = {2025}, }
• (2025). A federated sleep staging method based on adaptive re-aggregation and double prototype-contrastive using single-channel electroencephalogram. EAAI, 156, 111060. (www)

  @article{ , abstract = {The limited availability of sleep data collected by individual institutions poses a significant challenge in developing automatic sleep staging models, especially given that most existing models rely heavily on data-driven approaches. The advent of federated learning has introduced an innovative and reliable paradigm for inter-institutional collaboration, facilitating task knowledge sharing and implicit data augmentation while preserving privacy. However, in traditional federated learning methodologies, local models are typically initialized with the global model at the onset of each iteration. This global-knowledge-centric approach often results in catastrophic forgetting during local training and inadequately adapts to the diversity of local data, compromising the overall generalization performance of the model. In this paper, we introduce a novel re-aggregation strategy for local model initialization that synergizes the global model with historical local models, thereby mitigating the undue influence of global knowledge and preserving local task-specific information. The re-aggregation weights are adaptively determined based on the confidence of the global and historical models on local datasets. Furthermore, to address the inherent ambiguity among sleep stages, a dual prototype-contrastive learning module is proposed, comprising Prototype-Consistency Contrastive (PCC) and Prototype-Distinguishability Contrastive (PDC) components. Specifically, the PCC component is designed to ensure consistency between local prototypes and unbiased prototypes derived from re-aggregation, enhancing intra-class knowledge coherence. The goal of the PDC component is to strengthen the discriminability among various sleep stages, improving inter-class differentiation. Comprehensive experiments conducted on two public datasets demonstrate the effectiveness, superiority, and flexibility of the proposed method.}, archive = {J_EAAI}, author = {Bian Ma and Lijuan Duan and Zhaoyang Huang and Yuanhua Qiao}, doi = {10.1016/j.engappai.2025.111060}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111060}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A federated sleep staging method based on adaptive re-aggregation and double prototype-contrastive using single-channel electroencephalogram}, volume = {156}, year = {2025}, }
• (2025). An intelligent three-dimensional fault detection method based on multitask learning and multi-scale feature fusion. EAAI, 156, 111059. (www)

  @article{ , abstract = {Faults, as critical controlling factors in the formation and distribution of oil and gas resources, play a vital role in seismic data analysis. Deep learning, in three-dimensional (3D) fault detection, has made significant strides in recent years. However, current research relies primarily on synthetic seismic data, limiting model generalization across diverse geological conditions. Furthermore, the diversity and scale differences in fault morphology challenge traditional encoder–decoder detection networks. To overcome these challenges, we propose an intelligent three-dimensional fault detection method based on multitask learning and multi-scale feature fusion, named MTL-MSF-Fault3D, aiming to improve fault detection accuracy and generalization. The multitask learning (MTL) network processes fault detection as the primary task and edge detection as an auxiliary task, promoting feature sharing and improving model performance on field data. The multi-scale fusion (MSF) module dynamically integrates multi-scale features by dual-channel attention gates, preserving fault edge information and enhancing detection detail. The fault detection method proposed in this paper leverages artificial intelligence (AI) technology to optimize the feature extraction and decision-making processes for faults. By automatically adjusting network parameters, it ensures a significant improvement in the model’s adaptability across different geological environments. Experiments show MTL-MSF-Fault3D excels on synthetic data and provides reliable fault interpretation on complex field seismic data, all while maintaining low computational cost. This research introduces new techniques for fault detection, with significant practical value. It demonstrates the effectiveness, scalability, and robustness of artificial intelligence in fault detection within the oil and gas exploration field.}, archive = {J_EAAI}, author = {Guangyue Zhou and Kewen Li and Xinyuan Zhu}, doi = {10.1016/j.engappai.2025.111059}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111059}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An intelligent three-dimensional fault detection method based on multitask learning and multi-scale feature fusion}, volume = {156}, year = {2025}, }
• (2025). Decimal place separable complex convolutional neural network for wideband beamforming. EAAI, 156, 111057. (www)

  @article{ , abstract = {With the rapid development of synthetic aperture radar and increasing demand for remote sensing, wideband beamforming technology has been a hot spot. Recently, deep learning from computer science has given a hint for the next generation of beamforming. Many neural network-based beamformers have been reported. However, those methods show disadvantages on wideband signals. Due to the precision limitations of current computing devices, neural networks may encounter precision errors when generating extremely small numerical values. However, the performance of beamforming is still highly sensitive to small numerical values. Existing neural network-based methods produce decreased performance due to those errors. To enhance the precision of generation and achieve a better trade-off between performance and efficiency, we propose a generation mechanism with high precision at the framework level. In this paper, the decimal place separable complex convolutional neural network (DSCCNN) is proposed for wideband beamforming. Firstly, we apply different networks to handle distinct decimal places contributing to a mixture-of-experts framework, which can increase the fitting precision. Then, multilayer perceptrons are used to enhance the learning capabilities of the proposed network’s backbone referring to current popular computer vision network architectures. Last, an improved attention module is proposed to better process the different parts of complex-valued feature maps based on the squeeze-and-excitation module. Simulation experiments show the proposed beamforming method has excellent performance in anti-jamming. The computational complexity of the proposed method is low, which is beneficial for potential engineering applications. In addition, the proposed network can be trained within a very short time.}, archive = {J_EAAI}, author = {Hairui Zhu and Bi Wen and Cong Xue and Jie Luo and Jiali Li and Shurui Zhang}, doi = {10.1016/j.engappai.2025.111057}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111057}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Decimal place separable complex convolutional neural network for wideband beamforming}, volume = {156}, year = {2025}, }
• (2025). Continual learning for unknown domain fault diagnosis in rotating machinery via diffusion-integrated dynamic mixture experts. EAAI, 156, 111056. (www)

  @article{ , abstract = {Fault diagnosis in industrial rotating machinery faces significant challenges due to varying and unknown operating conditions. Traditional models often struggle with catastrophic forgetting when adapting to new domains, leading to decreased diagnostic accuracy over time. To address these issues, we propose a novel continual learning framework, termed Diffusion-Integrated Dynamic Mixture Experts (DIMIX), designed to enhance fault diagnosis in dynamic and unknown environments. The DIMIX model integrates three key components: a dynamic mixture experts framework to reduce catastrophic forgetting, a dynamic incremental clustering mechanism for autonomous recognition and adaptation to unknown domains, and a diffusion-based generative replay inspired by neuroscience to reinforce memory retention. Experimental results demonstrate that DIMIX not only limits the catastrophic forgetting rate to around 2%, but also achieves robust fault diagnosis across previously unseen operating conditions. These findings highlight the potential of DIMIX for improving the reliability and adaptability of fault diagnosis systems in complex industrial settings.}, archive = {J_EAAI}, author = {Tianjiao Lin and Liuyang Song and Lingli Cui and Huaqing Wang}, doi = {10.1016/j.engappai.2025.111056}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111056}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Continual learning for unknown domain fault diagnosis in rotating machinery via diffusion-integrated dynamic mixture experts}, volume = {156}, year = {2025}, }
• (2025). Erratum to “A robust semi-supervised learning scheme for development of within-batch quality prediction soft-sensors”[Eng. appl. artif. intell. 133 (2024) 107920]. EAAI, 156, 111051. (www)

  @article{ , archive = {J_EAAI}, author = {Yi Shan Lee and Junghui Chen}, doi = {10.1016/j.engappai.2025.111051}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111051}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Erratum to “A robust semi-supervised learning scheme for development of within-batch quality prediction soft-sensors”[Eng. appl. artif. intell. 133 (2024) 107920]}, volume = {156}, year = {2025}, }
• (2025). Weakly supervised temporal action localization via a multimodal feature map diffusion process. EAAI, 156, 111044. (www)

  @article{ , abstract = {With the continuous growth of massive video data, understanding video content has become increasingly important. Weakly supervised temporal action localization (WTAL), as a critical task, has received significant attention. The goal of WTAL is to learn temporal class activation maps (TCAMs) using only video-level annotations and perform temporal action localization via post-processing steps. However, due to the lack of detailed behavioral information in video-level annotations, the separability between foreground and background in the learned TCAM is poor, leading to incomplete action predictions. To this end, we leverage the inherent advantages of the Contrastive Language-Image Pre-training (CLIP) model in generating high-semantic visual features. By integrating CLIP-based visual information, we further enhance the representational capability of action features. We propose a novel multimodal feature map generation method based on diffusion models to fully exploit the complementary relationships between modalities. Specifically, we design a hard masking strategy to generate hard masks, which are then used as frame-level pseudo-ground truth inputs for the diffusion model. These masks are used to convey human behavior knowledge, enhancing the model’s generative capacity. Subsequently, the concatenated multimodal feature maps are employed as conditional inputs to guide the generation of diffusion feature maps. This design enables the model to extract rich action cues from diverse modalities. Experimental results demonstrate that our approach achieves state-of-the-art performance on two popular benchmarks. These results highlight the proposed method’s capability to achieve precise and efficient temporal action detection under weak supervision, making a significant contribution to the advancement in large-scale video data analysis.}, archive = {J_EAAI}, author = {Yuanbing Zou and Qingjie Zhao and Shanshan Li}, doi = {10.1016/j.engappai.2025.111044}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111044}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weakly supervised temporal action localization via a multimodal feature map diffusion process}, volume = {156}, year = {2025}, }
• (2025). Automated research methodology classification using machine learning. EAAI, 156, 111039. (www)

  @article{ , abstract = {Scientific papers have become the primary means for disseminating scientific research, and thus, the ability to classify research papers based on different aspects has become essential. Therefore, many works have developed classification approaches; however, they focused solely on research topic-based classification. In addition, no solution has been developed to classify papers based on the applied methodology, and finally, the accuracy of the existing paper classification methods is not satisfactory. In this study, a novel automated classification methodology using a refined Extreme Gradient boosting (XGBoost) model is presented to classify the research methods employed in scientific papers. Three article sets, including quantitative and qualitative research methods, were collected from the topics of tourism, medical science and information systems, consisting of 229, 557 and 787 papers, respectively. The classification problem was considered a binary classification task to maintain interpretability. The developed model was trained and tested on article set 1 (tourism) and 2 (medical science), and then, the proposed model was applied to article set 3, (information systems and tourism). The high accuracy achieved in different research fields (90%–95% accuracies on average) indicates that the proposed classification model is generalizable because it can be successfully applied in many disciplines. The automated classifier enables the rapid acquisition of vital information and the identification of significant differences among the applied methodologies in various research domains. A future development direction will be to increase the scalability of the proposed model to achieve efficient operations on large volumes of research papers.}, archive = {J_EAAI}, author = {Zsolt T. Kosztyán and Tünde Király and Tibor Csizmadia and Attila Imre Katona and Ágnes Vathy-Fogarassy}, doi = {10.1016/j.engappai.2025.111039}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111039}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated research methodology classification using machine learning}, volume = {156}, year = {2025}, }
• (2025). Multiple lung diseases detection using advanced deep learning model with attention mechanisms and upsampling features. EAAI, 156, 111038. (www)

  @article{ , abstract = {Lung diseases represent a major health challenge, necessitating accurate and timely diagnostic methods. X-ray imaging plays a critical role in diagnosing various lung conditions, with physicians traditionally relying on these images for disease identification. However, with rapid advancements in artificial intelligence and imaging technologies, the volume of medical image data has grown significantly, spurring the development of machine learning techniques, particularly deep learning, to improve diagnostic accuracy. Unlike previous studies that focused on binary or three-class classification tasks, this paper presents an efficient deep learning model capable of classifying six types of lung X-ray images. The model employs an efficient feature extraction approach by combining the Visual Geometry Group Network (VGG16) with a Convolutional Neural Network (CNN) feature fusion, further enhanced by a Squeeze-and-Excitation (SE) block and self-attention modules. This comprehensive framework enables effective classification of tuberculosis, pneumonia, COVID-19, normal lungs, lung opacity, and pneumothorax. To validate the model, we constructed a robust dataset consisting of 45,241 X-ray images sourced from multiple databases. The proposed model was evaluated in various conditions to verify its performance. It achieves a remarkable validation accuracy of 93% in detecting the six lung diseases from X-ray images.}, archive = {J_EAAI}, author = {Jie Zhu and Mohammed A.A. Al-qaness and Dalal AL-Alimi and Heng Zhi Tao and Saeed Hamood Alsamhi}, doi = {10.1016/j.engappai.2025.111038}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111038}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multiple lung diseases detection using advanced deep learning model with attention mechanisms and upsampling features}, volume = {156}, year = {2025}, }
• (2025). Hybrid transformer-convolution neural network architecture for emoji-aware hostile post classification with custom attention mechanisms. EAAI, 156, 111037. (www)

  @article{ , abstract = {In the current digital atmosphere, social media platforms have achieved a pervasive attendance, facilitating unprecedented levels of connectivity while simultaneously nurturing environments conducive to hostility. Hindi is the most spoken language in the Indian subcontinent and the third most spoken worldwide. Twitter has 30.3 million active users in India, with many communicating in Hindi, highlighting the need for specialized content moderation tools beyond English. To address the need to identify hostile content in Hindi, we developed a dataset of 8300 Devanagari tweets, annotated into five classes, providing a nuanced analytical framework. We introduce a novel Scaled Multi-Head Attention-Convolution Neural Network Fusion Model (SMA-CFM) to confront the complexities of multiclass classification in this domain. The proposed model extracts contextual features from text and emoji embeddings, refining them via Scaled Multi-Head Attention. Convolution Neural Network (CNN) layers further capture local dependencies, ensuring classification across different hostility categories. Finally, the fused features are processed through a dense layer for final classification. Compared to Long Short-Term Memory (LSTM) and CNN, our fusion SMA-CFM approach consistently outperforms existing techniques by effectively capturing both textual and emoji semantics. This improvement highlights the importance of integrating attention mechanisms with convolutional networks for hostile post-classification. Our Cross-Lingual Language Model with RoBERTa (XLM-RoBERTa) embedding with emoji vector achieves superior performance, with F1-scores of 79.86% (Abusive), 78.58% (Defamation), 81.08% (Hate), and 76.40% (Non-Hostile). Global vectors for word representation(GloVe) with emoji vector performs best for Offensive (78.58%). The findings confirm our approach effectively improves Hindi hostile post classification.}, archive = {J_EAAI}, author = {Santosh Rajak and Ujwala Baruah}, doi = {10.1016/j.engappai.2025.111037}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111037}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid transformer-convolution neural network architecture for emoji-aware hostile post classification with custom attention mechanisms}, volume = {156}, year = {2025}, }
• (2025). Prediction of concrete failure behavior under high impact using machine learning. EAAI, 156, 111036. (www)

  @article{ , abstract = {In the field of protective engineering, the rapid and accurate prediction of concrete failure behavior under high impact loads is of great significance. It can help assess the safety of concrete structures. Several methods including empirical formulas and fully connected networks (FCN) were proposed to achieve rapid prediction, but their accuracy was limited. Numerical simulation achieved high accuracy in prediction, but the predicting speed was limited. Developing a method that balances prediction accuracy and speed is urgent. This paper collected data on concrete failure behavior under projectile impact loads, performing feature selection and normalization. A modified residual network (ResNet) structure is proposed to predict concrete failure behavior under projectile penetration. The network replaces the convolutional layers in the traditional ResNet with fully connected layers, removes the pooling layers, and reduces the number of hidden layers, making the ResNet more suited to accommodate the concrete failure data. Meanwhile, SHapley Additive exPlanations (SHAP) analysis identifies the key penetration feature parameters affecting penetration performance; and a branch network is added to the ResNet to handle these significant penetration feature parameters, enhancing the network’s ability to predict concrete failure behavior under high impact loads. Compared with empirical formulas and FCN, the presented model achieves higher prediction accuracy and better generalization ability, while also surpasses numerical simulation methods in prediction speed.}, archive = {J_EAAI}, author = {Zihan Chen and Jianqiao Li and Xiangzhao Xu}, doi = {10.1016/j.engappai.2025.111036}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111036}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction of concrete failure behavior under high impact using machine learning}, volume = {156}, year = {2025}, }
• (2025). Promoting sustainable urban mobility: An integrated fuzzy decision-making model for assessing autonomous bus alternatives. EAAI, 156, 111034. (www)

  @article{ , abstract = {Today, in addition to the increasing pressures on urban transportation authorities to achieve sustainability goals, it has become necessary to comprehensively evaluate innovative transportation technologies such as autonomous (driverless) buses due to the increasing demand for public transportation applications that will contribute to making urban transportation more sustainable with its environmental, social, economic and social dimensions. In addition, the reservations and hesitations of decision-makers about integrating autonomous buses into urban transportation systems have not been eliminated. These hesitations and reservations are mainly due to critical research, theoretical gaps, and limitations in practice. Considering these gaps, this study presents an innovative model that integrates the fuzzy logarithm methodology of additive weights (LMAW) method and the fuzzy Dombi Bonferroni (DOBI) method to evaluate and rank 20 different autonomous bus alternatives with 33 sustainability criteria. The proposed integrated decision-making procedure can effectively manage complex uncertainties while examining whether autonomous bus alternatives can be integrated into urban transportation systems based on sustainability, considering four-dimensional sustainability criteria. This finding indicates that urban transportation's user-oriented and reliable nature is critical to achieving sustainability goals. In addition, the Proterra Catalyst (A9) is the autonomous bus with the highest sustainability performance for use in urban transport, followed by the Mercedes-Benz Future Bus (A18) and Mercedes-Benz eCitaro (A8). These results regarding alternatives underline the importance of advances in autonomous vehicle technology and making these vehicles more sustainable in evaluation processes.}, archive = {J_EAAI}, author = {Ömer Faruk Görçün and Mehmet Özçalıcı and Hasan Emin Gurler and Dragan Pamucar and Vladimir Simic}, doi = {10.1016/j.engappai.2025.111034}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111034}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Promoting sustainable urban mobility: An integrated fuzzy decision-making model for assessing autonomous bus alternatives}, volume = {156}, year = {2025}, }
• (2025). Convolutional neural network-attention-gate recurrent unit-attention hybrid framework for spindle thermal error modeling with joint feature analysis under complex variable speed conditions. EAAI, 156, 111033. (www)

  @article{ , abstract = {Deep learning-based spindle thermal error modeling and compensation methods can effectively enhance manufacturing precision. The complex variable working conditions and thermal hysteresis effects in machine tool machining bring significant challenges for high-precision thermal error modeling. To address this issue, a hybrid structure network based on the feature extraction capability of Convolutional Neural Network (CNN) and the thermal hysteresis effect resolution capability of deep Gate Recurrent Unit (GRU) is established. A dual-layer attention mechanism is introduced to enhance spatial features and temporal features for improving the model's robustness and accuracy. First, complex variable working conditions result in complexity and nonlinearity of data. CNN is employed to extract spatial features due to its powerful feature extraction capability. A self-attention mechanism is introduced after CNN block to further filter important features. Due to the influence of thermal hysteresis effects, a deep GRU block is established to extract temporal features. A channel attention mechanism is introduced as the final layer of the network to achieve feature selection across different temperature channels. Second, features extracted by two attention mechanism layers are visualized and analyzed using t-distributed stochastic neighbor embedding (t-SNE) algorithm to explain the effectiveness of the dual-layer attention mechanism structure. The probability density distribution of the predicted results is calculated by kernel density estimation. Model performance is analyzed from the perspective of data distribution. Finally, the proposed model is compared with advanced methods under complex working conditions on two machine tools. The effectiveness of the proposed model is further validated through actual cutting compensation.}, archive = {J_EAAI}, author = {Sen Mu and Guoqiang Fu and Yue Zheng and Xi Wang and Caijiang Lu and Jianzhong Fu}, doi = {10.1016/j.engappai.2025.111033}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111033}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Convolutional neural network-attention-gate recurrent unit-attention hybrid framework for spindle thermal error modeling with joint feature analysis under complex variable speed conditions}, volume = {156}, year = {2025}, }
• (2025). Microfluidic droplet detection using synthetic data generated by a generative adversarial network. EAAI, 156, 111032. (www)

  @article{ , abstract = {Cellular heterogeneity, even among genetically identical cells, results in variations in their properties and behaviors, making single-cell analysis crucial for obtaining detailed insights. However, isolating single cells from a cell population poses major challenges, as conventional laboratory techniques often risk cell damage and involve complex procedures. Droplet microfluidics has emerged as a promising approach for encapsulating cells, particularly single cells, into individual droplets without causing harm. Despite this, factors like cell sedimentation and aggregation can reduce encapsulation efficiency and lead to deviations from the expected Poisson distribution. To address these challenges, leveraging artificial intelligence and deep learning to monitor, detect, and regulate encapsulation conditions in real-time is critical for enhancing system performance. However, deep learning models require substantial training data, and issues like microfluidic channel clogging and the scarcity of certain cell types often limit data availability. To overcome this limitation, researchers are turning to synthetic data generation to supplement training datasets and address data scarcity challenges effectively. This study emphasizes the potential of integrating synthetic data with cutting-edge deep learning techniques to enhance the accuracy and efficiency of single-cell analysis within droplet microfluidic systems. A diverse dataset integrating synthetic and real images was used to train the YOLOv8s model for automated detection and classification of microfluidic droplets, enhancing accuracy and system performance. The model trained on a combination of real and synthetic data outperformed the one trained using conventional data augmentation methods, achieving an mAP 0.5 of 98% due to the increased diversity of training images. It also demonstrated faster and more stable training. Additionally, the YOLOv8 network, with a detection rate of approximately 2338 droplets per second, significantly improved processing speed compared to earlier YOLO versions.}, archive = {J_EAAI}, author = {Marziyeh Karimzadeh and Mohsen Alambardar Meybodi and Mohammadreza Shams and Zahra Goli-Malekabadi}, doi = {10.1016/j.engappai.2025.111032}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111032}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Microfluidic droplet detection using synthetic data generated by a generative adversarial network}, volume = {156}, year = {2025}, }
• (2025). A novel approach of causality matrix embedded into the graph neural network for forecasting the price of bitcoin. EAAI, 156, 111031. (www)

  @article{ , abstract = {Accurately forecasting Bitcoin prices presents significant challenges due to its high volatility and the complex interactions among macroeconomic and crypto-specific variables. Traditional forecasting models often rely on correlations, which fail to capture the intrinsic causal relationships that drive price fluctuations. In this paper, we propose a novel method that integrates a Cause & Effect (C&E) Matrix within a Graph Neural Network (GNN) to explicitly model these causal dependencies. Unlike correlations, causal relationships remain relatively stable even under changing market conditions, making them more reliable for robust and interpretable forecasting. Our approach begins with causal analysis to identify the key variables influencing Bitcoin’s price, after which these causal links are translated into directed graph structures. These structures allow for the extraction of spatio-temporal features via GNN, capturing the underlying dynamics of Bitcoin’s price movements. Experimental results demonstrate that our C&E embedded GNN significantly improves short-term Bitcoin price forecasts compared to baseline models, highlighting the critical role of causality in enhancing prediction accuracy and model interpretability in volatile markets.}, archive = {J_EAAI}, author = {Xinxin Luo and Wei Yin and Bo Xiao and Jia Cao}, doi = {10.1016/j.engappai.2025.111031}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111031}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel approach of causality matrix embedded into the graph neural network for forecasting the price of bitcoin}, volume = {156}, year = {2025}, }
• (2025). Adaptive disentangled target representation for unsupervised domain adaptation in remote sensing segmentation. EAAI, 156, 111029. (www)

  @article{ , abstract = {Due to the significant differences between the source and target domains, semantic segmentation models for remote sensing images trained on the source domain often struggle to generalize effectively to new target domains, leading to performance degradation. Most solutions disentangle features into domain-invariant features shared by both domains and domain-specific features unique to each domain, relying primarily on the domain-invariant features for model construction and inference. However, during this process, critical domain-specific features as essential information are often discarded as noise, leading to a reduction in segmentation accuracy. To address these issues, this study proposes an unsupervised domain adaptation, called ChameleonRS 1 . ChameleonRS separates the domain-invariant features and the domain-specific features by disentanglement learning, and introduces a multi-level feature alignment module to enhance the representation of domain-invariant features. More importantly, it aligns predictions based on domain-invariant features with those based on fused invariant and specific features using contrastive learning, effectively capturing target-specific features. This study explores two cross-domain semantic segmentation tasks: (I) Geographic location variation. (II) Geographic location and imaging mode variation. Results show that under setting (I), ChameleonRS outperforms the state-of-the-art method by 4.79% in Intersection over Union (IoU) and 6.51% in the F1 score. Under setting (II), it achieves improvements of 4.46% and 4.82% in IoU and F1 score. Results demonstrate the superiority and adaptability of ChameleonRS under varying geographical and imaging conditions, effectively mitigating the generalization issues caused by distribution differences, and thereby promoting the development and application of remote sensing technologies in practical scenarios.}, archive = {J_EAAI}, author = {Runuo Lu and Shoubin Dong and Jianxin Jia and Xusheng Wang and Kai Liu and Jinsong Chen and Shanxin Guo and Xiaorou Zheng}, doi = {10.1016/j.engappai.2025.111029}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111029}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive disentangled target representation for unsupervised domain adaptation in remote sensing segmentation}, volume = {156}, year = {2025}, }
• (2025). A diffusion model based on multi-scale spatial mamba for medical image segmentation. EAAI, 156, 111028. (www)

  @article{ , abstract = {Medical image segmentation plays a key role in disease diagnosis, treatment planning, and monitoring disease progression. Recently, denoising diffusion models have shown significant promise in generating accurate pixel-level semantic representations. In this study, we introduce a diffusion model based on multi-scale spatial mamba (MSM-Diff) designed for precise medical image segmentation. MSM-Diff integrates the strengths of diffusion models, Mamba architecture, and convolutional neural networks to efficiently capture both global and local contextual information from complex volumetric data. The core of MSM-Diff is the Mamba-based U-shaped feature encoder (MUFE), which combines the three-dimensional multi-scale spatial Mamba model (MS-Mamba) with extracted features for enhanced multi-scale and global feature extraction. By using the mamba architecture, the model maintains linear computational complexity. Additionally, MSM-Diff incorporates a multi-scale gated spatial convolution (MS-GSC) module within MUFE to further refine spatial feature representations. Extensive evaluations of three public datasets demonstrate that MSM-Diff consistently outperforms current methods, achieving state-of-the-art performance in DSC and HD95. This model provides a robust solution for medical image segmentation by effectively capturing global context and accurately delineating boundaries, thereby improving diagnostic and treatment planning outcomes for radiologists and clinicians.}, archive = {J_EAAI}, author = {Chun Li and Qiule Sun and Muqing Zhang and Jianxin Zhang}, doi = {10.1016/j.engappai.2025.111028}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111028}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A diffusion model based on multi-scale spatial mamba for medical image segmentation}, volume = {156}, year = {2025}, }
• (2025). Multi-class agent trajectory prediction with selective state spaces for autonomous driving. EAAI, 156, 111027. (www)

  @article{ , abstract = {Understanding and predicting multi-class agents’ movement has become more critical and challenging in diverse applications such as autonomous driving and urban intelligent monitoring. The current research mainly focuses on the motion trajectory of single-class agents. However, due to real traffic scenarios’ complexity and interactive behaviors’ variability, the motion patterns displayed by various classes of agents show inherent randomness. In this paper, inspired by the linear-time sequence model Mamba, we propose a Multi-class Agent Trajectory Prediction with Selective State Spaces (MTPSS) to model the interaction between different agents and better predict the trajectory of an individual. Specifically, MTPSS includes modeling relationships in both temporal and spatial dimensions. When encoding the spatial correlation within the trajectory graph, we construct a category-based sorting approach, which puts large-size category nodes behind to enhance contextual access. Then the sorted nodes are bi-directionally scanned through Mamba blocks, which makes the model more robust to permutations. In terms of temporal, considering the highly dynamic nature of rapidly moving agents, we utilize Mamba’s remarkable performance on sequential data to effectively conduct temporal scans to capture long-range temporal dependencies. Finally, to compute physically feasible trajectories, MTPSS employs the Neural Ordinary Differential Equation to smooth the predicted trajectory of the agent. We conducted extensive experiments on two publicly available traffic datasets and compared our method with state-of-the-art methods. Quantitative experiments show that our performance metrics are superior to state-of-the-art methods, and qualitative experiments demonstrate that the predicted trajectories have good diversity, which shows its potential in real-world traffic scenarios.}, archive = {J_EAAI}, author = {Jin Fan and Zhanwen Liu and Yong Fang and Zeyu Huang and Yang Liu and Shan Lin}, doi = {10.1016/j.engappai.2025.111027}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111027}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-class agent trajectory prediction with selective state spaces for autonomous driving}, volume = {156}, year = {2025}, }
• (2025). Attention-guided low-rank convolutional weighting for industrial missing data attacks. EAAI, 156, 111025. (www)

  @article{ , abstract = {The widespread deployment of data acquisition units in industrial settings has significantly improved the efficiency of control and anomaly detection. However, when facing high proportions of missing data attacks, data quality often deteriorates, limiting the potential of data-driven optimization. To address this issue, we propose an attention-guided low-rank convolutional weighting method aimed at effectively estimating missing elements impacted by attacks. The proposed method combines attention mechanisms with convolutional unfolding techniques to fully exploit the low-rank characteristics of the data and faithfully reconstruct the missing data structure. Specifically, a low-rank convolutional weighting regularizer is introduced to capture low-rank subspace features of the original data and effectively guide the data recovery process. Meanwhile, an attention mechanism is used to accurately handle sparse noise, enhancing global robustness to sparse missing data. By employing iterative optimization, closed-form updates are achieved for solving variables and regularizers. Experimental results demonstrate that the proposed method provides advantages across several industrial application scenarios, including monoclonal antibody production in perfusion bioreactors using Chinese hamster ovary cells, multi-mode wastewater treatment plants, and image datasets. Even under high missing rates and non-uniform missing attack patterns, our method outperforms existing state-of-the-art methods.}, archive = {J_EAAI}, author = {Ming-Qing Zhang and Liang-Yu Zhou and Guo-Yu Liu and Yan-Lin He and Qun-Xiong Zhu and Xin-Yi Cao and Yang Zhang and Yuan Xu}, doi = {10.1016/j.engappai.2025.111025}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111025}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Attention-guided low-rank convolutional weighting for industrial missing data attacks}, volume = {156}, year = {2025}, }
• (2025). An adaptive appliance identification method leveraging topological data analysis. EAAI, 156, 111024. (www)

  @article{ , abstract = {Non-intrusive load monitoring (NILM) is a technique that involves analyzing changes in voltage and current flowing through the main feeder to determine which appliances are in operation and their energy consumption. With the increasing amount and diversity of electric loads nowadays, it is becoming increasingly important to extract unique load signatures and build robust classification models for NILM. Topological data analysis (TDA) studies the properties of space that are preserved under continuous deformations, which can reveal the structure and relationships within complex datasets, such as networks, graphs, and manifolds. In this paper, we use TDA as feature extractor to mine vast of non-linear shape features from Voltage–Current (V-I) trajectory for appliance identification. Then, an adaptive feature selection method based on mutual information (MI) is proposed, as a result, the selected subset of features is more discriminative for the adopted dataset. Further, using selected features, the strategy of Frienemy Indecision Region Dynamic Ensemble Selection (FIRE-DES), which adaptively selects combination of classifiers according to data-samples by considering the improvement by indecision region, is employed for non-intrusive appliance classification. By such fusion of TDA and technique of ensemble learning, the experimental results on two public datasets prove efficacy of proposed method in both identification accuracy and computation time.}, archive = {J_EAAI}, author = {Yang Han and Keke Li and Hanju Cai and Wenpeng Luan and Bochao Zhao and Bo Liu}, doi = {10.1016/j.engappai.2025.111024}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111024}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive appliance identification method leveraging topological data analysis}, volume = {156}, year = {2025}, }
• (2025). A quality index for construction big data in shield tunneling. EAAI, 156, 111023. (www)

  @article{ , abstract = {The quality of the dataset underpinning the data-driven models predefines the upper limit for their performance yet lacks a quantitative way to be captured for the construction big data generated in earth pressure balance, i.e., EPB, shield tunneling. Herein, a quality index is proposed to fill this gap and formulated as an L 2 norm of a vector composed of three components, i.e., accuracy, inclusiveness, and informativeness. The accuracy component is the ratio of non-outlier samples, i.e., a dataset containing fewer outliers shows a higher accuracy, reflecting the extent to which the dataset represents the real construction conditions during the tunneling. The inclusiveness component is the normalized envelope area of the dataset being mapped into a two-dimensional space, reflecting the range of diverse construction scenarios that have been included in the dataset. The informativeness component is the dimensionless uncertainty reduction of given data-driven models by the dataset, reflecting the contribution of datasets to the given model’s prediction. The proposed quality index is comprehensively assessed using a big database collected from multiple tunneling projects. A series of sub-datasets deliberately divided from the big database are utilized to train data-driven models by three commonly used algorithms, i.e., random forest, neural network, and K-nearest neighbors, for mapping three target functions widely concerned in tunneling, i.e., torque, thrust, and penetration. It is shown that the proposed quality index of the training data unfailingly excellently correlates with the performance of the data-driven models (R-values > 0.91) regardless of algorithms, target functions, and sample sizes.The proposed quality index serves as a theoretical basis for a series of practical application scenarios, e.g., training data selection, and core dataset development. A practical application based on the Changsha project illustrated that the training dataset selected using the quality index can significantly boost the performance of the developed data-driven models by more than 38% and reduce training time by more than 26%.}, archive = {J_EAAI}, author = {Chao Zhang and Yuhao Ren and Qihang Huang and Renpeng Chen}, doi = {10.1016/j.engappai.2025.111023}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111023}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A quality index for construction big data in shield tunneling}, volume = {156}, year = {2025}, }
• (2025). Multidimensional chaotic signals generation using deep learning and its application in image encryption. EAAI, 156, 111017. (www)

  @article{ , abstract = {In this paper, we propose a novel artificial intelligence implemented approach to generate multi-dimensional chaotic signals using the Long- and Short-Term Time-Series Network (LSTNet) for a newly contrived Two-Stage pixel/bit level Scrambling and Dynamic Diffusion (TSSDD) color image encryption. Initially, we employ the hyperchaotic Lorenz and Chen chaotic systems to produce chaotic signals. Subsequently, the LSTNet model is trained to predict these produced multi-dimensional chaotic sequences and then it generates new multi-dimensional chaotic signals. Through analysis involving phase diagrams, largest Lyapunov exponent (LE), 0–1 test, Permutation Entropy (PE), Sample Entropy (SE), Correlation Dimension (CD) and National Institute of Standards and Technology (NIST), we observe that these applied artificial intelligence signals exhibit high chaotic states and randomness. Finally, we apply these signals to demonstrate the proposed TSSDD color image encryption wherein simulation experiments indicate competitive performance against common attacks.}, archive = {J_EAAI}, author = {Shuang Zhou and Zhiji Tao and Uğur Erkan and Abdurrahim Toktas and Herbert Ho-Ching Iu and Yingqian Zhang and Hao Zhang}, doi = {10.1016/j.engappai.2025.111017}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111017}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multidimensional chaotic signals generation using deep learning and its application in image encryption}, volume = {156}, year = {2025}, }
• (2025). Harnessing prototype networks for novel plant species and disease classification in open-world scenarios. EAAI, 156, 111016. (www)

  @article{ , abstract = {Identifying novel plant categories and diseases in open-world scenarios is crucial for modern agricultural production and applications. Recent studies have extended plant disease recognition from closed-set to open-set scenarios, aiming to reject samples that do not belong to known classes. However, beyond rejection, it is necessary to classify unknown samples rather than merely labeling them as ”unknown.” This paper assumes that images of unknown samples, rejected by open-set recognition algorithms, are available. We aim to classify these unlabeled samples by leveraging prior knowledge from a labeled set. To the best of our knowledge, no existing research has addressed the classification of unknown plant species and diseases. To fill this gap, we propose a novel prototype network that models the category space relationship between known and unknown classes. Specifically, we learn a prototype vector for each known category, enabling samples to obtain distance-based category probabilities by measuring their similarity to these prototypes. This approach captures complex class boundaries more effectively than linear classification models, offering greater flexibility and accuracy. Additionally, we employ a knowledge distillation loss to optimize the category space relationship and calculate a consistency loss to balance the model’s classification performance for both known and unknown classes. To further boost performance, we incorporate the pre-trained model dino-v2. Experiments on the large-scale plant specimen dataset Herbarium19 and the plant disease dataset Plant Village demonstrate that our method surpasses baseline approaches, improving novel class accuracy by 1%–30%. This research contributes to intelligent agriculture, and we will release the code to facilitate future work in the field.}, archive = {J_EAAI}, author = {Jiuqing Dong and Wei Jin and Alvaro Fuentes and Jaehwan Lee and Yongchae Jeong and Sook Yoon and Dong Sun Park}, doi = {10.1016/j.engappai.2025.111016}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111016}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Harnessing prototype networks for novel plant species and disease classification in open-world scenarios}, volume = {156}, year = {2025}, }
• (2025). Probabilistic intervals prediction based on adaptive regression with attention residual connections and covariance constraints. EAAI, 156, 111013. (www)

  @article{ , abstract = {This paper introduces a novel prediction interval method called Adaptive Regression with Attention Residual Connection and Covariance Constraint (AR-ARCC). By integrating Monte Carlo and Bayesian methods, we leverage the strengths of both to achieve a more flexible and accurate method for generating prediction intervals. Additionally, through the optimization of the loss function, introduction of penalty terms, and improvement of mean squared error calculations, the model’s performance in interval prediction tasks is enhanced. Finally, the integration of an interactive channel heterogeneous self-attention module, combined with residual blocks, enhances the modeling capability of the neural network. The comprehensive application of these methods results in superior performance of the model in handling uncertainty and local variations.}, archive = {J_EAAI}, author = {Fan Zhang and Min Wang and Lin Li and Yepeng Liu and Hua Wang}, doi = {10.1016/j.engappai.2025.111013}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111013}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Probabilistic intervals prediction based on adaptive regression with attention residual connections and covariance constraints}, volume = {156}, year = {2025}, }
• (2025). A unified region and concept-level explainable artificial intelligence method for explainability and active learning of defect segmentation model. EAAI, 156, 111009. (www)

  @article{ , abstract = {Objective: Despite the Artificial Intelligence (AI) method having achieved great progress in defect segmentation tasks, the explainability of AI method remains a challenge since its black-box property. To guarantee its prediction result can be understood and trusted by users, recent works attempted to explain the model’s decision process through Explainable Artificial Intelligence (XAI) methods. Challenges: However, the existing XAI methods still have some limitations: (1) these XAI methods only focus on explaining model decisions from a single perspective, which usually introduces biased explanations. (2) few works consider how to leverage the explanation mechanism of XAI methods to guide the active learning process of model, which limits the application of XAI methods. Methods: To address these issues, a unified region-level and concept-level explainable AI (RC-XAI) framework is proposed for the explainability and active learning of the defect segmentation model. Novelty: Firstly, RC-XAI incorporates region-level and concept-level explanators in a collaborative manner to provide comprehensive explanations for the model decision. It enhances the reliability and robustness of explanations. Secondly, RC-XAI proposes an explainability-driven representative sample selection (ED-RSS) module to guide the model’s active learning process for improving its final performance. Findings: Experimental results on three challenging datasets demonstrate the effectiveness and generalization of the proposed RC-XAI method. Our method provides better and more comprehensive explainability compared with other XAI methods. Additionally, experiments demonstrate the potential of applying the explanation mechanism of the RC-XAI method to the active learning process of defect segmentation models.}, archive = {J_EAAI}, author = {Huangyuan Wu and Bin Li and Lianfang Tian and Chao Dong and Wenzhi Liao}, doi = {10.1016/j.engappai.2025.111009}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111009}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A unified region and concept-level explainable artificial intelligence method for explainability and active learning of defect segmentation model}, volume = {156}, year = {2025}, }
• (2025). EVolutionary independent DEtermiNistiC explanation. EAAI, 156, 111008. (www)

  @article{ , abstract = {The widespread use of artificial intelligence deep neural networks (DNNs) in fields such as medicine and engineering necessitates understanding their decision-making processes. Current explainability methods often produce inconsistent results and struggle to highlight essential signals influencing model inferences. This paper introduces the Evolutionary Independent Deterministic Explanation (EVIDENCE) theory, a novel approach offering a deterministic, model-independent method for extracting significant signals from black-box models. EVIDENCE theory, grounded in robust mathematical formalization, is validated through empirical tests on diverse datasets, including COVID-19 audio diagnostics, Parkinson's disease voice recordings, and the George Tzanetakis music classification dataset (GTZAN). Practical applications of EVIDENCE include improving diagnostic accuracy in healthcare and enhancing audio signal analysis. For instance, in the COVID-19 use case, EVIDENCE-filtered spectrograms fed into a frozen Residual Network with 50 layers (ResNet50) improved precision by 32 % for positive cases and increased the Area Under the Curve (AUC) by 16 % compared to baseline models. For Parkinson's disease classification, EVIDENCE achieved near-perfect precision and sensitivity, with a macro average F1-Score of 0.997. In the GTZAN, EVIDENCE maintained a high AUC of 0.996, demonstrating its efficacy in filtering relevant features for accurate genre classification. EVIDENCE outperformed other Explainable Artificial Intelligence (XAI) methods such as Local Interpretable Model-agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP), and Gradient-weighted Class-Activation Mapping (GradCAM) in almost all metrics. These findings indicate that EVIDENCE not only improves classification accuracy but also provides a transparent and reproducible explanation mechanism, crucial for advancing the trustworthiness and applicability of AI systems in real-world settings.}, archive = {J_EAAI}, author = {Vincenzo Dentamaro and Paolo Giglio and Donato Impedovo and Giuseppe Pirlo}, doi = {10.1016/j.engappai.2025.111008}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111008}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {EVolutionary independent DEtermiNistiC explanation}, volume = {156}, year = {2025}, }
• (2025). VectorSketcher: Learning to create a vector-based free-hand sketch. EAAI, 156, 111005. (www)

  @article{ , abstract = {Sketch synthesis refers to converting a given content image into a sketch. Existing sketch synthesis methods are generally divided into pixel-based and vector-based sketch synthesis. Pixel-based sketch synthesis methods always introduce obvious artifacts and disharmonious patterns. Besides, they cannot support generating sketches with different levels of abstraction. The vector-based sketch synthesis methods have limitations in describing the structure and semantics of the content image. To tackle these problems, we propose a novel framework called VectorSketcher, which can create vectorized sketches that accurately describe the structure and semantics of the content images without introducing obvious artifacts and disharmonious patterns. Specifically, we proposed a Multi-scale Feature-based Stroke Initialization (MFSI) to speed up the optimization and essential visual details of the given image. We introduce a Controllable Score Distillation Sampling loss (CSDS) to further learn the content image’s detail. Extensive quantitative and qualitative experiments show that VectorSketcher can generate more accurate vector-based sketches than existing state-of-the-art (SOTA) sketch synthesis methods.}, archive = {J_EAAI}, author = {Zhanjie Zhang and Quanwei Zhang and Junsheng Luan and Mengyuan Yang and Yun Wang and Lei Zhao}, doi = {10.1016/j.engappai.2025.111005}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111005}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {VectorSketcher: Learning to create a vector-based free-hand sketch}, volume = {156}, year = {2025}, }
• (2025). Localizing state space for visual reinforcement learning in noisy environments. EAAI, 156, 110998. (www)

  @article{ , abstract = {Gaining robust policies is what the visual reinforcement learning community desires. In practical application, the noises in an environment lead to a larger variance in the perception of a reinforcement learning agent. This work introduces a non-differential module into deep reinforcement learning to localize the state space for agents, by which the impact of noises can be greatly reduced, and the learned policy can be explained implicitly. The proposed model leverages a hard attention module for localization, while an additional reinforcement learning process is built to update the localization module. We analyze the relationship between the non-differential module and agent, regarding the whole training as a hierarchical multi-agent reinforcement learning model, ensuring the convergence of policies by centralized evaluation. Moreover, to couple the localization policy and behavior policy, we modify the evaluation processes, gaining more direct coordination for them. The proposed method enables the agent to localize its observation or state in an explainable way, learning more advanced and robust policies by ignoring irrelevant data or changes in noisy environments. That is, it enhances reinforcement learning’s ability to disturbance rejection. Several experiments on simulation environments and Robot Arm suggest our localization module can be embedded into existing reinforcement learning models to enhance them in many respects.}, archive = {J_EAAI}, author = {Jing Cheng and Jingchen Li and Haobin Shi and Tao Zhang}, doi = {10.1016/j.engappai.2025.110998}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110998}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Localizing state space for visual reinforcement learning in noisy environments}, volume = {156}, year = {2025}, }
• (2025). Understanding question-answering systems: Evolution, applications, trends, and challenges. EAAI, 156, 110997. (www)

  @article{ , abstract = {Question answering (QA) systems have garnered significant attention in recent years due to their potential to bridge the gap between human language understanding and machine intelligence. Consequently, a wide variety of approaches have been developed, each tailored to specific tasks. In this survey paper, we provide a comprehensive overview of three prominent QA paradigms: Extractive , generative, and Visual QA. We discuss the underlying principles, methodologies, applications, challenges, and recent trends in each of these areas. By synthesizing insights from the existing literature and research findings, we aim to provide a holistic understanding of extractive, generative, and Visual QA systems and offer insights into their strengths, limitations, and future directions.}, archive = {J_EAAI}, author = {Amer Farea and Frank Emmert-Streib}, doi = {10.1016/j.engappai.2025.110997}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110997}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Understanding question-answering systems: Evolution, applications, trends, and challenges}, volume = {156}, year = {2025}, }
• (2025). BIGC-net: A body inter-intra-parts graph convolutional network for repetitive action counting. EAAI, 156, 110996. (www)

  @article{ , abstract = {With the continuous development of human pose estimation techniques, researchers have gradually applied them to the field of repetitive action counting, resulting in pose-level methods. However, the current researches on the pose-level are still limited. Therefore, this paper proposes a simple but efficient Body Inter-intra-parts Graph Convolutional Network (BIGC-Net). Specifically, two core modules are developed in BIGC-Net: the Global Inter-Part Feature Learning Module (GIFL-Module) and the Salient Intra-Part Feature Learning Module (SIFL-Module). Unlike previous pose-level methods, which only model human joints globally and ignore local details. Instead, we innovatively introduce the concept of body parts with Graph Convolutional Networks (GCN) to the repetitive action counting task. Based on the natural topology of the human body, we divide the joints into multiple inter-intra-parts, each of which is regarded as a subgraph to form the overall graph structure. The complete action is then achieved by the collaborative operation between different subgraphs, thus modelling the action execution process more accurately. Therefore, the GIFL-Module is designed to capture the global collaborative relationships between the subgraphs. However, since the body joints are segmented into multiple parts, this segmentation may ignore the variation of local detail information within the subgraphs. To address this issue, the SIFL-Module aims to capture the local interdependencies between joints within the subgraphs, and the ability to focus on the most salient features of the subgraphs as it moves. The collaboration of these two modules further enhances the feature representation capability. Finally, extensive experimental results on the challenging benchmark datasets (RepCount-pose, UCFRep-pose, and Countix-Fitness-pose) show that the proposed BIGC-Net achieves excellent performance.}, archive = {J_EAAI}, author = {Jun Li and Jinying Wu and Qiming Li and Bangshu Xiong}, doi = {10.1016/j.engappai.2025.110996}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110996}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {BIGC-net: A body inter-intra-parts graph convolutional network for repetitive action counting}, volume = {156}, year = {2025}, }
• (2025). Harnessing high-quality pseudo-labels for robust few-shot nested named entity recognition. EAAI, 156, 110992. (www)

  @article{ , abstract = {Few-shot Named Entity Recognition (NER) methods have shown initial effectiveness in flat NER tasks. However, these methods often prioritize optimizing models with a small annotated support set, neglecting the high-quality data within the unlabeled query set. Furthermore, existing few-shot NER models struggle with nested entity challenges due to linguistic or structural complexities. In this study, we introduce R etrieving h i gh-quality pseudo-label T uning, RiTNER, a framework designed to address few-shot nested named entity recognition tasks by leveraging high-quality data from the query set. RiTNER comprises two main components: (1) contrastive span classification, which clusters entities into corresponding prototypes and generates high-quality pseudo-labels from the unlabeled data, and (2) masked pseudo-data tuning, which generates a masked pseudo dataset and then uses it to optimize the model and enhance span classification. We train RiTNER on an English dataset and evaluate it on both English nested datasets and cross-lingual nested datasets. The results show that RiTNER outperforms the top-performing baseline models by 1.67%, and 3.04% in the English 5-shot task, as well as the cross-lingual 5-shot tasks, respectively.}, archive = {J_EAAI}, author = {Hong Ming and Jiaoyun Yang and Shuo Liu and Lili Jiang and Ning An}, doi = {10.1016/j.engappai.2025.110992}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110992}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Harnessing high-quality pseudo-labels for robust few-shot nested named entity recognition}, volume = {156}, year = {2025}, }
• (2025). Identifying change point in dynamic bearing degradation process: A retrospective approach based on visibility graph modeling. EAAI, 156, 110988. (www)

  @article{ , abstract = {With the explosive growth of data in industrial environments, the artificial intelligence (AI)-driven condition monitoring analysis has become increasingly prominent in bearing running status. Successful identification of early changes of the status relies on precise and reliable dynamic modeling of degradation process. The graph model has been recognized as an effective tool for characterization of dynamic running behaviors of rotary equipment, where the time complexity is an important but unavoidable issue in consideration of designing a real monitoring system. Following the research line of graph-based modeling, this paper introduces the visibility graph into dynamic modeling and investigates its performance on bearing condition monitoring. By virtue of visibility graph based modeling, the irrelevant attributes can be ignored which simplifies the constructed graph structure, and moreover a predefined penetration parameter is also used to strike a balance between validity and simplicity of the graph model. As such, the amount of effective information of the graph model can be controllable. In addition, its computational complexity of information extraction is also decreased. This paper adopts a machine learning based retrospective analysis mechanism to measure the dissimilarity of model health indicators to enhance the generalization ability and feasibility, such that the change decision can be made by a common null hypothesis testing. Experimental results conducted on two datasets show the effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Guangyuan Chen and Ke Li and Keyan Yuan and Zixu Chen and Teng Wang and Guoliang Lu}, doi = {10.1016/j.engappai.2025.110988}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110988}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identifying change point in dynamic bearing degradation process: A retrospective approach based on visibility graph modeling}, volume = {156}, year = {2025}, }
• (2025). An efficient technique for reversible data hiding using bidirectional histogram shifting and multistage embedding. EAAI, 156, 110986. (www)

  @article{ , abstract = {Reversible Data Hiding has been extensively investigated due to its myriad applications in different fields such as defense, medical, cloud storage, and secure data communication over public networks. However, most of the techniques suffer due to concerns about data capacity, communication overhead, and data security. To solve these concerns, a novel technique has been proposed that ensures high security and optimum capacity, wherein no additional overhead is required to be transmitted through a separate channel. For this, bidirectional embedding of secret data has been proposed wherein both the left and right peaks from the main histogram peak have been exploited for data embedding, thereby achieving high embedding capacity. Data embedding is performed in multistage in both plain and encrypted domains to achieve not only optimum quality but also high security. The secret data embedding capacity is also tunable, wherein the number of bits per pixel can be determined from the size of the cover and the amount of secret data. In addition, block-wise embedding of secret data further enhances the data embedding capacity. The proposed technique is demonstrated to perform better in comparison to the state-of-the-art techniques while, experimental analysis over different types of cover by different performance metrics was done. On average, the PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index) values of 43.22 dB and 0.9984 respectively, is obtained for one bit per peak pixel embedding.}, archive = {J_EAAI}, author = {Sanjay Kumar and Gurjit Singh Walia and Anjana Gupta}, doi = {10.1016/j.engappai.2025.110986}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110986}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient technique for reversible data hiding using bidirectional histogram shifting and multistage embedding}, volume = {156}, year = {2025}, }
• (2025). Ground motion inversion utilizing sensor-based structural response time histories with a multi-head attention-enhanced temporal convolutional network. EAAI, 156, 110955. (www)

  @article{ , abstract = {Accurate measurement of ground motion (GM) time histories is essential for precise structural analysis and safety evaluation. However, high costs, limited availability, and maintenance demands of GM recording devices lead to several challenges. This paper aims to present the Multi-Head Attention-Enhanced Temporal Convolutional Network (MHA-TCN Net), a framework developed for inverting GM from recorded structural response (SR) data during earthquakes. A case study on complex multi-degree-of-freedom (MDOF) structures demonstrates the MHA-TCN Net's ability to predict GM time histories across elastic and various plastic deformation ranges, showcasing its potential for real-time GM inversion. By incorporating a multi-head attention mechanism along with an overlapping subset strategy, the MHA-TCN Net achieves high precision in GM inversion. It consistently demonstrates low validation losses, with the mean absolute error not exceeding 0.063, exhibits test correlation coefficients as high as 0.95, and maintains a mean square error below 0.4 on unseen test datasets. Comparative experiments against the Temporal Convolutional Network, Long Short-Term Memory, Convolutional Neural Network, Gated Recurrent Unit, and Transformer models demonstrate the MHA-TCN Net's faster convergence and superior accuracy for both training and validation phases. Importantly, the proposed approach can handle inherent sensor noise, ensuring reliable performance under challenging field conditions, thereby highlighting the contribution of advanced artificial intelligence methods for engineering applications with constrained data environments. Furthermore, without retuning the network, the model demonstrates high accuracy with limited real-time training datasets, proving its versatility across various real-world datasets, especially in areas with SR sensors but few seismic instrumentations.}, archive = {J_EAAI}, author = {Ali Zar and Shuang Li and Changqing Li and Yicheng Chen and Muhammad Akbar}, doi = {10.1016/j.engappai.2025.110955}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110955}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ground motion inversion utilizing sensor-based structural response time histories with a multi-head attention-enhanced temporal convolutional network}, volume = {156}, year = {2025}, }
• (2025). Deep learning-driven false data injection attack in renewable integrated smart grids. EAAI, 156, 110953. (www)

  @article{ , abstract = {The modernisation of power grids accompanies growing reliance on information and communication infrastructure, raising the vulnerability to cyber threats. Even a trivial, undetectable manipulation of big data can seriously compromise the data integrity of the targeted power system, resulting in adverse consequences, e.g., power theft, misleading control actions, false tripping, etc. The integration of distributed energy resources further amplifies data complexity, introducing new challenges in ensuring data security. This paper proposes a novel deep learning-based False Data Injection attack scheme utilising Conditional Generative Adversarial Networks to assess potential breaches in data integrity within renewable-integrated power networks. The study evaluates the effectiveness of the proposed attack against advanced False Data Detection mechanisms in complex smart grid environments. The IEEE 5-bus, 30-bus, and 118-bus systems incorporating photovoltaic and wind-based distributed generation with realistic load and generation dynamics are modelled as target networks. Simulation results critically analyse the limitations and effectiveness of the proposed attack, demonstrating its capability to bypass state-of-the-art False Data Detection safeguards.}, archive = {J_EAAI}, author = {Rehan Nawaz and Rabbaya Akhtar and Saad Ullah Khan and Siqi Bu and Muhammad Habib Mahmood}, doi = {10.1016/j.engappai.2025.110953}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110953}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning-driven false data injection attack in renewable integrated smart grids}, volume = {156}, year = {2025}, }
• (2025). Fuzzy reinforcement learning prescribed-time algorithm for the rigid–flexible coupled robotic mechanisms with input deadzone. EAAI, 156, 110939. (www)

  @article{ , abstract = {In the horizontal plane, the dynamic model for rigid–flexible coupled robotic mechanisms under large beam-deformations are determined through the utilization of a comprehensive modeling approach based on the virtual work concept. To track the desired angular positions of such robotic mechanisms with input nonsymmetric deadzone, the fuzzy nonsymmetric deadzone compensation based prescribed time adaptive reinforcement learning control strategy, incorporated with virtual robust linear quadratic state feedback input is proposed. To handle the unknown nonsymmetric input deadzone and uncertain system dynamics, an actor prescribed time fuzzy law is adopted. For further reduce the large vibration modes and tracking errors simultaneously, a virtual input and the proposition of a robust linear quadratic state feedback controller are developed. With the Lyapunov direct strategy, the angular position tracking errors and the flexible vibration of robotic mechanisms are demonstrated to converge to a tiny confined compact set. In numerical scenarios, the proposed fuzzy nonsymmetric deadzone compensation-based prescribed time adaptive reinforcement learning strategy simultaneously reduced mean angular tracking errors in a preset time and flexible vibration when compared respectively to virtual robust state feedback-free and backstepping mode control baselines.}, archive = {J_EAAI}, author = {Xingyu Zhou and Haoping Wang and Yang Tian}, doi = {10.1016/j.engappai.2025.110939}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110939}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzy reinforcement learning prescribed-time algorithm for the rigid–flexible coupled robotic mechanisms with input deadzone}, volume = {156}, year = {2025}, }
• (2025). Spatiotemporal effect driven landslide susceptibility mapping at fine scales: A deep learning model based on multidimensional feature fusion and source data adaptation. EAAI, 156, 110924. (www)

  @article{ , abstract = {Landslides, as frequent natural hazards, pose severe threats to human life and property safety. This study proposes a deep learning model named Local and Global Feature Convolutional Network with Interferometric Synthetic Aperture Radar (LGCR-Net). The model aims to address the issues of multi-source data, heterogeneity, and interference present in Landslide Susceptibility Mapping (LSM), while also compensating for the timeliness limitations of LSM. Firstly, the model introduces a parallel structure for extracting multidimensional features, achieving effective feature fusion through a Multidimensional Feature Fusion Module(MF). Subsequently, an image enhancement module is incorporated to process the source data, significantly enhancing the model's stability and source data adaptability. Furthermore, by integrating the Local and Global Features Convolutional Network (LGC-Net) model with Interferometric Synthetic Aperture Radar (InSAR) technology, the LGCR-Net model is formed, enabling the output LSM to possess timeliness. This model has been successfully applied to the study of the Baihetan reservoir and its surrounding areas in China. Experimental results indicate that after incorporating the MF and image enhancement module, the model's Precision increased by 0.0151 and 0.0213, respectively, while Recall improved by 0.0239 and 0.0334, respectively. Compared to traditional deep learning models, the LGC-Net model demonstrates superior practicality and predictive reliability. With the integration of InSAR technology, the improved LSM not only addresses the model's sensitivity shortcomings in medium and low susceptibility areas but also maintains predictive performance in high susceptibility areas, providing a more comprehensive and accurate representation of the potential landslide risks within the study area.}, archive = {J_EAAI}, author = {Quan Shi and Xiaoliang Xu and Jianlin Li and Huafeng Deng and Qinghai Zhang and Delin Tan and Yu He}, doi = {10.1016/j.engappai.2025.110924}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110924}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatiotemporal effect driven landslide susceptibility mapping at fine scales: A deep learning model based on multidimensional feature fusion and source data adaptation}, volume = {156}, year = {2025}, }
• (2025). Overcoming catastrophic forgetting in tabular data classification: A pseudorehearsal-based approach. EAAI, 156, 110908. (www)

  @article{ , abstract = {Continual learning (CL) poses the important challenge of adapting to evolving data distributions without forgetting previously acquired knowledge while consolidating new knowledge. In this paper, we introduce a new methodology, coined the Tabular-data Rehearsal-based Incremental Lifelong Learning framework (TRIL3), designed to address the phenomenon of catastrophic forgetting in an online, task-free setting. TRIL3 utilizes an Incremental Learning Vector Quantization (ILVQ) algorithm as an efficient prototype-based incremental generative model to store and generate synthetic data to preserve knowledge over time, and the Deep Neural Decision Forest (DNDF) algorithm, which was modified to run incrementally to learn supervised classification tasks for tabular data. Based on tests conducted to determine the optimal percentage of synthetic data and comparisons with other available task-free CL proposals, we conclude that TRIL3 outperforms other methods in the literature using only 50% of synthetic data.}, archive = {J_EAAI}, author = {Pablo García-Santaclara and Bruno Fernández-Castro and Rebeca P. Díaz-Redondo}, doi = {10.1016/j.engappai.2025.110908}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110908}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Overcoming catastrophic forgetting in tabular data classification: A pseudorehearsal-based approach}, volume = {156}, year = {2025}, }
• (2025). Implementation of reinforcement learning for enhanced pressure control in a 190,000-barrel crude distillation unit: The first full-scale commercial deployment. EAAI, 156, 110907. (www)

  @article{ , abstract = {We introduce the first successful full-scale implementation of a Reinforcement Learning-based Pressure Control System (RLPCS) within a Crude Distillation Unit (CDU) in the petroleum refining industry. Our approach utilizes an offline reinforcement learning (RL) technique that leverages log data from multiple operators’ manual operations to derive an optimal automated operational model. This method is safe as it does not interfere with the system currently in operation and does not incur additional operational costs. The trained model automates pressure control decisions, effectively resolving problems associated with manual control methods, such as high labor intensity and inconsistent operator control. To achieve this, we overcame various challenges related to training artificial intelligence (AI) in complex industrial environments. Our approach involved designing an appropriate Markov Decision Process (MDP) and incorporating a legacy control system for continuous data integration. We trained the Soft Actor–Critic (SAC) algorithm using offline data. However, applying the standard SAC algorithm to offline data and deploying this model directly in the online system posed substantial estimation problems. To address these, we integrated conservative loss and Return to Go into the SAC algorithm. Consequently, our model reduced operator intervention time by 84%, and variability in pressure control decreased significantly, resulting in a 12.8% reduction in cumulative errors. This demonstrates that RLPCS has reached the stage where it can be applied to permanent operations in real-world refining processes, not just a temporary demonstration. Building on this proven operational readiness, our work represents the industry’s first large-scale commercial deployment of an RL-based pressure control model in a CDU. The novelty lies in the comprehensive integration of MDP formulation, model architecture, and algorithmic strategies, each carefully tailored to the complex industrial environment, thereby bridging the gap between theoretical advances and real-world application.}, archive = {J_EAAI}, author = {Dongchan Seo and Dongil Kim and Hyoeun Son and Yusung Kim}, doi = {10.1016/j.engappai.2025.110907}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110907}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Implementation of reinforcement learning for enhanced pressure control in a 190,000-barrel crude distillation unit: The first full-scale commercial deployment}, volume = {156}, year = {2025}, }
• (2025). Multi-pulse superposition for droplet volume control in inkjet printing based on model and data fusion. EAAI, 156, 110884. (www)

  @article{ , abstract = {Inkjet printing technology for fabricating organic light-emitting diode display panels offers advantages such as high material utilization and the capability for large-area manufacturing. When printing display panels with varying resolutions, ejecting droplets of different sizes from the nozzle is often necessary to balance print quality and efficiency. However, due to nozzle size limitations, the volume range of stable droplets produced by a single-pulse driving waveform is relatively narrow, with the maximum volume being less than twice the minimum volume. Therefore, approaches based on superposition of multi-pulse waveforms have attracted attention, but existing studies only implement manual design of waveforms based on experimental laws and rarely involve automatic regulation of multi-pulse waveform parameters, which is not favorable for industrial applications. Based on combining a meniscus vibration model with industrial ejection data, this paper extracts control strategies from historical data using deep reinforcement learning, and recommends initial waveform parameters through a fuzzy system. Then, the multi-pulse waveform parameters are automatically adjusted in real-time based on the observed droplet volume to fuse the droplets at the nozzle, enabling a wider range of droplet volume closed-loop control. Experiments on industrial inkjet printing equipment implemented intelligent closed-loop regulation of multi-pulse driving waveforms, successfully controlling droplets of different sizes such as 2, 4, and 8 picoliter with an error accuracy of less than ± 4 % . This approach applies artificial intelligence algorithms to inkjet printing engineering and intelligently adjusts the multi-pulse waveform parameters to enhance the controllable range of droplet volumes.}, archive = {J_EAAI}, author = {Xiao Yue and Xin Li and Jiankui Chen and Wei Chen and Hua Yang and Jincheng Gao and Zhouping Yin}, doi = {10.1016/j.engappai.2025.110884}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110884}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-pulse superposition for droplet volume control in inkjet printing based on model and data fusion}, volume = {156}, year = {2025}, }
• (2025). Prediction of high-performance concrete compressive strength using decision tree-guided artificial neural network pretraining approach. EAAI, 156, 110828. (www)

  @article{ , abstract = {Deep learning models exhibit substantial nonlinear learning capabilities, enabling them to predict the compressive strength of high-performance concrete based on its mix proportions and to discern the intricate nonlinear relationships between these proportions and compressive strength. However, the data-hungry nature of deep learning necessitates extensive training datasets on mix proportions, and generating such large-scale, high-quality concrete mix datasets incurs significant costs. To address this issue, this study proposes a Decision Tree-guided Artificial Neural Network Pretraining (TANNP) approach, aimed at training high-accuracy concrete compressive strength prediction models using small-scale datasets. This approach synergizes the advantages of decision trees in managing small datasets and the powerful learning capabilities of neural networks. The TANNP method was experimentally validated on a concrete compressive strength dataset. The results demonstrate that compared to directly employing neural networks, the TANNP approach significantly improves performance metrics: the coefficient of determination increased from 0.91 to 0.97, the root mean square error decreased from 5.10 MPa to 3.26 MPa (a reduction of approximately 36.1%), the mean absolute error was reduced from 3.00 MPa to 2.10 MPa, the mean absolute percentage error dropped from 10.72% to 7.24%, and the A20-Index improved from 0.88 to 0.92, achieving state-of-the-art results on this dataset. These findings indicate that the TANNP approach effectively mitigates the data-hunger issue inherent in deep learning models and significantly enhances the accuracy of concrete compressive strength predictions.}, archive = {J_EAAI}, author = {Yang Zhang and Xinghai Yuan and Xuanpeng Zhang and Heng Wang and Pan He and Ling Luo and Chuanyun Xu}, doi = {10.1016/j.engappai.2025.110828}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110828}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction of high-performance concrete compressive strength using decision tree-guided artificial neural network pretraining approach}, volume = {156}, year = {2025}, }
• (2025). Efficient real-time object detection for embedded and industrial systems:The you only look once-compact approach. EAAI, 156, 109927. (www)

  @article{ , abstract = {The You Only Look Once (YOLO) series has been widely adopted in engineering applications; however, obtaining sufficient data in the engineering field is often challenging. Existing YOLO models frequently struggle to achieve good detection performance on small and medium-sized datasets and face challenges when deploying high-resolution inputs (e.g., 640x640 or higher) on embedded devices. To address these issues, we propose a new YOLO network, called You Only Look Once-Compact (YOLO-C), by improving the backbone and head of the YOLO series. We systematically explore the impact of different backbone architectures, neck designs, detection heads, loss functions, and label assignment methods on experimental performance. Specifically, we optimize the High-Performance Graphics Processing Unit Network version 2 (HGNetv2) by introducing structural reparameterization techniques, which enhance the model's ability to learn more efficient and robust features. Additionally, we incorporate large kernel convolutions to capture broader contextual information and use a shared-weight detection head to reduce model complexity while maintaining high performance. Experimental studies on seven diverse datasets demonstrate that YOLO-C achieves superior generalization and detection results across various scenarios. Notably, YOLO-C can achieve over 140 frames per second (FPS) when deployed on embedded devices, even with high-resolution inputs. On the Northeastern University surface defect (NEU-DET) dataset, a distilled and pruned nano version of YOLO-C outperforms the Real-Time Detection Transformer (RT-DETR) with only 3.03 % of the parameters and 6.14 % of the computational cost, while achieving 2.6 % higher detection accuracy.}, archive = {J_EAAI}, author = {YuRan Chen and BaiTing Zhao and XiaoFen Jia and TianBing Ma}, doi = {10.1016/j.engappai.2024.109927}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {109927}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient real-time object detection for embedded and industrial systems:The you only look once-compact approach}, volume = {156}, year = {2025}, }
• (2025). A novel method for surface defect inspection: Unsupervised continual learning for gradually varying domains. EAAI, 155, 111150. (www)

  @article{ , abstract = {Current surface defect detection methods hinder the improvement of industrial product quality and efficiency due to low domain adaptability and low real-time problems. This study proposes a novel defect inspection method with gradually varying domain adaptation and lightweight design to overcome current limitations. First, the lightweight MobileNetV4 and dynamic feature pyramid network are integrated as the backbone network to reduce defect inspection time and resource consumption. Then, a novel domain adaptation method based on batch streaming, buffering and sample replay is proposed for setting up source, intermediate, target and fusion domains to mitigate catastrophic forgetting due to changes in different inspection scenarios. The proposed method is trained on an industrial defect dataset containing 6 defect types, 5 classes of inspection scenarios and 17,336 images. The performance of this method is compared with other state-of-the-art methods in terms of accuracy, parameters, computational effort, model size and speed. The results show that the proposed method performs well in terms of accuracy and real-time performance over other methods for the gradually varying domain adaptation problem, reaching 82.13 %, 3.86 million, 2.37 billion and 25.61 frames per second (f/s) for mean average precision (mAP), parameters, computations, and speed, respectively. This work bridges the research gap between model light weighting and unsupervised domain adaptation, solves the challenges of low accuracy and slow speed due to complex models and limited generalization, and provides a solution for fast and accurate inspection of surface defects in industrial products.}, archive = {J_EAAI}, author = {Chen Renfei and Li Zhongwen and Zhao Zeyang}, doi = {10.1016/j.engappai.2025.111150}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111150}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel method for surface defect inspection: Unsupervised continual learning for gradually varying domains}, volume = {155}, year = {2025}, }
• (2025). Multi-objective optimization of lightweight and crashworthiness of automotive front-end structures based on stacked regression surrogate. EAAI, 155, 111138. (www)

  @article{ , abstract = {To improve the efficiency of multi-objective optimization design for automobile lightweighting and crashworthiness, this article takes the front-end structure of two passenger sedan as the research object and carries out research on the surrogate model based on the Stacked regression (SR) and carries out multi-objective optimization based on the surrogate model. The base regressors of SR support vector regression (SVR) and Gaussian process regression (GPR), and the meta regressor is elastic net regression (ENR). Based on the Bayesian optimization method, the hyperparameters of the surrogate model are determined and the performance of the single model is compared with that of the stacked regressor. The results show that the nonlinear response prediction performance of the SR is significantly better than the single SVR and GPR as well as the response surface and kriging with a small sample size (50 sample points). As the sample size increases (100 sample points), the performance improvement of the SR is not significant but is at least close to the best performing base regressor. SR is a robust surrogate model with strong nonlinear fitting capability that can be used for automotive lightweighting and crashworthiness design and other complex engineering prediction problems with strong nonlinearity and high sample acquisition cost.}, archive = {J_EAAI}, author = {Fengmin Lian and Dengfeng Wang and Zifeng Zhang and Hongli Chen and Wenchao Xu}, doi = {10.1016/j.engappai.2025.111138}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111138}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective optimization of lightweight and crashworthiness of automotive front-end structures based on stacked regression surrogate}, volume = {155}, year = {2025}, }
• (2025). The end-to-end chip surface defect segmentation method based on the diffusion model and attention mechanism. EAAI, 155, 111131. (www)

  @article{ , abstract = {Surface defect detection is critical in chip manufacturing and plays a key role in product quality control. However, due to the complexity of surface defects, existing segmentation methods still face challenges in robustness and noise resistance, and the encoding of features does not balance local detail information and global dependency well, leading to lower accuracy. Recently, diffusion models have gained traction in various vision tasks for their strong generative and noise-resistant capabilities. To address these challenges, this study introduces an innovative end-to-end defect segmentation method based on a diffusion model and attention mechanism (DASeg-Net). First, the High-frequency and Low-frequency attention mechanism (HiLo) is introduced into the feature encoder, which employs a dual branch to enhance both the global dependency and local detail information of the defect features, thereby achieving balanced encoding. Second, an enhanced feature fusion module, Spatial Attention Fusion Module (SAFM), is designed to strengthen the integration of features across different levels from a spatial perspective, facilitating the effective merging of deep semantic features with shallow ones. Third, an auxiliary edge branch is introduced to bolster the focus on the subtle features of defective edges. Fourth, a transformer-based denoising decoder is employed to combine the encoded features with noise masks, progressively denoising to generate high-quality segmentation masks. Finally, extensive comparative experiments were conducted on our chip defect segmentation and other surface defect segmentation datasets. Specifically, the proposed method achieved mean Intersection over Union (mIoU) of 82.06 %, 86.04 %, 76.22 %, and 83.39 % on the four datasets. The results demonstrate that the proposed method exhibits superior robustness and achieves state-of-the-art performance.}, archive = {J_EAAI}, author = {Zilin Xia and Yufan Zhao and Jinan Gu and Wenbo Wang and Zedong Huang and Yan Gao and Peiyue Sun}, doi = {10.1016/j.engappai.2025.111131}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111131}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The end-to-end chip surface defect segmentation method based on the diffusion model and attention mechanism}, volume = {155}, year = {2025}, }
• (2025). A lightweight defect detection transformer for printed circuit boards combining image feature augmentation and refined cross-scale feature fusion. EAAI, 155, 111128. (www)

  @article{ , abstract = {End-to-end object detectors based on the Transformer excel at detecting tiny defects on printed circuit board (PCB), but are limited by complex structures and poor generalization. To address the issues, a lightweight framework for PCB surface tiny-defect detection is proposed based on real-time detection transformer (Lite-DETR). Firstly, Lite-DETR designs a lightweight and efficient backbone network (LEBN) to preserve more delicate details and enhance attention to the contextual information of defects. Secondly, Lite-DETR devises an image feature augmentation module (IFAM) to improve data generalization. Finally, the refined cross-scale feature fusion module (RCFFM) is introduced to selectively propagate the information between features, enabling the model to capture multi-scale defect features effectively. At the same time, Lite-DETR employs Intersection over Union with minimum point distance and normalized Wasserstein distance to optimize bounding box matching. Lite-DETR achieves 99.0 % and 99.2 % mean average precision on two different PCB defect datasets, outperforming other advanced defect detection models. Notably, Lite-DETR has a low number of parameters, with only 5.2 million, reducing 74 % compared to real-time detection transformer. The experimental results show that Lite-DETR effectively balances accuracy and complexity, providing a reliable solution for tiny-defect detection in complex industrial environments.}, archive = {J_EAAI}, author = {Tao Luo and Yongbing Zhou and Donglin Shi and Qinglin Yun and Shuying Wang and Jian Zhang and Guofu Ding}, doi = {10.1016/j.engappai.2025.111128}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111128}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight defect detection transformer for printed circuit boards combining image feature augmentation and refined cross-scale feature fusion}, volume = {155}, year = {2025}, }
• (2025). Illustration of low oscillating magnetic field on sodium alginate-based hybrid nanofluid flow between two revolving disks: An artificial neural network-based study. EAAI, 155, 111101. (www)

  @article{ , abstract = {The application of hybrid nanofluids is pivotal in improving heat transfer, and aiding in the thermal regulation of electronic devices, heat exchangers, and cutting-edge manufacturing processes. The integration of hybrid nanofluids and artificial neural networks enhances the development of smart cooling systems, optimizing thermal regulation. This study examines the flow of magnetite ( F e 3 O 4 ) and graphene ( G r ) nanoparticles with a base fluid sodium alginate over a porous coaxial disk in a low oscillating magnetic field in conjugation with thermal radiation. The Runge-Kutta 4th order scheme is operated to analyze the characteristics of the flow fields followed by the suitable transformation used for the conversion of dimensional governing equations to their corresponding non-dimensional form. In a novel way, this research utilizes artificial neural networks (ANN) to design an effective predictive model for thermal transfer rate, comparing its performance with response surface methodology (RSM) and validating the results through analysis of variance (ANOVA). The outcomes are the velocity distribution decreases with a higher magnetization parameter and a large Reynolds number results in a higher velocity. Nusselt number achieves it most efficient thermal transfer rate at 88 epoch with gradient of 9.9805 × 10 − 8 .}, archive = {J_EAAI}, author = {Ram Prakash Sharma and Bimal Kumar Barik and V. Vinay Kumar and Abhishek Sharma}, doi = {10.1016/j.engappai.2025.111101}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111101}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Illustration of low oscillating magnetic field on sodium alginate-based hybrid nanofluid flow between two revolving disks: An artificial neural network-based study}, volume = {155}, year = {2025}, }
• (2025). Compact-sparse prototype calibration network for few-shot continual fault diagnosis of rotating machinery. EAAI, 155, 111099. (www)

  @article{ , abstract = {Rotating machinery inevitably generates only a few samples of new fault categories during long-term operation, which requires the fault diagnosis model to incrementally learn few new categories and retain the existing fault knowledge. Recent few-shot continual fault diagnosis (FSCFD) methods mainly rely on constructing prototype classifiers and generating virtual samples to address the challenges of catastrophic forgetting and overfitting. However, this would ignore the rich feature information in the base session, while the limited incremental data makes it difficult to accurately depict the new category feature information. Therefore, a compact-sparse prototype calibration network (CSPCN) is proposed to improve the diagnosis capacity for new category faults in the FSCFD scenario. First, a compact-sparse base loss (CSBL) is employed to reserve sufficient space for new fault categories by maximizing the variance distribution among base prototypes. Second, an incremental prototype calibration classifier (IPCC) is designed to improve the ability to distinguish new categories by integrating new prototypes with the weighted base prototypes in real time. Extensive experiments conducted on the subway train bogie and variable load gearbox dataset validate the proposed method's exceptional diagnostic performance. Through multidimensional comparisons with state-of-the-art FSCFD methods and rigorous ablation experiments, CSPCN demonstrates significant improvements in effectively identifying and distinguishing new fault categories.}, archive = {J_EAAI}, author = {Shen Yan and Haidong Shao and Xinyi Wang and Haomiao Zhang and Yiming Xiao and Bin Liu}, doi = {10.1016/j.engappai.2025.111099}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111099}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Compact-sparse prototype calibration network for few-shot continual fault diagnosis of rotating machinery}, volume = {155}, year = {2025}, }
• (2025). Computational analysis for thermal and mass transfer in pumping flow of ellis fluid along with solid particles: Morlet wavelet neural networks approach. EAAI, 155, 111096. (www)

  @article{ , abstract = {The current study conducts a parametric analysis of thermal transport and mass transfer during peristaltic circulation within a symmetric channel, utilizing numerical methods and artificial intelligence techniques. This study examines a multi-phase flow involving a non-Newtonian Ellis fluid with suspended solid particles. Additionally, the research explores the influences of chemical reactions along with the electroosmotic phenomena on the flow dynamics. Using the approximation of the lubrication hypothesis, the flow problem is modeled with the assumptions of lengthy wavelength and tiny Reynolds coefficient. The combined ordinary differential equations of the fluid and particulate phases are solved by the Morlet-wavelet neural networks integrated with particle swarm optimization and neural networks algorithm. To validate the intelligence-based especially mean square error results are further compared using the numerical NDSolve and physics informed neural networks with Adam optimizer as a reference solution. The values of Morlet-wavelet neural networks with particle swarm, Morlet-wavelet neural networks with neural networks algorithm and hybrid of Morlet-wavelet with particle swarm and neural networks algorithms based fitness functions are ranging from 10 − 03 − 10 − 04 , 10 − 04 − 10 − 05 and 10 − 05 − 10 − 07 respectively. The absolute error between the proposed three approaches and reference solution between the 10 − 02 − 10 − 04 , 10 − 04 − 10 − 05 and 10 − 05 − 10 − 07 . The mean and median values of mean squared error over 70 independent runs are ranging between 10 − 06 − 10 − 07 and 10 − 07 − 10 − 09 for 12 distinct cases visualized in through figures. The velocity function enhances with the gain in the values of electroosmotic factor and Ellis fluid parameter. Further, statistical metrics calculated over multiple runs to check the convergence and efficiency of the proposed approaches. This study can be used in chemical processing, biomedical engineering, and environmental engineering industries, where the transportation of non-Newtonian fluids with solid particles is important. It optimizes heat and mass transfer in reactors, drug delivery, water treatment, and slurry transport, utilizing AI-based simulations for enhanced efficiency and performance.}, archive = {J_EAAI}, author = {Muhammad Naeem Aslam and Arshad Riaz and Mehpara Shehzadi and Safia Akram and M.M. Bhatti}, doi = {10.1016/j.engappai.2025.111096}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111096}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Computational analysis for thermal and mass transfer in pumping flow of ellis fluid along with solid particles: Morlet wavelet neural networks approach}, volume = {155}, year = {2025}, }
• (2025). Artificial intelligence-based predictive models for shear wave velocity of soils: A comprehensive review. EAAI, 155, 111095. (www)

  @article{ , abstract = {Shear wave velocity (V s ) of soils is a crucial property in geotechnical engineering practice, affecting seismic site response analysis, seismic hazard assessment, and dynamic soil-structure interaction. The precise determination of V s is crucial in assessing the dynamic behavior of soils during seismic events, as it markedly influences the amplification and attenuation of ground motions. While several empirical equations have been proposed thus far for estimating V s in earthen materials, the majority of them lack the required accuracy and predictive capability. As a result, there has been a growing tendency among practicing engineers towards utilizing Artificial Intelligence (AI) for V s prediction. This paper presents an extensive overview of the developments in deploying AI and its subsets, including Machine Learning (ML) and Deep Learning (DL) techniques, for the precise estimation of V s in soil deposits. Notably, despite the importance of shear wave velocity as a key geotechnical parameter, no prior review study has exclusively focused on evaluating it using AI-based models. This review systematically examines various AI-based methodologies employed by researchers to enhance the reliability and precision of V s predictions using soil properties and in-situ test data. The advantages of ML techniques over conventional empirical correlations are thoroughly analyzed and critically compared. Additionally, the paper discusses the relative performance of different AI-based approaches, outlining their strengths and limitations in V s estimation. The review also provides a general qualitative assessment of V s measurement methods, offering guidance on selecting the most appropriate approach based on project-specific requirements and constraints. Finally, through a critical evaluation of existing literature, key knowledge gaps are identified, and potential directions for future research in this domain are proposed.}, archive = {J_EAAI}, author = {Meghdad Payan and Parsa Asadi and Amirhossein Jamaldar and Mahdi Salimi and Payam Zanganeh Ranjbar and Danial Jahed Armaghani and Xuzhen He and Daichao Sheng}, doi = {10.1016/j.engappai.2025.111095}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111095}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence-based predictive models for shear wave velocity of soils: A comprehensive review}, volume = {155}, year = {2025}, }
• (2025). Multi-scale fuzzy self-attention network for hyperspectral image classification with small-samples. EAAI, 155, 111092. (www)

  @article{ , abstract = {Abstract—With the widespread application of convolutional neural networks (CNN), significant breakthroughs have been achieved in hyperspectral image (HSI) classification. However, the lack of training samples remains one of the primary factors contributing to low classification performance. Moreover, HSIs may be affected by factors such as lighting conditions, environmental variations, and photographing distances during the collecting process, inevitably introducing noise. This noise significantly increases the uncertainty of the classification process, particularly in small sample scenarios. To alleviate these issues, a multi-scale fuzzy self-attention network (MFSAN) is proposed. Firstly, a multi-scale fuzzy embedding module (MFEM) is designed to effectively model fuzzy dependencies between features through three parallel paths, mitigating noise-induced uncertainty across multiple scales. Secondly, inspired by the principles of Transformers, a weighted dual-distance combined self-attention module (WD 2 CAM) is proposed to enhance global context representation by utilizing a novel spectral similarity measure. Finally, a new multiple feature extraction module is developed to fully exploit the rich information of HSIs under limited training samples. This module extracts spectral, spatial, and spectral-spatial features using multiple three-dimensional convolutions with different receptive fields. Experimental results on three datasets demonstrate that MFSAN achieves superior classification accuracy compared to state-of-the-art methods, with performance improvements in overall accuracy (OA) ranging from 1.43 % to 9.84 % under 1 % training samples.}, archive = {J_EAAI}, author = {Ziyi Li and Jindong Xu and Qianpeng Chong and Yu Yan}, doi = {10.1016/j.engappai.2025.111092}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111092}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale fuzzy self-attention network for hyperspectral image classification with small-samples}, volume = {155}, year = {2025}, }
• (2025). A comparative study of methods for measuring node influence in complex networks. EAAI, 155, 111088. (www)

  @article{ , abstract = {Identifying influential nodes in complex networks has attracted significant research attention due to its applications in diverse fields ranging from viral marketing to epidemics and behavioral analytics. Most of the existing algorithms primarily assess node influence by measuring their spreading power, encompassing a spectrum from structural centrality metrics to machine learning and graph neural network based approaches. This study first provides a comprehensive review of various influential node identification methods and subsequently conducts an extensive comparative analysis. Using a unified platform applied to multiple standard real-world networks, we evaluate the performance, predictive accuracy, and computational cost of these methods. Our findings indicate that no single algorithm consistently outperforms others across all network types. While learning-based methods, particularly graph neural network, generally achieve higher predictive accuracy, they exhibit lower interpretability and impose greater computational costs. Therefore, selecting an appropriate method necessitates a balance between complexity and accuracy, tailored to the specific application context.}, archive = {J_EAAI}, author = {Seyed Amir Sheikh Ahmadi and Laleh Tafakori and Mahdi Jalili and Parham Moradi}, doi = {10.1016/j.engappai.2025.111088}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111088}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A comparative study of methods for measuring node influence in complex networks}, volume = {155}, year = {2025}, }
• (2025). Robustness-driven belief rule base for complex systems. EAAI, 155, 111087. (www)

  @article{ , abstract = {Robustness is a critical aspect of modeling and evaluating complex systems, as it ensures the stability and reliability of engineering decisions in uncertain and dynamic environments. Despite their increasing application in system prediction, evaluation, and diagnosis, belief rule base (BRB) models often suffer from reduced robustness when confronted with environmental fluctuations and noisy input data. To address these challenges, this study introduces a novel methodological framework that significantly enhances the robustness of belief rule base models. The proposed framework makes the following contributions: (1) a dynamic Gaussian membership function replacing the traditional triangular membership function to improve the stability of input transformations; (2) a weighted average computation method for the Lipschitz constant, offering a more precise and comprehensive evaluation of robustness; and (3) the integration of Lipschitz regularization with a multiattribute optimization algorithm to achieve an optimal balance between robustness, interpretability, and accuracy. Through experimental validation of aircraft relay health assessment and lithium battery life prediction, as well as six benchmark case studies, the results show that the proposed model has higher robustness and interpretability than the traditional belief rule base model, while maintaining high prediction accuracy. These findings highlight the potential of the proposed approach to advance decision-making reliability in dynamic and high-stakes engineering systems.}, archive = {J_EAAI}, author = {Aosen Gong and Wei He and You Cao and Guohui Zhou and Hailong Zhu and Ning Ma}, doi = {10.1016/j.engappai.2025.111087}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111087}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robustness-driven belief rule base for complex systems}, volume = {155}, year = {2025}, }
• (2025). Parallel-wise global and local attention vision transformer-based generative adversarial network using fourier transform loss for generating fake iris image. EAAI, 155, 111080. (www)

  @article{ , abstract = {The technological progress in generative adversarial network (GAN) has facilitated the creation of highly realistic counterfeit images. This, in turn, has paved the way for presentation attacks, named as spoof attacks, targeting a range of biometric recognition systems, including those for faces, fingerprints, veins, and irises. To mitigate these risks, significant efforts have been directed toward developing presentation attack detection (PAD) methods specifically tailored to identify fake iris images generated by the technology of GAN in the field of iris recognition. However, most of them used in the existing research do not consider the difference in the frequency domain although the difference in frequency domain makes the fake iris image easier to distinguish from the true (real) one. To tackle these problems and create a more robust PAD model, we propose a parallel-wise global and local attention vision transformer-based generative adversarial network (PGLAV-GAN) that considers both global and local context, and can generate fake iris images with similar frequency domain features to true (real) ones through Fourier transform (FT) loss. The generator in PGLAV-GAN is designed to minimize the loss of feature information while allowing feature information at different resolutions to be effectively utilized, through the incorporation of nested U-shaped network (UNet++) and deep supervision as the backbone network architecture. Moreover, the performance of PAD was assessed by applying various additional post-processing techniques, confirming that the proposed method surpassed the state-of-the-art (SOTA) presentation attack (PA) methods. When the proposed model was validated through an experiment using two open databases, the average classification error rate of the PAD model was 1.51 % for the liveness detection-iris competition 2017 (LiveDet-Iris-2017)-Notre Dame dataset and 1.9588 % for the LiveDet-Iris-2017-Warsaw dataset. This indicates that the proposed model outperformed SOTA models, and produced fake iris images that are more authentic and sophisticated for PA, enabling more robust PAD model to be constructed with the fake iris images generated by our method and enhancing the consequent security level of iris recognition system.}, archive = {J_EAAI}, author = {Jung Soo Kim and Jin Seong Hong and Seung Gu Kim and Kang Ryoung Park}, doi = {10.1016/j.engappai.2025.111080}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111080}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Parallel-wise global and local attention vision transformer-based generative adversarial network using fourier transform loss for generating fake iris image}, volume = {155}, year = {2025}, }
• (2025). One-dimensional convolutional neural network model driven intelligent operation phases identification of hydraulic press using energy data. EAAI, 155, 111079. (www)

  @article{ , abstract = {Continuously monitoring the equipment's working condition is critical to control the quality variations of the stamping metal. In the conventional monitoring platform, numerous sensors are retrofitted to the hydraulic press to monitor its condition with the high cost and complexity. In contrast to the method of retrofitting the hydraulic press, in this work, we propose an intelligent monitoring system to effortlessly and accurately assess and determine the operation phases of hydraulic press machinery with minimal retrofitting. In this system, a one-dimensional convolutional neural network (1-DCNN) model was established to deeply mine the underlay information contained in the equipment's power consumption data to predict the working status. Subsequently, Bayesian hyperparameters optimization algorithm was adopted to systematically explore the best combination of structural parameters of the 1-DCNN model for better performance. Concurrently, sensitivity analysis of these hyperparameters was conducted to quantitatively ascertain their influence on the model performance. Furthermore, some machine learning (ML) models were compared to the proposed 1-DCNN model and this proposed intelligent monitoring system was implemented on the cloud-based server, ultimately, to achieve real-time classification of operation phases of the hydraulic press. The study results revealed that the kernel size of convolution layers can significantly influence the classification accuracy of the 1-DCNN model compared to other hyperparameters. Compared to these traditional ML models, the proposed 1-DCNN model performed better with a commendable classification accuracy of 98.89 % in predicting operation phases. With the leverage of the proposed system, accurate quantitative assessments of energy consumption of each operation phase of the targeted hydraulic press were achieved, thereby offering invaluable insights for the formulation of energy-saving strategies.}, archive = {J_EAAI}, author = {Yanxiong Liu and Yuwen Shu and Zhicheng Xu and Senbo Han and Changbang Zhang and Mingzhang Chen}, doi = {10.1016/j.engappai.2025.111079}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111079}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {One-dimensional convolutional neural network model driven intelligent operation phases identification of hydraulic press using energy data}, volume = {155}, year = {2025}, }
• (2025). High-speed portrait video segmentation using the hybrid combination of deep-learning models and boundary movement adjustment. EAAI, 155, 111077. (www)

  @article{ , abstract = {As global warming intensifies, the development of energy-efficient Artificial Intelligence (AI) technologies has become crucial. Additionally, the growing demand for on-device AI in smartphones, extended reality devices, and autonomous vehicles necessitates AI systems that can function effectively on low-performance hardware. To address these needs, this study proposes hybrid methods in the field of Portrait Video Segmentation (PVS). Our proposed hybrid models leverage Deep-learning based Segmentation Models (DSMs) and a novel Boundary Movement Adjustment (BMA) process to achieve speed and accuracy balance. The Hybrid Serial Model (HSM) not only accelerates PVS but also improves energy efficiency while maintaining a similar level of accuracy. On the other hand, the Hybrid Parallel Model (HPM) enables high-performance PVS even on low-performance devices, ensuring no video frames are lost during high-speed segmentation processing. Tests conducted on Jetson Nano, Raspberry Pi, and a desktop PC demonstrate the effectiveness of these models, showing improvements in PVS speed while maintaining accuracy close to that of traditional DSMs. HSM increased PVS speed from 15.2 Frames Per Second (FPS) to 25.1 FPS on a desktop PC with a 0.5 % accuracy loss, and from 6.3 FPS to 16.5 FPS on a Jetson Nano with a 1 % loss. HPM reached 30 FPS on a desktop PC with a 0.05 % loss, and 29.7 FPS on a Jetson Nano with a 1 % loss. On the Raspberry Pi, the HPM method improved speed from 2.9 FPS to 29.8 FPS, demonstrating its adaptability for low-performance devices.}, archive = {J_EAAI}, author = {Yong-Woon Kim and Yung-Cheol Byun and Addapalli V.N. Krishna and Balachandran Krishnan}, doi = {10.1016/j.engappai.2025.111077}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111077}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {High-speed portrait video segmentation using the hybrid combination of deep-learning models and boundary movement adjustment}, volume = {155}, year = {2025}, }
• (2025). A novel method for predicting plant leaf disease based on machine learning and deep learning techniques. EAAI, 155, 111071. (www)

  @article{ , abstract = {Plants are valued for their practical and for their medical drives (food, medicine, etc.) thus considered indispensable. Diseases aiming at the plat growth may result in inflicting havoc to both the crop viability and economical values. Consequently, the ability to detect and to classify the disease occurring in plants are important in cases of making a healthier crop productions. Nonetheless, this sector procures an effective planning and well-performing approaches for easier deployment in fetching better results. Both Deep Learning (DL) and Machine Learning (ML) are employed in the detection of diseases in the plants and are highly valued for their favorable results. However, they layback at certain points when the detection is being done for a larger area of disease detection. To resolve these issues the proposed study aims in deploying, Multi-Gradient Deep Convolutional Neural Network with Adaptive Support Vector Machine (M-D-C-A-S) as a primitive approach in leaf disease detections. The proposed model aims in achieving higher rates of accuracy and to outperform than the traditional and conventional forms of approaches. To evaluate the efficacy of the prediction of leaf diseases, certain performance metrics are employed, such as accuracy rates, precision levels, f1-score, and recall rates.}, archive = {J_EAAI}, author = {A. Dinesh and S.P. Balakannan and M. Maragatharajan}, doi = {10.1016/j.engappai.2025.111071}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111071}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel method for predicting plant leaf disease based on machine learning and deep learning techniques}, volume = {155}, year = {2025}, }
• (2025). Intelligent adaptive control of ship dynamic positioning using extreme learning machine and disturbance observer. EAAI, 155, 111061. (www)

  @article{ , abstract = {Ships are often affected by many different types of disturbances during their voyage. In this paper, a prescribed performance control scheme based on adaptive extreme learning machine (ELM) and disturbance observer is proposed to solve the control problem of ship dynamic positioning (DP) with multiple disturbances. These include unknown nonlinearities in the ship model and wave disturbances modeled using nonlinear exosystems. The unknown nonlinear function is included in the exosystem, which makes the wave model more realistic. ELM is a machine learning method based on feedforward neural network, which is faster than the traditional single-hidden layer feedforward neural network and ensures high learning accuracy. First, the ELM network is applied to approximate the unknown nonlinear functions in the DP and exosystems. Subsequently, a neural network disturbance observer based on ELM is designed to estimate the external disturbances. With an accurate estimate, one can design a controller to compensate for the disturbances. In order to improve the closed-loop control precision and dynamic performance, performance functions are designed to constrain the error vector of the system. This design can ensure the convergence of the state error of the ship system in a predetermined time. Using the ELM disturbance observer technique, prescribed performance functions, and the command filter backstepping method, a neural network-based anti-disturbance control scheme with prescribed performance for ship DP system is developed here. The closed-loop stability of the combined DP ship system with unknown model nonlinearities and disturbances modeled using a nonlinear exosystem is analyzed via Lyapunov function. Virtual experiments showed that the DP ship controlled by the proposed method can quickly track the reference trajectory and maintain stability when unknown nonlinearities of the model and external wave disturbances are included. The proposed method improves the anti-disturbance ability of DP ships and helps maintain smooth sailing missions and exploration activities.}, archive = {J_EAAI}, author = {Hongdu Wang and Wenwen Zhao and Mansour Karkoub and Junrong Wang}, doi = {10.1016/j.engappai.2025.111061}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111061}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent adaptive control of ship dynamic positioning using extreme learning machine and disturbance observer}, volume = {155}, year = {2025}, }
• (2025). A regularized transformer with adaptive token fusion for alzheimer's disease diagnosis in brain magnetic resonance images. EAAI, 155, 111058. (www)

  @article{ , abstract = {Alzheimer's Disease (AD) poses significant challenges in neuroimaging, where early and accurate diagnosis is essential for timely intervention. Magnetic Resonance Imaging (MRI) serves as a key modality for AD diagnosis, providing detailed anatomical insights without radiation exposure. While traditional Convolutional Neural Networks (CNNs) excel at local feature extraction from MRI data, they fail to capture global contextual information critical for AD detection. To address this limitation, we propose the Lightweight Robust Alzheimer's Disease Vision Transformer (LRAD-ViT), a computationally efficient framework tailored for early AD detection. LRAD-ViT enhances the global learning capabilities of Vision Transformers through a novel adaptive token fusion technique. The proposed method selectively identifies and merges non-essential tokens within brain MRI scans, optimizing computational efficiency without sacrificing diagnostic performance. By dynamically adapting the token structure during model operation, the method prioritizes diagnostically relevant regions in the images. Additionally, randomized learning regularization improves learning dynamics and model robustness. Additionally, randomized learning regularization improves learning dynamics and model robustness. Validation on two brain MRI datasets from the Alzheimer's Disease Neuroimaging Initiative (ADNI) demonstrates that LRAD-ViT achieves a diagnostic accuracy of 93.41 % on AD versus cognitive normal (CN) and 90.95 % on CN versus mild cognitive impairment (MCI), all while reducing computational demands by around 40 %. These metrics substantiate LRAD-ViT's superior diagnostic performance and computational efficiency, making it a promising tool for the early and accurate diagnosis of Alzheimer's Disease, with significant implications for clinical practice.}, archive = {J_EAAI}, author = {Si-Yuan Lu and Yu-Dong Zhang and Yu-Dong Yao}, doi = {10.1016/j.engappai.2025.111058}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111058}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A regularized transformer with adaptive token fusion for alzheimer's disease diagnosis in brain magnetic resonance images}, volume = {155}, year = {2025}, }
• (2025). A semantic-embedded multi-view graph to sequence learning approach for air traffic flow prediction in the terminal maneuvering area of multi-airport system. EAAI, 155, 111054. (www)

  @article{ , abstract = {The primary objective of this study is to predict the sector-level air traffic flow in the terminal maneuvering area of multi-airport system (MAS-TMA) with a novel semantic-embedded multi-view graph to sequence learning approach (SE-MV-G2S). To characterize the complex airport operation interactions and highly coupled traffic flows within the MAS-TMA, a topic model is employed in the proposed SE-MV-G2S method to identify the hidden air traffic flow patterns and the semantic similarity between sectors. Then, a multi-view graph convolution network is built to fully capture the spatial heterogeneous correlations between sectors, including a geographical view, air route view, and semantic view. The bidirectional long short-term memory network and local spatial-temporal attention mechanism are further utilized to encode the temporal dependencies among sector-level air traffic flow variables and to capture the cross-spatial-temporal effects across multiple time steps. Three-month automatic dependent surveillance-broadcast (ADS-B) trajectory data is collected from a representative multi-airport system in Central and South China to illustrate the procedure. The results indicate that the developed SE-MV-G2S network can collaboratively encode the spatiotemporal characteristics in the sector-level air traffic flow data, which achieves more accurate and robust prediction performance than other compared models during multiple look-ahead time steps. The results of this study could provide insightful suggestions to air traffic authorities for developing effective coordination and congestion mitigation strategies, such as reallocating routes and optimizing aircraft sequences to enhance the efficiency and safety of multi-airport operations.}, archive = {J_EAAI}, author = {Jie Bao and Junfeng Zhang and Yixuan Chen and Zijie Zhang and Jiawei Kang}, doi = {10.1016/j.engappai.2025.111054}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111054}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A semantic-embedded multi-view graph to sequence learning approach for air traffic flow prediction in the terminal maneuvering area of multi-airport system}, volume = {155}, year = {2025}, }
• (2025). A novel landslide susceptibility mapping method based on a backpropagation neural network algorithm with optimized non-landslide samples and hyperparameters. EAAI, 155, 111052. (www)

  @article{ , abstract = {Landslide susceptibility mapping is essential for early warning and hazard mitigation. To address the uncertainty in non-landslide sample selection and optimize model hyperparameters, this study introduces a multi-sample label learning (MSLL) approach. Additionally, a meta-heuristic optimization algorithm, Gradient-based optimizer (GBO), is used to optimize the Back Propagation Neural Network (BPNN) model. A spatial database is constructed using 167 historical landslide events and 12 evaluation factors. Non-landslide samples are selected with buffer control sampling (BCS) and MSLL methods. The GBO algorithm is then applied to optimize the hyperparameters of the MSLL-BPNN model. Model performance is evaluated using Area Under the Curve (AUC), Confusion Matrix, Accuracy, and Precision. Results show that the MSLL method improves the AUC by approximately 3 % for both training and testing samples, compared to BCS, indicating a more reliable selection of non-landslide samples. GBO optimization further increases the AUC of the BPNN model by 4 % for training and 3 % for testing, demonstrating the importance of hyperparameter optimization in improving model accuracy. Overall, the MSLL method effectively reduces uncertainty in non-landslide sample selection, while GBO optimization significantly enhances the BPNN model's performance, both are crucial for accurate landslide susceptibility mapping.}, archive = {J_EAAI}, author = {Shandong Liu and Xingyuan Jiang and Xudong He and Haijun Qiu and Yi Yang and Rongqian Zhao and Hongsen Hu}, doi = {10.1016/j.engappai.2025.111052}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111052}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel landslide susceptibility mapping method based on a backpropagation neural network algorithm with optimized non-landslide samples and hyperparameters}, volume = {155}, year = {2025}, }
• (2025). Applying convolutional attention mechanisms and human memory search for effective english-urdu translation. EAAI, 155, 111043. (www)

  @article{ , abstract = {In recent decades, machine translation has become a prominent area of research, with the primary objective of overcoming language barriers. Early approaches primarily centered on word-for-word translation between source and target languages. However, there has been a shift towards data-driven models, such as neural machine translation and statistical methods, driven by advancements in Artificial Intelligence (AI), communication and computing technologies. This paper introduces a novel Convolutional Attention Mechanism-based Human Memory Search (CAM-HMS) algorithm for translating English into Urdu, to achieve high-quality and effective machine translation. The proposed model consists of several key phases, including pre-processing, sentence padding, word embedding, encoding, decoding, and target text generation. A new spider web-based search strategy is also incorporated to enhance the translation process in neural machine translation. The performance is evaluated using a UMC005 English-Urdu dataset, Parallel corpus for English & Urdu language, and the Roman Urdu to English Translation Dataset. Various automatic evaluation metrics such as the Bilingual Evaluation Understudy (BLEU), National Institute of Standards and Technology (NIST), Word Error Rate (WER), Accuracy, Precision, Recall, and F-Measure are used to assess the model's efficiency, and its output is compared to that of Google Translate. The proposed model achieves an average BLEU score of 82.14 %, NIST of 79.51 %, WER of 2.77 %, Accuracy of 98.99 %, Precision of 98.95 %, Recall of 98.90 %, and F-Measure of 98.92 %, demonstrating its effectiveness in producing high-quality machine translation results.}, archive = {J_EAAI}, author = {K. Jayasakthi Velmurugan and H. Faheem Nikhat and K. Suresh and S. Hemavathi and V. Kavitha}, doi = {10.1016/j.engappai.2025.111043}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111043}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Applying convolutional attention mechanisms and human memory search for effective english-urdu translation}, volume = {155}, year = {2025}, }
• (2025). Scheduling and route planning for forests rescue: Applications with a novel ant colony optimization algorithm. EAAI, 155, 111042. (www)

  @article{ , abstract = {With the increasing frequency and intensity of forest fires due to global climate change and human activities, timely and efficient fire rescue has become critical. Helicopter inspections are a preventive measure that significantly improves fire incident resolution efficiency. Existing approaches to helicopter route planning often focus on single-stage scheduling (either offline pre-planning or online reactive adjustments), which fail to dynamically balance pre-scheduled routes with real-time emergency responses. Additionally, traditional optimization algorithms like the Ant Colony Optimization (ACO) suffer from premature convergence and instability caused by overly random initial search solutions and restrictive tabu lists in later search phases. To address these gaps, this paper proposes a two-stage dynamic path planning model that integrates offline pre-scheduling with online adaptive adjustments. The model first optimizes inspection routes for all points using a novel Improved Ant Colony Optimization (IACO) algorithm, which introduces dynamic pheromone initialization and a stage-dependent pheromone update strategy to enhance exploration-exploitation balance. During real-time operations, when emergencies occur, the model dynamically dispatches the nearest available helicopter while reconfiguring remaining inspection paths. Two representative scenarios validate the model's effectiveness: (1) independent regional dispatch where a helicopter adjusts its route to handle an emergency within its assigned cluster, and (2) cross-region joint dispatch involving multi-helicopter coordination across clusters. The IACO algorithm outperforms traditional ACO by reducing total flight distances by up to 4.1 % in offline planning and improving emergency response efficiency by 7.3 % in dynamic scenarios. This work provides a robust framework for balancing routine inspections with agile emergency management in forest fire rescue operations.}, archive = {J_EAAI}, author = {Xu Wangying and Xie Naiming}, doi = {10.1016/j.engappai.2025.111042}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111042}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Scheduling and route planning for forests rescue: Applications with a novel ant colony optimization algorithm}, volume = {155}, year = {2025}, }
• (2025). Data-driven disc cutter wear prediction for shield tunneling in complex strata: Model comparison and interpretation. EAAI, 155, 111041. (www)

  @article{ , abstract = {Real-time and accurate prediction of cutter wear can reduce costs and minimize the risks associated with frequent maintenance. Machine learning models have been widely used in cutter wear analysis for shield tunneling, whereas selection and interpretation of these models under complex strata are rarely discussed. Here, a dataset collected from a large-diameter shield tunneling project encountering complex strata is established. 12 input parameters are selected, characterizing Tunnel Boring Machine (TBM) operation, cutter mechanical, and geological condition to the output parameter – cutter wear, quantified by reduction in cutter radius. Test set results show Convolutional Neural Network (CNN) has the best performance with a coefficient of determination (R 2 ) of 0.8749 followed by Gradient Boosting Machine (GBM) model with an R 2 of 0.8427 and K-Nearest Neighbors (KNN) model with an R 2 of 0.8329. An empirical log-log regression model is established with an R 2 of 0.751, enabling rapid calculations for on-site assessments. Data perturbation tests further confirm the robustness of both the CNN and KNN models. We also interpret the model performance with SHapley Additive exPlanations (SHAP) and Pearson correlation coefficients, finding that the cutter installation radius contributes the most. This study provides reference for future development of Artificial Intelligence (AI)-driven cutter wear analysis and prediction, supporting safe and cost-effective on-site shield tunneling.}, archive = {J_EAAI}, author = {Linxuan Yuan and Xiaobin Ding and Arnold Yuxuan Xie and Weiran Huang}, doi = {10.1016/j.engappai.2025.111041}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111041}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven disc cutter wear prediction for shield tunneling in complex strata: Model comparison and interpretation}, volume = {155}, year = {2025}, }
• (2025). Enhancing two dimensional irregular pattern packing using a modified genetic algorithm and image processing. EAAI, 155, 111040. (www)

  @article{ , abstract = {This paper presents a novel artificial intelligence (AI)-driven approach to two-dimensional irregular pattern packing by integrating advanced image processing with an enhanced Genetic Algorithm (GA). From meta-heuristic optimization methods, the GA is adapted to improve material utilization and the compactness of pattern placement. We introduce a simple and effective overlap detection technique based on pixel counting, alongside a two-dimensional packing optimization strategy guided by each pattern's expanded area. These innovations make the placement process more flexible and precise, leading to improved overall packing efficiency. The results demonstrate that the refined GA combined with the expanded-area-based optimization strategy significantly enhances material utilization and pattern packing performance. This research not only addresses key challenges in irregular packing through an AI-based approach but also establishes a new benchmark for future optimization techniques in this domain.}, archive = {J_EAAI}, author = {Longhui Meng and Liang Ding and Maen Atli and Aqib Mashood Khan and Yicai Shan and Muhammad Umar Farooq}, doi = {10.1016/j.engappai.2025.111040}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111040}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing two dimensional irregular pattern packing using a modified genetic algorithm and image processing}, volume = {155}, year = {2025}, }
• (2025). Dynamic customer behavior prediction in subscription services using causal reinforcement learning. EAAI, 155, 111030. (www)

  @article{ , abstract = {This study addresses a critical gap in predictive analytics for subscription-based services by integrating causal inference with a Modified Markov Decision Process model, enhanced by reinforcement learning, to predict customer behavior dynamically. While traditional approaches in big data analytics focus on customer segmentation and churn probability, they often fail to make adaptable, forward-looking predictions. By identifying causal variables and defining customer states, this research advances segmentation accuracy and captures the evolving value of a customer over time. The proposed methodology, which computes transition probability matrices and initial state probabilities, demonstrates superior predictive accuracy, achieving mean absolute error of 4.27 and root mean squared error of 4.24 on the wikipedia web traffic dataset, and mean absolute error of 10.89 and root mean squared error of 4.25 on a multinational organization dataset. These results signify the efficacy of the proposed method in outperforming existing models, offering valuable insights into the dynamic nature of customer behavior in subscription services.}, archive = {J_EAAI}, author = {Manoranjan Gandhudi and Alphonse P.J.A. and Vasanth Velayudham and Leeladhar Nagineni and Gangadharan G.R.}, doi = {10.1016/j.engappai.2025.111030}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111030}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic customer behavior prediction in subscription services using causal reinforcement learning}, volume = {155}, year = {2025}, }
• (2025). Calibration-free, machine learning based wavelength modulation spectroscopy sensor for fast gas concentration measurements. EAAI, 155, 111026. (www)

  @article{ , abstract = {The paper presents the application of machine learning (ML) regression algorithms in the gas laser sensor providing concentration determination using a wavelength-modulation-spectroscopy (WMS). The first advantage, the most important from the practical point of view, is the ability to perform measurements without troublesome, complex and expensive calibrations (calibration-free) using standard samples of mixtures with different concentrations, pressures and temperatures. This was achieved while ensuring the measurement accuracy required in this type of application and significantly shortening the sensor response time - up to 600 times compared to other known solutions. The second advantage that has a significant impact on the innovativeness of the solution is the use of a new method for simulating large sets of WMS-specific signal data characterized by high compliance with experimental results. The data set takes into account the variability of parameters such as the concentration of gas molecules, their temperature and pressure, which have a significant impact on the absorption of laser radiation by the tested molecules. To accurately determine the absorption value, it is necessary to know the sensor responses to all possible combinations of these parameters. The solution presented in this paper is the answer to this problem. For this purpose, several pre-selected machine learning models were tested, among which the best performance was achieved by custom convolutional neural network (CNN), providing an accuracy described by a mean absolute percentage error (MAPE) of less than 1 % for the validation set and 9.7 % for the test set, using experimental signals collected for reference gas mixtures characterized by an expanded uncertainty of 7.8 %. This aspect should be considered as the third advantage of the presented research, namely its comprehensiveness, i.e. the transition from theoretical research to experiment using a custom-designed laboratory setup of the WMS methane sensor, consisting of, among others, a quantum cascade laser (QCL) with a wavelength of 8014 nm (nm) and a multi-pass-cell providing a path length of 349 cm (cm). The obtained results proved that the deep learning (DL) approach can be used for accurate and calibration-free determination of gas concentration based on the analysis of the absorption signals, ensuring a detection repeatability corresponding to an absorption coefficient of 5.6∙10 −6 1/cm and reducing the time required for this process from 37 s (s) to 0.06 s.}, archive = {J_EAAI}, author = {Filip Musiałek and Jacek Wojtas}, doi = {10.1016/j.engappai.2025.111026}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111026}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Calibration-free, machine learning based wavelength modulation spectroscopy sensor for fast gas concentration measurements}, volume = {155}, year = {2025}, }
• (2025). A robust maximal margin hypersphere support vector machine with pinball loss. EAAI, 155, 111020. (www)

  @article{ , abstract = {Noisy data poses a significant challenge in classification tasks, often degrading the performance of traditional machine learning models. Although the Maximum Margin Hyper-Sphere Support Vector Machine captures data distribution patterns effectively, its sensitivity to outliers limits its robustness in noisy environments. To address this issue, we propose an enhanced model—the Maximal Margin Hyper-Sphere Support Vector Machine based on Pinball loss. By integrating the Pinball loss function, our model improves noise robustness through two key mechanisms: minimizing the hypersphere radius to enhance intra-class compactness and maximizing the distance between sphere centers to strengthen inter-class separation. The asymmetric penalty mechanism of the Pinball loss further bolsters the ability of our model to handle noisy data. Extensive experiments are conducted on artificially generated noisy datasets, as well as three types of complex datasets: class-imbalanced, hierarchical nested, and non-spherical highly overlapping datasets. The results demonstrate that the proposed method not only exhibits strong noise resistance but also maintains superior performance and robustness across diverse complex environments. Furthermore, experiments on 14 benchmark datasets further validate the superior performance of our model, with an average improvement of 7.57 % over existing baseline models methods.}, archive = {J_EAAI}, author = {Ting Ke and Mingzhu Meng and Mengyan Wu and Jianfeng Qin}, doi = {10.1016/j.engappai.2025.111020}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111020}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust maximal margin hypersphere support vector machine with pinball loss}, volume = {155}, year = {2025}, }
• (2025). A multi-objective adaptive memetic algorithm and engineering application for a double-floor layout problem with separate human and vehicle transport elevators. EAAI, 155, 111010. (www)

  @article{ , abstract = {Planning the use and location of elevators is crucial in ensuring the safety and cost-efficiency of multi-floor workshop layouts. To this end, this study proposes a double-floor layout problem considering separate human and vehicle transport elevators. The constructed mixed-integer programming model analyzes flexible position constraints of elevators and different types of transport routes, intending to minimize employee movement and material handling costs. Subsequently, a multi-objective adaptive memory algorithm that includes a unique three-layer encoding and decoding strategy and adaptive local search is developed to solve this problem. Experimental results on three small-scale benchmark instances validate the accuracy of the proposed model and algorithm. Moreover, the developed algorithm obtains the Pareto frontier solution in all benchmark comparison experiments, indicating that it outperforms the other four competitive algorithms in the existing literature. Finally, the proposed separated elevator layout structure and intelligent algorithm are applied to a real-world reducer workshop containing 20 departments. The results demonstrate the superiority of the proposed layout under safety and low-cost conditions.}, archive = {J_EAAI}, author = {Dan Ji and Zeqiang Zhang and Minjie Zhao and Zongxing He and Zihan Guo}, doi = {10.1016/j.engappai.2025.111010}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111010}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-objective adaptive memetic algorithm and engineering application for a double-floor layout problem with separate human and vehicle transport elevators}, volume = {155}, year = {2025}, }
• (2025). A framework for super-resolution of side-scan sonar images: Combination of variational bayes and regional feature selection. EAAI, 155, 111007. (www)

  @article{ , abstract = {Side-scan sonar is widely used in ocean exploration due to its broad search range and strong identification capabilities. However, the inherent characteristics of acoustic images often result in poor image quality, negatively impacting subsequent downstream tasks’ accuracy. Image super-resolution (SR) technology based on deep learning technology is employed to address this issue. Despite this, existing SR models face two main challenges when applied to side-scan sonar images: (1) less data in side-scan sonar images causes the model overfitting problem; (2) less effective features in side-scan sonar images cause lower efficiency. To overcome these challenges, this paper proposes a deep learning framework that integrates a Bayesian structure with region-based feature selection. First, we introduce a rolling region selection method to extract key features of interest from side-scan sonar images, enhancing efficiency without compromising quality. Additionally, we replace traditional Convolutional Neural Networks (CNN) with Variational Bayes Convolutional Neural Networks (VB-CNN) to perform the SR task, improving generalization on small datasets and mitigating the risk of overfitting. Experiments conducted on the Side-Scan Sonar Visual Object Classes (SSS-VOC) dataset and other datasets demonstrate our proposed approach’s effectiveness through both qualitative and quantitative comparisons.}, archive = {J_EAAI}, author = {Xin Wen and Chensheng Cheng and Lu Li and Feihu Zhang and Guang Pan}, doi = {10.1016/j.engappai.2025.111007}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111007}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A framework for super-resolution of side-scan sonar images: Combination of variational bayes and regional feature selection}, volume = {155}, year = {2025}, }
• (2025). A predefined-time stability control method based on fuzzy compensation for Euler–Lagrange systems and its applications to manipulators. EAAI, 155, 111006. (www)

  @article{ , abstract = {This paper presents an approach to predefined-time terminal sliding mode control (PT-TSMC) for Euler–Lagrange systems (ELSs) with actuator control input saturation, leveraging fuzzy compensation techniques. Initially, the paper focuses on tunable predefined-time stability (PTS), providing adjustable parameters to fine-tune the system’s stability time, thereby enhancing the adaptability of the controller design. Based on this approach, a control input system is formulated to ensure predefined-time stability for Euler–Lagrange systems. Additionally, a strategy is proposed to enhance system performance, including robustness improvement, chattering reduction, and singularity elimination, through the design of an adaptive fuzzy logic system (AFLS). The AFLS estimates unstructured model uncertainties and compounded disturbances, seamlessly integrating them into the control system. Notably, this approach effectively addresses the challenge of handling unknown model data. Finally, comprehensive comparative simulations demonstrate the effectiveness of the proposed method, showcasing its commendable control performance.}, archive = {J_EAAI}, author = {Anh Tuan Vo and Thanh Nguyen Truong and Hee-Jun Kang and Ngoc Hoai An Nguyen}, doi = {10.1016/j.engappai.2025.111006}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111006}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A predefined-time stability control method based on fuzzy compensation for Euler–Lagrange systems and its applications to manipulators}, volume = {155}, year = {2025}, }
• (2025). Guiding reinforcement learning with shaping rewards provided by the vision–language model. EAAI, 155, 111004. (www)

  @article{ , abstract = {Enabling robots to learn manipulation tasks is a practical engineering application, and reinforcement learning is one of the key artificial intelligence methods to achieve it, but reinforcement learning always faces the trade-off between the ease of designing reward functions and the ease of learning from rewards. Reward shaping provides a solution and recent works have designed rewards using images and language descriptions, which is a simple and convenient shaping method for non-expert users. However, some of them only adopt the pretrained model without fine-tuning, while others train the reward model with absolute score labels, which all have difficulties in capturing the spatial relationships within the images, so the performance of reward shaping models is limited. In this work, we propose a novel reward shaping method to generate additional rewards from task descriptions and scene images. We utilize the pretrained vision–language model as backbone for efficient cross-modal information fusion and design a downstream task trained with pair-wise comparison for reward shaping. Extensive experiments are conducted to demonstrate the effectiveness of each component in the method. The approach is validated in the Meta-World environment and the results demonstrate that it outperforms standard reinforcement learning and the existing work in terms of the policy learning efficiency.}, archive = {J_EAAI}, author = {Kaiyi Wang and Yongjia Zhao and Yichen He and Shuling Dai and Ning Zhang and Minghao Yang}, doi = {10.1016/j.engappai.2025.111004}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {111004}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Guiding reinforcement learning with shaping rewards provided by the vision–language model}, volume = {155}, year = {2025}, }
• (2025). A multi-agent deep distribution approximation strategy optimization algorithm with multi-threaded parallel computing mechanism suitable for large-scale and complex urban road networks. EAAI, 155, 110999. (www)

  @article{ , abstract = {In order to alleviate the traffic congestion of urban road networks, a multi-agent hierarchical heterogeneous control framework with cooperation mechanism is proposed in this paper. Secondly, the maximum vehicle delay is introduced as the road congestion level judgment standard, and the road congestion level is discretely divided. Thirdly, a multi-agent joint reward mechanism is designed to enhance the multi-agent collaboration ability and improve the cumulative income of the system. Finally, a multi-agent deep distribution approximation strategy optimization algorithm based on the regional adaptive green wave control mechanism (RAG-MADPPO) is proposed to solve the problems of the long vehicle waiting time in the process of large-scale traffic signal control. Moreover, the multi-threaded parallel computing strategy is proposed to improve the training efficiency of the RAG-MADPPO algorithm. In addition to testing the different strategies on the Manhattan road network scenario, the RAG-MADPPO algorithm is applied to the Oakland road network to verify its effectiveness further. The results show that the proposed RAG-MADPPO algorithm has significant advantages in evaluation indicators and model performance.}, archive = {J_EAAI}, author = {Liyue Fu and Yanhai Xu and Shan Gao}, doi = {10.1016/j.engappai.2025.110999}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110999}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-agent deep distribution approximation strategy optimization algorithm with multi-threaded parallel computing mechanism suitable for large-scale and complex urban road networks}, volume = {155}, year = {2025}, }
• (2025). Cross-database facial expression recognition based on multi-feature representation multi-layer domain adaptive fusion network. EAAI, 155, 110995. (www)

  @article{ , abstract = {In order to extract richer comprehensive facial expression features and reduce domain distribution differences in cross-database experiments, this paper proposes a Multi-feature Representation Multi-layer Domain Adaptive Fusion Network (MRMDAFNet) and applies it to cross-database facial expression recognition research. MRMDAFNet extracts three different types of features including residual network features, Haar features, and Graph Convolutional Network features. Residual network features represent deep feature information, Haar features represent local contrast information, and Graph Convolutional Network features represent topological structure information. The three types of features are aligned between the source domain and the target domain through the application of sub-domain adaptation. Moreover, MRMDAFNet integrates these features using a multi-head attention mechanism and employs global adaptation to align features, and further presents a Multi-layer Domain Adaptation module to achieve domain adaptation at different levels. A variety of transfer learning tasks and extensive cross-database facial expression recognition experiments are implemented on the Extended Cohn-Kanade dataset (CK+), Facial Expression Recognition 2013 dataset (FER2013), and Real-world Affective Faces database (RAF). The experimental results demonstrate the effectiveness of MRMDAFNet, achieving accuracies of 73.58%, 56.59%, and 63.25% in cross-database facial expression recognition experiments with the three databases as target domains respectively. The proposed MRMDAFNet model may be applicable to intelligent human–computer interaction in artificial intelligence.}, archive = {J_EAAI}, author = {Jingjie Yan and Chengkun Du and Biao Li and Xiaoyang Zhou and Ying Liu}, doi = {10.1016/j.engappai.2025.110995}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110995}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-database facial expression recognition based on multi-feature representation multi-layer domain adaptive fusion network}, volume = {155}, year = {2025}, }
• (2025). Cross-domain fault diagnosis of marine diesel engines based on stepwise diffusion and iterative bidirectional optimization. EAAI, 155, 110994. (www)

  @article{ , abstract = {Cross-domain fault diagnosis of marine diesel engines presents significant challenges due to variations in data distribution and the limited availability of labeled fault samples under different operating conditions. To address this, an unsupervised domain-adaptive diagnostic framework is proposed, integrating stepwise diffusion and iterative bidirectional optimization to enhance fault identification. First, the quadratic axial attention transformer introduces a fourth weight in the axial computation to effectively capture the long-range spatio-temporal correlations in the time–frequency representations and strengthen the cross-axis contextual dependence. Next, the domain stepwise diffusion bridge utilizes Markov transform to gradually refine the significant distributional differences across domains into continuous sub-distributions, ensuring a smoother adaptation process. Finally, an iterative bidirectional optimization strategy is proposed to dynamically coordinate the interaction between stepwise diffusion and fault classification, where two complementary learning directions are alternately executed to preserve the semantic integrity of features. Experimental validation on a self-constructed dataset covering multiple operating conditions demonstrates the effectiveness of the proposed approach, achieving 93.80 % average accuracy, 93.75 % precision, and 93.45 % recall. This approach not only breaks through the limitations of existing domain alignment methods and provides a brand new solution for cross-domain fault diagnosis, but also provides a wide range of implications for future research and applications in this field. The code and model are available at: https://github.com/lazyJzr/UDAtask .}, archive = {J_EAAI}, author = {Zhen Zhao and Ziru Jin and Xin Xin and Yutong Fu and Xiaotong Huang and Liang Li and Hongyan Qin and Chong Wei and Yongwei Li and Yang Liu}, doi = {10.1016/j.engappai.2025.110994}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110994}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-domain fault diagnosis of marine diesel engines based on stepwise diffusion and iterative bidirectional optimization}, volume = {155}, year = {2025}, }
• (2025). Layered and scalable image coding with semantic features for human and machine. EAAI, 155, 110982. (www)

  @article{ , abstract = {In recent years, there has been an emergence of numerous edge devices capturing visual data primarily for machine vision analysis, with occasional viewing by humans. While scalable frameworks have been proposed to meet these needs, existing frameworks often suffer from information redundancy between different tasks. To mitigate this redundancy and enhance compression performance, we propose a layered and scalable end-to-end image codec consisting of semantic layer for machine inference and reconstructed layer for image reconstruction, which can be flexibly deployed as needed. Specifically, we introduce a learning-based compression model (LCM) in the semantic layer to extract the semantic features, and obtain the transformed features matching the selected machine vision back-end network through the feature transform module (FTM). Additionally, knowledge distillation is introduced to enhance machine task performance. In the reconstructed layer, the reconstructed features are fused with the semantic features from the semantic layer to achieve human-eye-friendly high perceptual quality image reconstruction. Experimental results show that our proposed image codec saves up to 86.81% bitrate on object detection compared to Versatile Video Coding (VVC). Additionally, the perceptual quality of the reconstructed images outperforms the state-of-the-art image codecs. Code will be released at https://github.com/ombremer/ScalableImageCoding .}, archive = {J_EAAI}, author = {Jiao Wei and Ping An and Shipei Wang and Kunqiang Huang and Chao Yang and Xinpeng Huang}, doi = {10.1016/j.engappai.2025.110982}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110982}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Layered and scalable image coding with semantic features for human and machine}, volume = {155}, year = {2025}, }
• (2025). Textual prompt guided image restoration. EAAI, 155, 110981. (www)

  @article{ , abstract = {Image restoration has been a cutting-edge topic in the academic and industrial fields of computer vision. Since degradation signals are often random and diverse, “all-in-one” models that can do blind image restoration have been concerned in recent years. Early works require training specialized headers and tails to handle each degradation of concern, which are manually cumbersome. Moreover, it lacks naturalness in practical application. Recent works focus on learning visual prompts from data distribution to identify degradation type. However, the prompts employed in most of models are non-text, lacking sufficient emphasis on the importance of human-in-the-loop. In this paper, an effective textual prompt guided image restoration model has been proposed. In this model, task-specific BERT (Bidirectional Encoder Representations from Transformers) is fine-tuned to accurately understand user’s instructions and generate textual prompt guidance. Depth-wise multi-head transposed attentions and gated convolution modules are designed to bridge the gap between textual prompts and visual features. The proposed model has innovatively introduced semantic prompts into low-level visual domain. It highlights the potential to provide a natural, precise, and controllable way to perform image restoration tasks. Extensive experiments have been done on public denoising, dehazing and deraining datasets. The experiment results demonstrate that, compared with popular state-of-the-art methods, the proposed model can obtain much more superior performance, achieving accurate recognition and removal of degradation without increasing model’s complexity. Related source codes and data will be publicly available on github site https://github.com/MoTong-AI-studio/TextPromptIR .}, archive = {J_EAAI}, author = {Qiuhai Yan and Aiwen Jiang and Kang Chen and Long Peng and Qiaosi Yi and Chunjie Zhang}, doi = {10.1016/j.engappai.2025.110981}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110981}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Textual prompt guided image restoration}, volume = {155}, year = {2025}, }
• (2025). Calculating of stomatal index for tomato and lettuce based on you only look once version 8 and improved high-resolution network. EAAI, 155, 110979. (www)

  @article{ , abstract = {The stomatal index (SI), the ratio of stomata to the total of stomata and pavement cells, is a crucial indicator of plant growth status. However, automatic counting of SI in lettuce and tomato presents significant challenges due to the difficulties in accurately segmenting and counting pavement cells with complex morphology. A novel artificial intelligence (AI)-driven architecture for calculating the SI is proposed. The improved High-Resolution Network (Imp_HRNet) integrates a Multi-level Data-dependent Feature Aggregation (MDFA) module to enhance cell segmentation, and a connected domain algorithm was used to count the segmented pavement cells. Compared to the HRNet, the Imp_HRNet demonstrates significant improvements: (1) a 0.18% increase in pavement cell segmentation accuracy; (2) enhanced R 2 values, achieving 0.9991 for lettuce and 0.9985 for tomato; and (3) reduced mean absolute percentage errors ( MAPE ) by 9.45% for lettuce and 4.74% for tomato. The You Only Look Once version 8 (YOLOv8) was employed for stomata counting, achieving R 2 values of 0.9964 for lettuce and 0.9916 for tomato, with corresponding M A P E of 0.83% and 2.53%, respectively. The SI was calculated from stomata and pavement cells counts and achieved R 2 values of 0.9653 for lettuce and 0.9685 for tomato, with M A P E of 2.22% and 3.41%, respectively. These results showed that the proposed AI-driven method enables efficient and precise SI estimation, even for complex pavement cell shapes.}, archive = {J_EAAI}, author = {Yang Xu and Li Du and Qingrui Zhu and Can Wang and Liyuan Zhang and Yaxiao Niu and Qi Li and Danyan Chen and Junhua Zhang}, doi = {10.1016/j.engappai.2025.110979}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110979}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Calculating of stomatal index for tomato and lettuce based on you only look once version 8 and improved high-resolution network}, volume = {155}, year = {2025}, }
• (2025). Deep reinforcement learning of group consciousness for multi-robot pathfinding. EAAI, 155, 110978. (www)

  @article{ , abstract = {Deep Reinforcement Learning (DRL) is highly effective in tackling complex environments through individual decision-making. It offers a novel and powerful approach to multi-robot pathfinding (MRPF). Building on DRL principles, this paper proposes a two-layer collaborative planning framework based on group consciousness (MACCRPF). The framework addresses the unique challenges of MRPF, where robots must not only independently complete their tasks but also coordinate to avoid conflicts during execution. Specifically, the proposed two-layer group consciousness mechanism encompasses: Basic layer group consensus, which emphasizes real-time information sharing and local task scheduling among robots. This layer ensures individual decisions are optimized through dynamic interaction and coordination. Top-layer group consensus, guided by the basic layer consensus, incorporates group strategies and evaluation mechanisms to adaptively adjust pathfinding in complex environments. Additionally, a hierarchical reward mechanism is designed to balance the demands of the two-layer planning framework. This mechanism significantly enhances inter-robot coordination efficiency and task completion rates. Experimental results demonstrate the efficacy of our approach, achieving over 20% improvement in pathfinding success rates compared to state-of-the-art methods. Furthermore, the framework exhibits strong transferability and generalization, maintaining high efficiency across diverse environments. This method provides a technical pathway for efficient collaboration in multi-robot systems.}, archive = {J_EAAI}, author = {Lin Huo and Jianlin Mao and Hongjun San and Ruiqi Li and Shufan Zhang}, doi = {10.1016/j.engappai.2025.110978}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110978}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep reinforcement learning of group consciousness for multi-robot pathfinding}, volume = {155}, year = {2025}, }
• (2025). Deep learning methods for protein representation and function prediction: A comprehensive overview. EAAI, 155, 110977. (www)

  @article{ , abstract = {Deep learning has revolutionized protein function prediction by capturing intricate protein relationships, yet a comprehensive survey of its methodologies remains elusive. In this review, we systematically dissect recent advances by addressing three pivotal questions: (a) which modalities are most critical for various function prediction tasks, (b) which deep learning strategies optimally model these modalities, (c) what common and task-specific challenges persist. We categorize protein data into eight distinct types – from fundamental representations to specialized expert knowledge – and provide an exhaustive analysis of state-of-the-art deep learning models alongside emerging self-supervised learning strategies. Moreover, we compare the evolution of architectures across different modeling paradigms, highlighting their respective strengths and limitations. Our investigation spans over fifteen downstream tasks across five key research areas, including protein function annotation, protein–protein interactions, protein–ligand interactions, mutation effect prediction, and remote homology detection. Finally, we discuss current challenges and propose potential solutions, offering strategic guidance for data selection, methodological innovation, and future research directions in the application of deep learning to protein function prediction.}, archive = {J_EAAI}, author = {Mingqing Wang and Zhiwei Nie and Yonghong He and Athanasios V. Vasilakos and Qiang (Shawn) Cheng and Zhixiang Ren}, doi = {10.1016/j.engappai.2025.110977}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110977}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning methods for protein representation and function prediction: A comprehensive overview}, volume = {155}, year = {2025}, }
• (2025). Weakly-supervised anomaly detection based on adversarial transfer learning. EAAI, 155, 110975. (www)

  @article{ , abstract = {While transfer learning-based anomaly detection techniques are popular, they often focus on known normal samples, neglecting insights from known anomalies. To address this gap, we present a novel weakly-supervised anomaly detection model using adversarial transfer learning. This model aims to achieve high performance despite limited known and unknown samples. Our approach introduces a dual-domain balanced centralization strategy in the anomaly detection module, combining the support vector description algorithm with transfer learning to strengthen hypersphere constraints in a weakly-supervised setting. In the adversarial learning phase for domain adaptation, we introduce a sample-based gradient reversal strategy to effectively integrate normal samples and segregate abnormal ones, optimizing scarce abnormal data utilization. To enhance the use of unknown samples, we employ a weight calculation method based on the Cumulative Distribution Function within the weakly-supervised anomaly detection module, leveraging sample feature distribution patterns for improved anomaly detection precision. We conducted three sets of anomaly detection experiments to validate our methodology across diverse scenarios, encompassing two image recognition and one process industry dataset. The experimental outcomes underscore the superiority of our proposed algorithm over current state-of-the-art methods in anomaly detection, affirming its efficacy and practical utility.}, archive = {J_EAAI}, author = {Jingkai Chi and Zhizhong Mao}, doi = {10.1016/j.engappai.2025.110975}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110975}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weakly-supervised anomaly detection based on adversarial transfer learning}, volume = {155}, year = {2025}, }
• (2025). Time-frequency informed stacked long short-term memory-based generative adversarial network for missing data imputation in sensor networks. EAAI, 155, 110973. (www)

  @article{ , abstract = {To monitor the health condition of civil infrastructures, the continuous acquisition of high-quality sensor data is crucial. However, in harsh environments, data can be lost due to sensor faults, data acquisition system malfunctions, or communication errors. In this study, we propose a stacked long short-term memory (LSTM)-based generative adversarial network (GAN) approach to impute missing acceleration data from faulty sensors. The stacked LSTM-based GAN model is trained by incorporating time-frequency domain information and minimizing both reconstruction and adversarial losses. The reconstruction loss is calculated based on errors in acceleration data and power spectral density (PSD). The GAN’s generator employs a stacked LSTM network to capture long-term dependencies in sequential data. The performance of our proposed approach is compared with multiple state-of-the-art methods based on GANs and variational autoencoders (VAEs). When used as baseline approaches, GANs employ a deep convolutional autoencoder (CAE) in their generators, utilizing skip connections to facilitate the flow of information from the encoder to the decoder. The proposed approach is numerically studied using a three-span continuous bridge model and the American Society of Civil Engineers (ASCE) benchmark model and experimentally validated using the physical ASCE benchmark structure and the Qatar University Grandstand Simulator (QUGS) benchmark structure. Our approach demonstrates promising performance in reconstructing missing data in both the time and frequency domains, showcasing its potential to enhance sensor fault tolerance in safety-critical systems by accurately restoring faulty sensor data.}, archive = {J_EAAI}, author = {Zixin Wang and Malleswari Kachireddy and Tarutal Ghosh Mondal and Wen Tang and Mohammad R. Jahanshahi}, doi = {10.1016/j.engappai.2025.110973}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110973}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Time-frequency informed stacked long short-term memory-based generative adversarial network for missing data imputation in sensor networks}, volume = {155}, year = {2025}, }
• (2025). A deep learning framework for the segmentation and quantitative analysis of pulmonary embolism. EAAI, 155, 110972. (www)

  @article{ , abstract = {Pulmonary Embolism (PE) is a medical emergency arising from an abrupt, occlusive, or partial pulmonary artery blockage in the respiratory system. This is a medical emergency and can be fatal if not detected early. A two-stage architecture is proposed for efficient segmentation and quantification of pulmonary embolism. In the first stage, an attention-based lightweight deep architecture is developed to segment emboli. The Computed Tomography Pulmonary Angiogram (CTPA) image dataset is retrospectively collected from Aster Medcity Multi-speciality Hospital, Kochi. The architecture segmented pulmonary embolisms with an accuracy of 97.00% and specificity of 97.54%. The sensitivity and mean dice scores of the proposed model were 97.15% and 0.96, respectively. The predicted mask is quantified in the second stage by drawing the contour, area, perimeter, and bounding rectangle. Since a diagonal equally bisects a rectangle into two, a diagonal is drawn to measure the extent of emboli. This quantitative analysis enables doctors to assess clot size accurately, evaluating disease progression and medication effectiveness. The model is compared with existing state-of-the-art methodologies. Another public external dataset was utilized to generalize the model. On this dataset, the proposed model segmented emboli with accuracy, specificity, sensitivity, and mean dice scores as 95.28%, 96.00%, 95.62%, and 0.94, respectively. The quantification is performed on this dataset to assess the extent of clots.}, archive = {J_EAAI}, author = {Nitha V.R. and Vinod Chandra S.S. and Praveen Valsalan K. and Arjun S.}, doi = {10.1016/j.engappai.2025.110972}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110972}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep learning framework for the segmentation and quantitative analysis of pulmonary embolism}, volume = {155}, year = {2025}, }
• (2025). TOEformer: Temporal order enhanced transformer for time series forecasting. EAAI, 155, 110965. (www)

  @article{ , abstract = {Transformers have shown excellent performance in time series forecasting, but a recent study points out that existing Transformer-based time series forecasting models cannot preserve temporal order well. To address this problem, we preserve temporal order information by integrating linear and Transformer structures and predicting trend and seasonal data, respectively, which compensates for the inadequacy of the Transformer in this regard by leveraging linear models to preserve the temporal order information of the original sequence. Besides, directly combining the seasonal and trend part prediction results increases the overall prediction error. For this problem, this paper proposes a new loss function, the relative value exponential loss, which reduces the impact of the problem on predictions by constraining the network to generate more seasonal-trend pairs that can offset each other’s errors. Experimental results on six benchmark public datasets show that our temporal order enhanced Transformer ( TOEformer ) can preserve temporal order information and produce better forecasting performance in more than 85% of multivariate prediction setups compared to state-of-the-art time series forecasting models. The proposed relative value exponential loss effectively mitigates the overall error growth problem.}, archive = {J_EAAI}, author = {Qiang Li and Jiwei Qin and Dacheng Wang and Xizhong Qin and Daishun Cui and Jiachen Xie and Dezhi Sun}, doi = {10.1016/j.engappai.2025.110965}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110965}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {TOEformer: Temporal order enhanced transformer for time series forecasting}, volume = {155}, year = {2025}, }
• (2025). A dynamic probabilistic graphical model for mapping tasks to virtual machines in data centers. EAAI, 155, 110963. (www)

  @article{ , abstract = {Efficient mapping of tasks to Virtual Machines is significant in optimizing resource utilization and performance in Virtual Data Centers. In this work, a novel approach for mapping tasks to Virtual Machines by leveraging a bespoke probabilistic graphical model is proposed. The proposed framework maps tasks to Virtual Machines in two phases: constructing a Dynamic Bayesian Network to model scheduling factors and solving the mapping as an inference problem using the Junction Tree algorithm. The model updates the states of Virtual and Physical Machines during mapping, allowing for optimal mapping decisions based on evolving conditions. Experimental findings demonstrate that the proposed model outperforms the state-of-the-art models across various counts, achieving a makespan of 18.87 s, a degree of load imbalance of 0.22, a task guarantee ratio of 83%, an average Service Level Agreement violation rate of 0.32, and an energy consumption of 67 Kilo Joules. Compared to the strongest meta-heuristic baseline, the proposed model reduces makespan by 6.89%, decreases load imbalance by 11.08%, lowers Service Level Agreement violations by 12.44%, reduces energy consumption by 2.50%, and improves the task guarantee ratio by 4.28%.}, archive = {J_EAAI}, author = {Korrapati Sindhu and Karthick Seshadri}, doi = {10.1016/j.engappai.2025.110963}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110963}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dynamic probabilistic graphical model for mapping tasks to virtual machines in data centers}, volume = {155}, year = {2025}, }
• (2025). Demultiplexing and electron-muon identification in different micropattern readout planes with common U-net approach. EAAI, 155, 110962. (www)

  @article{ , abstract = {Micropattern Gaseous Detectors, like Micromegas, are used in particle physics to detect charged particles like muons, through the ionization of the gas they contain. Micromegas detectors provide good spatial resolution that allows doing muography, an imaging method using the natural muon flux to study very opaque objects like pyramids, volcanoes or nuclear reactors. One of those detectors, called the multiplexed Micromegas detector, allows to significantly reduce the volume and cost of the electronics. Multiplexed Micromegas need less electronics but produce ambiguous data. Nevertheless, a demultiplexing step is required: a dedicated analysis finds the true position of the particle(s) on the detector. In this work, we propose a new demultiplexing method, with a denoising approach using a U-Net architecture. We show that the same method allows to analyze two types of Micromegas detectors: a one-dimensional (1D) detector with strips and a two-dimensional (2D) detector with pixels in a Time Projection Chamber. We demonstrate that only a few changes must be made to adapt to the 1D and 2D analysis. In the 1D case, data from Khufu’s pyramid muography was re-analyzed, resulting in an image that better matched the pyramid’s simulation. This makes this common method, with a U-Net, easily adaptable for a wide range of other high-granularity particle detectors, even other than gaseous ones, and for different sizes and dimensionalities. Moreover, we show that U-Net’s capabilities also allow distinguishing muon tracks from electron tracks in the 2D detector.}, archive = {J_EAAI}, author = {Baptiste Lefevre and Etienne Gozillon and David Attié and Héctor Gomez and Irakli Mandjavidze and Philippe Mas}, doi = {10.1016/j.engappai.2025.110962}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110962}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Demultiplexing and electron-muon identification in different micropattern readout planes with common U-net approach}, volume = {155}, year = {2025}, }
• (2025). Machine learning applications for risk assessment in maritime transport: Current status and future directions. EAAI, 155, 110959. (www)

  @article{ , abstract = {As the maritime transportation system is inherently complex and vulnerable to potential hazards, it is critical to conduct risk assessments via applicable methodologies. Recently, machine learning (ML) algorithms have attracted tremendous attention due to their ability to analyze risks effectively. Nevertheless, there is a lack of a systematic summarization of ML applications in maritime transport risk assessment (MTRA). Hence, this review aims to encapsulate the current status, issues, considerations, and future directions of ML applications using the systematic reviews and meta-analyses method. In particular, the status is summarized from the following three dimensions: advantages, disadvantages, and corresponding applications. Moreover, several issues are recognized, including dataset processing and methods utilization, and considerations from the perspective of sensitivity analysis and evaluation methods. Regarding future directions, promising opportunities in terms of data and method improvements are identified. Overall, this review contributes to MTRA by presenting the existing research status with a framework and providing suggestions on model selection and method improvement for future research.}, archive = {J_EAAI}, author = {Yuqing Lin and Xue Li and Kum Fai Yuen}, doi = {10.1016/j.engappai.2025.110959}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110959}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning applications for risk assessment in maritime transport: Current status and future directions}, volume = {155}, year = {2025}, }
• (2025). Learning multi-level graph attentional representation for thermal infrared object tracking. EAAI, 155, 110957. (www)

  @article{ , abstract = {Thermal infrared (TIR) object tracking is a fundamental task in computer vision that is not affected by changes in lighting conditions. It performs better than visible light trackers in extreme environments such as nighttime, heavy rain, haze, and sandstorms. However, TIR object tracking also faces challenges such as occlusion, thermal crossover, motion blur, and similarity interference. Unlike visual tracking, TIR images lack color information and texture features. These factors make it challenging to learn detailed and shape features of the targets, making it hard to distinguish between targets and interference effectively. In this study, we propose a graph-based deep learning model, SiamMLGR , within the Siamese framework for stable TIR object tracking to address these issues. Specifically, to extract more fine-grained features of TIR targets, we propose a multiple graph attention module (MGAM) to replace the global matching information transmission method in the Siamese framework. This module constructs a graph structure to establish local and global connections between the target and the search area. Furthermore, to retain more of the features learned by the MGAM, we propose a spatial graph convolutional module (SGCM), which uses an explicit graph adjacency matrix to propagate information between the attention graphs. Additionally, we incorporate large-scale datasets from the visual tracking field into the model training process. By mixing these with TIR datasets, we address the sample imbalance issue present in pure TIR datasets. Extensive experimental results indicate that the proposed method achieves state-of-the-art performance.}, archive = {J_EAAI}, author = {Peng Gao and Shi-Min Li and Fei Wang and Hamido Fujita and Hanan Aljuaid and Ru-Yue Yuan}, doi = {10.1016/j.engappai.2025.110957}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110957}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Learning multi-level graph attentional representation for thermal infrared object tracking}, volume = {155}, year = {2025}, }
• (2025). Data collection in deep reinforcement learning-enhanced reconfigurable intelligent surface-assisted wireless networks. EAAI, 155, 110952. (www)

  @article{ , abstract = {In the evolving sixth generation (6G) landscape, the integration of reconfigurable intelligent surfaces (RIS) with unmanned aerial vehicles (UAVs) offers a revolutionary opportunity to optimize data collection in the Internet of things (IoT) through deep reinforcement learning (DRL) and improve energy efficiency and network performance. This paper aims to study how reconfigurable intelligent surfaces and deep reinforcement learning can help increase throughput and energy efficiency in unmanned aerial vehicle-controlled Internet of things networks. The focus is on improving the capabilities of unmanned aerial vehicles to efficiently collect data in different regions and ensure safe landings. Divided into two phases, the study first improves the directional capacity and flexibility of unmanned aerial vehicles and then evaluates the integration of reconfigurable intelligent surface technology. We introduce two deep reinforcement learning models, namely the directional capacity and flexible reconnaissance (DCFR) model and the reconfigurable intelligent surface model, and compare them with a benchmark model. We found significant improvements in communication and data collection efficiency. The simulation results show an 8.18% increase in data collection performance and a 6.92% increase in collected data per unit energy when using reconfigurable intelligent surfaces, with a 10.59% increase in collection performance and a 22.64% increase in energy efficiency. Furthermore, an unmanned aerial vehicle optimized with the double deep Q-network algorithm effectively identified optimal trajectories for data collection, confirming the significant benefits of reconfigurable intelligent surfaces in unmanned aerial vehicle-controlled Internet of things networks.}, archive = {J_EAAI}, author = {Idris Ertas and Halil Yetgin}, doi = {10.1016/j.engappai.2025.110952}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110952}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data collection in deep reinforcement learning-enhanced reconfigurable intelligent surface-assisted wireless networks}, volume = {155}, year = {2025}, }
• (2025). Adaptive coordinated motion planning for lower limb exoskeleton robots with a robotic walker. EAAI, 155, 110949. (www)

  @article{ , abstract = {Repetitive gait training is helpful for patients with gait disabilities and can be effectively facilitated by lower limb exoskeleton robots (LLEs). At the early stages of rehabilitation, the deployment of a mobile robotic walker to complement exoskeleton-assisted gait training is indispensable, particularly for patients with diminished muscle strength. It is a critical issue to coordinated control the exoskeleton robot and the robotic walker to achieve a natural walking posture due to the variations of gait patterns with different walking speeds. In this paper, a novel Adaptive Coordinated Motion Planning (ACMP) approach is proposed to tackle the coordinated control problem of the LLEs with a robotic walker, which comprises of three parts: the gait patterns generation for LLEs, the optimization for adaption to the desired walking speed, and the coordinated control for both the exoskeleton robot and the robotic walker. The main contribution of this paper is twofold: a new knee-stretched gait patterns generation approach with the given foot locations; based on the reduced order dynamics model for the human–exoskeleton system, an optimization problem is constructed to generate reference joint angles for the human–exoskeleton–walker system and adapt to different walking speeds. Experimental validation was conducted on the CoppeliaSim simulation platform. Results demonstrate that the proposed approach achieves coordinated motion control across a wide speed range (0–0.8 m/s), with the mean displacement tracking error of the center of mass constrained within 0.01 m. This ensures accurate locomotion tracking of the LLEs while maintaining natural walking postures. Compared to non-adaptive baselines, the proposed method achieves a tracking accuracy improvement of over 92%, highlighting its effectiveness in dynamic human–robot coordination tasks.}, archive = {J_EAAI}, author = {Chaobin Zou and Yi Ren and Rui Huang and Xinhao Zhang and Jingting Zhang and Zhinan Peng and Hong Cheng}, doi = {10.1016/j.engappai.2025.110949}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110949}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive coordinated motion planning for lower limb exoskeleton robots with a robotic walker}, volume = {155}, year = {2025}, }
• (2025). Curriculum learning-based missile guidance law for intercepting maneuvering targets with high-speed. EAAI, 155, 110948. (www)

  @article{ , abstract = {In this paper, a missile guidance strategy is designed by using curriculum-based reinforcement learning for intercepting high-speed maneuvering targets. To enhance the exploration capabilities of agents within the state space, an exploration ability index is introduced. The proposed index, integrated with the entropy of the Soft Actor-Critic algorithm, dynamically adjusts the training process to augment exploration. In the proposed curriculum-based reinforcement learning, multiple sub-tasks are designed to progressively train the agent from simple to complex scenarios. This incremental learning framework enables the agent to efficiently master the task, expediting the overall training process. Through comprehensive simulations, we validate the efficacy of the proposed guidance law. Comparing the results obtained in other reinforcement learning methods, our findings highlight the advantages of curriculum learning in enhancing the efficiency and effectiveness of missile guidance systems.}, archive = {J_EAAI}, author = {He Tao and Zhengxuan Jia and Bing Zhu and Tingyu Lin}, doi = {10.1016/j.engappai.2025.110948}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110948}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Curriculum learning-based missile guidance law for intercepting maneuvering targets with high-speed}, volume = {155}, year = {2025}, }
• (2025). A sample average approximation-based approach for the last mile delivery and pickup problem with load-dependent travel time under uncertainties. EAAI, 155, 110944. (www)

  @article{ , abstract = {This paper studies an electric cargo bicycle last mile Delivery and Pickup Problem with Load-Dependent Travel Time under Uncertainties (DPLTTU), in which the travel speed depends on the gradient of the road and the load of the electric cargo bicycle. Moreover, the uncertainties in actual transportation and service processes are considered in the DPLTTU. We formulate the studied DPLTTU as a Stochastic Programming with Recourse (SPR) model. To solve this SPR model, the Sample Average Approximation (SAA)-based algorithms are proposed, where a Simulated Annealing (SA) algorithm and an Adaptive Large Neighborhood Search (ALNS) algorithm are proposed to solve the integer programming model converted by SPR model. An effective parallel computing strategy is applied into speeding up the solving processed of SAA-based algorithms. In order to validate the efficiency and effectiveness of the developed SPR model and SAA-based algorithms, a large number of numerical experiments are conducted. The experimental results highlight the performance of the developed model and algorithms, and demonstrate that the proposed model and algorithms can generate a more risk-resistant solution when uncertain elements are taken into account, which will help logistics companies make suitable decisions when addressing the issue of last mile delivery and pickup.}, archive = {J_EAAI}, author = {Hongyuan Luo and Deyun Wang and Yanhui Li and Xinyuan Lu and Mingyun Gao and Hao Chen}, doi = {10.1016/j.engappai.2025.110944}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110944}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A sample average approximation-based approach for the last mile delivery and pickup problem with load-dependent travel time under uncertainties}, volume = {155}, year = {2025}, }
• (2025). Time-delayed fractional grey bernoulli model with independent fractional orders for fossil energy consumption forecasting. EAAI, 155, 110942. (www)

  @article{ , abstract = {Fossil fuels serve as the primary energy source in the global energy landscape. Through an in-depth understanding and forecasting of fossil fuel consumption, it becomes possible to address energy and environmental challenges more effectively. This study contributes to artificial intelligence by proposing a novel time-delayed fractional grey Bernoulli model with independent fractional orders, which leverages the structural properties of the Bernoulli equation, the cumulative nature of fractional orders, and the driving influence of time-delay term. The model parameters are optimized using the particle swarm optimization algorithm, enhancing its adaptability and accuracy. The engineering application of this model focuses on forecasting fossil fuel consumption, a critical challenge in energy and environmental engineering. Using real datasets from 2000 to 2022, the proposed model is applied to predict the consumption of natural gas, coal and oil in the Middle East and North America, alongside ten benchmark models. The results demonstrate the superior performance of the proposed model, achieving the mean absolute percentage errors of 2.632991%, 5.793513%, 5.432220% and 2.816756% across four case studies, significantly outperforming other models. These findings highlight the potential of the proposed model as a robust and reliable decision-support tool in energy engineering.}, archive = {J_EAAI}, author = {Xin Ma and Qingping He and Wanpeng Li and Wenqing Wu}, doi = {10.1016/j.engappai.2025.110942}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110942}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Time-delayed fractional grey bernoulli model with independent fractional orders for fossil energy consumption forecasting}, volume = {155}, year = {2025}, }
• (2025). DynLeafNet: A dynamic lightweight architecture for plant disease classification using dynamic residual network with explainable artificial intelligence. EAAI, 155, 110937. (www)

  @article{ , abstract = {Agriculture is a vital sector contributing to food security and economic growth, but it faces significant hurdles in food production due to declining crop yields, with plant diseases being a key factor. Moreover, early detection of plant diseases remains a major challenge due to class imbalance in datasets and the difficulty in identifying initial symptoms. To tackle these challenges, this study proposes a novel lightweight dynamic model called DynLeafNet, a deep learning-based architecture specifically designed for efficient and accurate recognition of plant disease. DynLeafNet leverages residual Convolutional Neural Networks (CNNs) enhanced with advanced dynamic techniques such as Learned Group Convolutions (LGConv), Early Exiting, and Grid Search to optimize performance and computational efficiency. Moreover, the proposed model uses data augmentation to address the Class Imbalance Problem (CIP) by calculating class weights in the datasets. The efficacy of the DynLeafNet model is validated using the publicly available PlantVillage and Rice Leaf Disease datasets, where it demonstrated superior accuracy and reduced computational cost (calculated in terms of Floating Point Operations per Second (FLOPs)) compared to existing state-of-the-art models. For the PlantVillage dataset, DynLeafNet achieved a training accuracy of 99.34%, a testing accuracy of 98.33%, and a computational complexity of 17.64M FLOPs. For the Rice Leaf Disease dataset, the model achieved a training accuracy of 97.81%, a testing accuracy of 95.64%, and a computational complexity of 35.07M FLOPs. In addition, single image prediction is performed using Gradient Weighted Class Activation Mapping ++ (Grad-CAM ++), to visualize the model’s focus areas and enhance interpretability.}, archive = {J_EAAI}, author = {Preeti Raj Verma and Deepika Pantola and Navneet Pratap Singh}, doi = {10.1016/j.engappai.2025.110937}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110937}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DynLeafNet: A dynamic lightweight architecture for plant disease classification using dynamic residual network with explainable artificial intelligence}, volume = {155}, year = {2025}, }
• (2025). Model-based contextual reinforcement learning for robotic cooperative manipulation. EAAI, 155, 110919. (www)

  @article{ , abstract = {In large-scale manufacturing and automation, cooperative robotic manipulation requires lower-level controllers to generate specific position or force commands based on real-time information. However, adapting these controllers for different objects or trajectories often demands a complete re-learning process. Contextual Reinforcement Learning (CRL) offers a method to generalize controllers across various contexts, yet it frequently faces instability and inefficiency due to uncertain contact dynamics and external disturbances, especially in real-world industrial environments. To enhance process efficiency and sustainability in robotic systems, a Model-Based Contextual Reinforcement Learning (MCRL) approach is proposed, leveraging metaheuristic solutions to optimize cooperative manipulation. The approach includes a parameterized lower-level control policy, designed using projected inverse dynamics for decoupled control in both constrained and unconstrained subspaces. A probabilistic forward model further enhances the cooperative manipulation process by calculating expected returns and mitigating risk-sensitive biases. Through these optimizations, the MCRL framework efficiently generalizes lower-level policies and improves operational performance. Evaluation on a multi-robot simulation system and a physical dual-arm platform demonstrates the approach’s effectiveness in achieving high-quality, sustainable policies for cooperative manipulation tasks, aligning with the goals of improved efficiency and resource utilization in manufacturing.}, archive = {J_EAAI}, author = {Wenrui Wang and Qirong Tang and Hao Yang and Chao Yang and Baoping Ma and Shuqing Wang and Rongfu Lin}, doi = {10.1016/j.engappai.2025.110919}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110919}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Model-based contextual reinforcement learning for robotic cooperative manipulation}, volume = {155}, year = {2025}, }
• (2025). Lightweight transformer-driven multi-scale trapezoidal attention network for saliency detection. EAAI, 155, 110917. (www)

  @article{ , abstract = {State-of-the-art (SOTA) methods in salient object detection (SOD) often struggle to balance computational efficiency with performance. They encounter challenges in capturing complex features due to insufficient attention mechanisms and ineffective integration of multi-scale features. While deep convolutional neural networks (CNNs) provide lightweight architectures, they lack an understanding of global contexts. Transformers excel at capturing global context, but they require considerable resources, limiting their use on constrained platforms. To overcome these limitations, we present a novel saliency detection (SD) network that utilizes a pyramid vision transformer backbone to efficiently extract multi-scale features. The intermediate multi-scale features are refined through contextual feature refinement blocks (CFRBs) with dilated convolutions, which capture rich contextual information at each scale and enhance feature representation. Furthermore, these features are passed on to our newly proposed trapezoidal attention module (TAM), which integrates effective attention blocks. The Adaptive Spatial Coordinate Attention (ASCA) blocks are employed to highlight the importance of spatial locations in the initial two feature maps. As the high-level refined features require channel adjustments, we introduce Compact Channel Gate (CCG) blocks that adaptively recalibrate channel-wise feature responses. Additionally, the medium-scale two feature maps are processed through Feature-Aware Multi-Head Attention (FAMHA) blocks to capture long-range dependencies and global context. The extracted features are progressively upsampled and concatenated to create a high resolution saliency map, which is used for the final predictions. Extensive empirical analysis on six benchmark SD datasets demonstrates the effectiveness of our network, surpassing over 26 SOTA SD methods while maintaining a lightweight architecture. Code, qualitative results, and trained models will be available at: https://github.com/TalhaUsman-ZERO/TRSNet .}, archive = {J_EAAI}, author = {Muhammad Talha Usman and Habib Khan and Imad Rida and JaKeoung Koo}, doi = {10.1016/j.engappai.2025.110917}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110917}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight transformer-driven multi-scale trapezoidal attention network for saliency detection}, volume = {155}, year = {2025}, }
• (2025). Progressive enhancement dehazing for object detection in extreme weather. EAAI, 155, 110903. (www)

  @article{ , abstract = {Existing object detection technologies have made significant advances in perception systems for autonomous driving. However, accurately detecting objects in extreme weather conditions, especially fog, remains a significant challenge. On the one hand, current methods struggle to balance image enhancement and object detection, leading to the neglect of essential information that could improve detection accuracy. On the other hand, the lack of mature datasets in this area limits many studies to synthetic fog data generated from the prior-based atmospheric scattering model, which inevitably restricts further performance improvements. To address these challenges, we propose the P rogressive E nhancement D ehazing Y ou O nly L ook O nce (PED-YOLO) method, which processes images in real time under foggy conditions to improve object detection. We design a novel progressive image processing module that follows a unique paradigm of progressive supervised learning, which gradually processes the image from small size to large size, to effectively overcome the challenge of processing large-sized images at once. Moreover, we develop a small convolutional network module that focuses on each channel of the image and enables more accurate adaptive prediction of the parameters of the filters. In addition, we develop a novel style transfer model to generate simulated fog images that more closely resemble real fog images and train them together to bring the synthetic fog domain closer to the real fog domain and improve the generalizability. We evaluate our method extensively on several popular datasets, and the experimental results show the superior performance of PED-YOLO, highlighting its potential to advance autonomous driving.}, archive = {J_EAAI}, author = {Zhiying Li and Junhao Wu and Shuyuan Lin and Zheng Wang and Xiaobo Jin and Guanggang Geng and Feiran Huang and Jian Weng}, doi = {10.1016/j.engappai.2025.110903}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110903}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Progressive enhancement dehazing for object detection in extreme weather}, volume = {155}, year = {2025}, }
• (2025). Hybrid prediction model for the friction performance of asphalt pavements by combining multi-scale decomposition and transformer network. EAAI, 155, 110896. (www)

  @article{ , abstract = {The monitoring of asphalt pavements is key to ensuring traffic safety in modern road environments. This study introduces a hybrid model combining two-dimensional (2D) Fourier transform and a transformer network to predict the skid resistance of asphalt pavements. Precise surface texture data were collected using the Laser Scanner-40s (LS-40s) three-dimensional (3D) laser texture measurement device, and skid resistance tests were conducted using the British pendulum tester. The collected data were denoised using bilateral filtering and the median absolute deviation method. Random rotation cropping was used for texture segmentation, resulting in 5650 sets of 3D texture data. A 2D Fourier transform was applied for the multi-scale decomposition of the surface textures. Three network architectures—Visual Geometry Group 16 (VGG16), Residual Network 34 (Resnet34), and Transformer—were employed for feature extraction and then compared. The transformer network outperformed the other two network architectures, with an average absolute error of 3.7178 and a coefficient of determination of 0.9461. The proposed method enables a rapid, accurate, non-contact prediction of the skid resistance of asphalt pavements, suitable for large-scale pavement performance assessments.}, archive = {J_EAAI}, author = {You Zhan and Zilong Nie and Xiuquan Lin and Allen A. Zhang and Changfa Ai}, doi = {10.1016/j.engappai.2025.110896}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110896}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid prediction model for the friction performance of asphalt pavements by combining multi-scale decomposition and transformer network}, volume = {155}, year = {2025}, }
• (2025). A lightweight network for category-level open-vocabulary object pose estimation with enhanced cross implicit space transformation. EAAI, 155, 110890. (www)

  @article{ , abstract = {Category-level object pose estimation has emerged as a crucial task in fields such as robotic manipulation, augmented reality, and autonomous driving. Unlike instance-level and unseen-level estimation, category-level estimation does not rely on computer-aided design models for guidance, making it highly applicable to real-world scenarios. However, current methods face some challenges, including high computational complexity and poor generalization across object instances with diverse features within the same category. In this paper, we introduce a lightweight network designed specifically for prior-free category-level pose estimation. The network addresses existing limitations through a plug-and-play lightweight color and geometric feature extraction module optimized for real-time pose tracking, an enhanced cross-implicit space transformation module utilizing a dual-branch bidirectional cross-attention mechanism to capture feature variations effectively, and a category-level open-vocabulary pose estimation inference network that integrates the Segment Anything model and Vision Language model for handling novel objects. Our method significantly reduces the complexity of the model and achieves a inference speed of 120.1 Frames Per Second (3-5 times faster than existing methods), while maintaining high accuracy, achieving state-of-the-art results on the benchmark datasets. Furthermore, robotic grasping experiments conducted in unrestricted semantic understanding environments demonstrate the network’s capability to segment and perform pose regression on objects in any scene without the need for computer-aided design models. This robustness and efficiency make the network deployable in real-time scenarios and on resource-limited devices, thus advancing the broader field of pose estimation applications.}, archive = {J_EAAI}, author = {Pihong Hou and Yongfang Zhang and Wei Zhou and Boxuan Ye and Yi Wu}, doi = {10.1016/j.engappai.2025.110890}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110890}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight network for category-level open-vocabulary object pose estimation with enhanced cross implicit space transformation}, volume = {155}, year = {2025}, }
• (2025). Flow regime identification from acoustic sensors in pipe flow using deep neural networks. EAAI, 155, 110885. (www)

  @article{ , abstract = {In contemporary industrial processes, precise flow signal measurement is crucial for several reasons, such as enabling condition-based monitoring of pipeline health, preventing damage to expensive materials/equipment, minimizing environmental pollution, and ensuring the safety of operational personnel. In this work, a novel method for flow regime identification using advanced signal denoising techniques and deep neural networks is proposed. The time series data is collected from passive acoustic sensors that capture flow-generated acoustic signals on pipes within an experimental flow loop. The data is obtained under various flow conditions, with particular emphasis on slug flows, given their inherently high-noise nature. The denoising methods used here are an amalgamation of whitening procedures, autocorrelation filtering, bandpass filtering, and highly sophisticated deep neural networks. Notably, the first three methods are ingeniously adapted to facilitate the training of deep neural networks, eliminating the necessity for field data during the training process. Subsequently, the application of deep neural networks proves highly effective in refining signals associated with slug flows amidst extreme noise. This refinement allows for accurate determination of the slug regime type and its associated frequencies. Our approach to deep neural networks is particularly intriguing, as it leverages time–frequency segments for data denoising and incorporates artifical noise during training. This stands in contrast to existing neural network methodologies that predominantly focus on mapping inputs to outputs. Consequently, this technique introduces a cheap and fast method for flow lines monitoring where high levels of process noise are present, compared to existing extensive other approaches such as fiber optics monitoring or intrusive sensing. The experimental results demonstrate that the proposed technique benchmarking advanced methods, including the independent component analysis (ICA) and the eigenvalue decomposition (EGD) based approaches. In the presence of strong random noise, the presented methodology achieved an SNR of 30.2 dB, whereas the EGD and ICA methods attained SNR values of 15.3 dB and 17.5 dB, respectively, highlighting the superiority of the proposed approach.}, archive = {J_EAAI}, author = {Naveed Iqbal and Azzedine Zerguine and Mohamed Nabil Noui-Mehidi and Mohamed Larbi Zeghlache and Abdulmajid Lawal and Ali Al-Shaikhi}, doi = {10.1016/j.engappai.2025.110885}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110885}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Flow regime identification from acoustic sensors in pipe flow using deep neural networks}, volume = {155}, year = {2025}, }
• (2025). Rehabilitation exercise score prediction through integration of spatial feature with graph structural learning. EAAI, 155, 110833. (www)

  @article{ , abstract = {Automated exercise assessment is crucial for streamlining rehabilitation processes, especially for individuals with limited access to therapists or facing high expenses. This research addresses these challenges by proposing an innovative approach that combines the strengths of two dimensional Convolutional Neural Networks (CNN) and Graph Neural Networks (GNN). Traditional models, such as Long Short-Term Memory (LSTM)-CNN architectures, often struggle to effectively capture both spatial and temporal dependencies within dynamic exercise data. By integrating CNN, which excel at extracting spatial features from image data, with GNN, which enhance the model’s ability to capture intricate spatial relationships, we offer a robust solution to these limitations. The primary contribution of this research lies in integrating the structural patterns of graph networks with the visual capabilities of CNN, specifically designed for predicting exercise scores. This proposed Convolutional-Graphical Neural Network (C-GNN) architecture demonstrates promising results, achieving a mean absolute deviation of 0.132 on the KInematic Assessment of MOvement and Clinical Scores for Remote Monitoring of Physical REhabilitation (KIMORE) dataset and 0.029 on the University of Idaho - Physical Rehabilitation Movements Data (UI-PRMD) dataset. These findings underscore the C-GNN’s effectiveness in accurately assessing exercise performance across diverse rehabilitation contexts. These results highlight the C-GNN architecture’s potential for improving the accessibility and efficiency of rehabilitation processes, particularly in resource-constrained settings. In future work, we aim to improve model interpretability by incorporating techniques like SHapley Additive exPlanations (SHAP) or Local Interpretable Model-Agnostic Explanations (LIME), which can provide deeper insights into the decision-making process.}, archive = {J_EAAI}, author = {Md. Johir Raihan and Md. Golam Mustafa and Ali Saleh Alammary and Abdullah-Al Nahid}, doi = {10.1016/j.engappai.2025.110833}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110833}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rehabilitation exercise score prediction through integration of spatial feature with graph structural learning}, volume = {155}, year = {2025}, }
• (2025). Bridging building information systems: A parameter-efficient semantic approach to construction data interoperability. EAAI, 155, 110680. (www)

  @article{ , abstract = {This study investigates the application of semantic text similarity embedding models to address data interoperability challenges in the construction and civil engineering sectors. Research focuses on improving embedding models to produce domain-specific embeddings that precisely capture the complex terminology and relationships among various building systems, including Building Information Modeling, Building Energy Modeling, and Building Management Systems. In this study, we explored the use of embedding models to enhance semantic text similarity applications within the construction industry. The aim is to augment general-purpose language models by training them on a comprehensive construction and civil engineering data corpus to promote enhanced data integration and interoperability among diverse sources. Our methodology includes a comprehensive data collection and preprocessing pipeline, followed by a multi-stage fine-tuning approach that employs continuous pre-training, task-specific fine-tuning, and parameter-efficient adaptation algorithms. We assess the effectiveness of our models through intrinsic evaluations, including semantic similarity measures and domain-specific benchmarks. The test results demonstrate that our domain-adapted models, especially the Weight-Decomposed Low-Rank Adaptation method, substantially surpass the fundamental language models in domain-specific tasks, including understanding technical terms and data alignment across domains. The results of the models show significant performance enhancements with minimal resource investment, making them appropriate for practical use. Our research demonstrates that domain-specific embedding models can substantially improve data integration and enable more advanced, data-driven applications in the building industry.}, archive = {J_EAAI}, author = {Hamed Asadollahi and Rani El Meouche and Zhiyu Zheng and Mojtaba Eslahi and Elham Farazdaghi}, doi = {10.1016/j.engappai.2025.110680}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110680}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bridging building information systems: A parameter-efficient semantic approach to construction data interoperability}, volume = {155}, year = {2025}, }
• (2025). Towards explainable artificial intelligence in machine learning: A study on efficient perturbation-based explanations. EAAI, 155, 110664. (www)

  @article{ , abstract = {Explainable Artificial Intelligence (XAI) is critical for validating and trusting the decisions made by Machine Learning (ML) models, especially in high-stakes domains such as healthcare and finance. However, existing XAI methods often face significant computational challenges. To address this gap, this paper introduces a novel Perturbation-Based Explanation (PeBEx) method and comprehensively evaluates it versus local interpretable model-agnostic explanation approach (LIME) and SHapley Additive exPlanations (SHAP) across multiple datasets and ML models to assess explanation quality and computational efficiency. PeBEx leverages perturbation-based strategies to systematically alter input features and observe changes in model predictions to determine feature importance. This method not only offers superior computational efficiency, leading to scalability and efficiency for complex models on large datasets. Through testing on both synthetic and public datasets using eight ML models, we uncover the relative strengths and limitations of each XAI method in terms of explanation accuracy, fidelity, and computational demands. Our results show that while SHAP and LIME provide detailed explanations, they often suffer from high computational costs, particularly with complex models like Multi-Layer Perceptron (MLP). Conversely, PeBEx demonstrates superior efficiency and scalability, making it particularly suitable for applications that require rapid response times without compromising explanation quality. We conclude by proposing potential enhancements for PeBEx, including its adoption in a wider array of large-scale models. This study not only advances our understanding of the computational aspects of XAI but also proposes PeBEx as a viable solution for improving the efficiency, scalability, and applicability of explainability in ML.}, archive = {J_EAAI}, author = {Ismael Gómez-Talal and Mana Azizsoltani and Luis Bote-Curiel and José Luis Rojo-Álvarez and Ashok Singh}, doi = {10.1016/j.engappai.2025.110664}, journal = {Engineering Applications of Artificial Intelligence}, month = {9}, pages = {110664}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards explainable artificial intelligence in machine learning: A study on efficient perturbation-based explanations}, volume = {155}, year = {2025}, }
• (2025). Using machine and deep learning models, on-chain data, and technical analysis for predicting bitcoin price direction and magnitude. EAAI, 154, 111086. (www)

  @article{ , abstract = {This study investigates the performance of Machine Learning (ML) and Deep Learning (DL) models in predicting both Bitcoin price magnitude and movements using a unique combination of feature selection, input types, and output types. The model inputs are Bitcoin price, on-chain metrics, and technical analysis indicators (TA). The model outputs are Bitcoin price direction and magnitude. The ML and DL models developed in this paper include Support Vector Machines (SVM), Random Forest, Gradient Boosting Machines, Long Short-Term Memory, Convolutional Neural Network-Long Short-Term Memory, Gated Recurrent Units, Temporal Convolutional Network, and Long and Short-Term Time-series Network. An important aspect of this study was the use of the Boruta feature selection method to select the most relevant subset of features. When predicting price movement, the findings revealed that the SVM model achieves the best performance with an accuracy of 83% and F1-Score of 82%. When predicting price magnitude, the model achieved the lowest RMSE (1,531.3) and an impressive R 2 (0.9856). Generally, the DL models gave competitive performance for the classification tasks, but they mostly performed poorly with regression tasks. It was discovered that the inclusion of the Boruta feature selection method significantly improved performance of the models. On-chain data also significantly improved model performance, especially for classification tasks. Finally, the study assessed the models’ profitability through backtesting, and Boruta-SVM emerged as the most profitable model. The classification models mostly obtained positive returns, while the regression models yielded losses or marginal gains. This study emphasizes the importance of feature selection, model choice, and target type in Bitcoin price prediction. Analysts can use these insights to improve investment decisions, thereby maximizing returns.}, archive = {J_EAAI}, author = {Oluwadamilare Omole and David Enke}, doi = {10.1016/j.engappai.2025.111086}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111086}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Using machine and deep learning models, on-chain data, and technical analysis for predicting bitcoin price direction and magnitude}, volume = {154}, year = {2025}, }
• (2025). Rolling bearings remaining useful life estimation using digital twin and physics-informed methods with uncertainty quantification. EAAI, 154, 111070. (www)

  @article{ , abstract = {This paper presents an innovative framework for monitoring the degradation process of rolling bearings, predicting their remaining service life and quantifying the associated uncertainties. By integrating digital twin technology with artificial intelligence, the framework significantly enhances the accuracy and reliability of remaining service life predictions for rolling bearings. First, an advanced dynamic model of the rolling bearing is developed, forming the foundation for the digital twin to accurately simulate the bearing's operating conditions. Next, a dynamic update method for the digital twin model is proposed, based on a backpropagation neural network, which ensuring real-time adaptation to changes in actual bearing conditions. Building on this, an innovative uncertainty-quantifying physics-informed neural network is introduced, which leverages a dropout technique. This network takes the output of the digital twin model as input to predict the remaining service life of the bearing while quantifying the uncertainty. The performance of the proposed framework is evaluated through experiments, demonstrating its ability to accurately reflect the bearing's operational state and predict its remaining service life with quantified uncertainty. The experimental results show that the mean absolute error and root mean square error of this framework are 0.071 and 0.084, respectively, which are both significantly better than the 0.100 and 0.118 observed for other methods, fully validating the superiority of this method.}, archive = {J_EAAI}, author = {Fengjin Gong and Ping Ma and Hongli Zhang and Cong Wang and Xinkai Li and Yinfei Wu}, doi = {10.1016/j.engappai.2025.111070}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111070}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rolling bearings remaining useful life estimation using digital twin and physics-informed methods with uncertainty quantification}, volume = {154}, year = {2025}, }
• (2025). Unsupervised multiple-target domain adaptation for bearing fault diagnosis. EAAI, 154, 111063. (www)

  @article{ , abstract = {Data-driven bearing fault diagnosis methods have gained significant attention due to their high accuracy and minimal reliance on expert knowledge. However, models trained on the data collected under single working conditions often assume independent and identically distributed data, limiting their generalization ability in real-world scenarios where varying working conditions introduce challenges such as domain shift and unlabeled data. Unsupervised domain adaptation is an effective method for addressing these challenges by training classifiers using labeled source domain data and unlabeled target domain data. However, current domain adaptation fault diagnosis methods are typically limited to single-source and single-target scenarios. When multiple unlabeled working conditions are involved, it becomes difficult to simultaneously ensure the domain invariance and the fault relevance of features. This paper proposes a fault diagnosis method across multiple unlabeled working conditions. The method extracts features from image data transformed from time-domain signals and ensures the correlation between the extracted features and fault labels, as well as domain invariance, through the gradient reversal mechanism and a corresponding loss function. The proposed method achieves an average accuracy of 99.89 % and 95.93 % in multi-target domain fault diagnosis tasks on two public benchmark bearing datasets, which illustrates its effectiveness and advantages compared with the existing representative methods. The source code is available at https://github.com/WhiteGL/DA_IRP/tree/master .}, archive = {J_EAAI}, author = {Guoli Bai and Tonghao Xing and Wei Sun and Huashan Chi and Zhidan Zhong and Qingchao Sun and Liang Sun}, doi = {10.1016/j.engappai.2025.111063}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111063}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised multiple-target domain adaptation for bearing fault diagnosis}, volume = {154}, year = {2025}, }
• (2025). Toward transforming space exploration with artificial intelligence neuromorphic computing. EAAI, 154, 111055. (www)

  @article{ , abstract = {With space missions venturing farther from Earth and becoming increasingly complex, the integration of neuromorphic systems is imperative. Inspired by the architecture and function of the human brain, neuromorphic computing offers significant advantages in terms of power efficiency, real-time processing, and autonomous decision making. This review explores the emerging role of artificial intelligence (AI) neuromorphic computing in space exploration, highlighting its current state and potential future strategies. Recent advancements have demonstrated its potential for long-duration missions that demand autonomous decision making and real-time data processing. For example, in the BrainStack project, neuromorphic processors operated with a power consumption of less than 1.5 W (W), exhibiting approximately 90 % greater energy efficiency over 15 W consumption by conventional von Neumann processors. However, several challenges persist, including physical constraints imposed by the space environment, such as radiation exposure, extreme temperatures, and vacuum conditions, as well as greater integration with existing spacecraft systems and further improvements in power efficiency for long-duration missions. Therefore, we identify environmental and technical challenges, as well as future directions for deploying neuromorphic technologies in space. To achieve enhanced energy efficiency and operational reliability, future strategies should focus on improving hardware and software capabilities, optimizing spiking neural network algorithms, and incorporating neuromorphic computing into mission design and validation phases. This review is intended to serve as a reference for advancing AI neuromorphic computing as a strategic technology that enables sustainable mission execution and autonomous exploration in extreme space environments.}, archive = {J_EAAI}, author = {Ari Yu and Seungwan Woo and Hyojung Ahn}, doi = {10.1016/j.engappai.2025.111055}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111055}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Toward transforming space exploration with artificial intelligence neuromorphic computing}, volume = {154}, year = {2025}, }
• (2025). Intelligent optimization of air-floating piston core parameters for homemade frictionless pneumatic actuators based on a new multi-objective particle swarm optimization algorithm with gaussian mutation and fuzzy logic. EAAI, 154, 111053. (www)

  @article{ , abstract = {This paper proposes a multi-objective design method for air-floating pistons based on particle swarm optimization algorithm, aiming to efficiently design an ideal piston with strong load-carrying capacity and low air consumption. First, an optimization model for load-carrying capacity and air consumption is constructed based on the air flow characteristics in the air film. Secondly, a new multi-objective particle swarm optimization approach with Gaussian mutation and fuzzy logic is introduced to enhance optimization efficiency. The test results on 15 benchmark functions show that this algorithm has higher convergence and diversity compared with seven other multi-objective evolutionary algorithms. This algorithm is then applied to solve the air-floating piston design problem. The optimization results show that compared with the standard multi-objective particle swarm optimization algorithm, the introduced new algorithm provides solutions with higher quality and more uniform distribution. A set of ideal and a set of non-ideal structural parameters are selected based on the Pareto optimal solutions obtained by the new algorithm, and the corresponding pistons are manufactured. Finally, the load-carrying capacity and air consumption of the air-floating pistons are experimentally tested. It is found that compared with the piston with non-ideal parameters, the piston with ideal parameters not only reduces the air consumption by approximately 50 %, but also increase the load-carrying capacity by more than 10 %. The experimental results are basically consistent with results obtained by the proposed method, verifying the effectiveness of the optimization design method.}, archive = {J_EAAI}, author = {Chenwei Pu and Yifan Jia and Zhihao Zhang and Hui Luo and Mingyuan Ren and Jinhu Wang and Pengfei Qian}, doi = {10.1016/j.engappai.2025.111053}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111053}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent optimization of air-floating piston core parameters for homemade frictionless pneumatic actuators based on a new multi-objective particle swarm optimization algorithm with gaussian mutation and fuzzy logic}, volume = {154}, year = {2025}, }
• (2025). Multimodal stock market emotion recognition model trained with a large language model. EAAI, 154, 111035. (www)

  @article{ , abstract = {Stock market emotion recognition models are often trained on public textual datasets. These datasets are well-labelled but may not have the same distribution as real-world data and consequently fail to reflect the real world. Furthermore, stock market emotion is reflected not only in textual comments but also in images; emotion recognition on the basis of textual comments alone is one-sided. Considering these issues, in this paper, we propose a multimodal emotion recognition framework that is trained via imitation learning from a large language model. The proposed framework contains two main innovations. First, by leveraging large language model's compositionality capability, the proposed framework generates pseudo labels from textual comments. Through imitating the patterns of the pseudo labels, the framework can be trained directly on unlabelled real-world data, which addresses current models' distribution drift between public datasets and the real-world. Second, multimodal fusion is equipped with the proposed framework, enabling emotion recognition from textual stock market comments and images simultaneously. Compared with existing methods, the proposed method in this paper significantly improves the performance of stock market emotion recognition by leveraging large language models and multimodal stock market data. The experimental results demonstrate that the emotion recognition framework proposed in this paper outperforms the existing single-mode models and multimodal models, with the accuracy, precision, recall, and F1 score being 82.97 %, 83.03 %, 83.05 %, and 82.88 % respectively. This framework provides a new pathway for multimodal and unsupervised/semi-supervised emotion recognition in the stock market.}, archive = {J_EAAI}, author = {Chao Liu and Yuxia Miao and Qi Zhao and Chao Wang and Xiangyu Zhu}, doi = {10.1016/j.engappai.2025.111035}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111035}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodal stock market emotion recognition model trained with a large language model}, volume = {154}, year = {2025}, }
• (2025). Intelligent traffic rerouting with dynamic vehicle selection for congestion mitigation. EAAI, 154, 111022. (www)

  @article{ , abstract = {Traffic congestion is an inevitable issue faced by urban cities due to the high population density and transportation requirements. Although public transport and infrastructure improvements can alleviate congestion, spatial limitations restrict these solutions. Therefore, an additional system is necessary to mitigate traffic congestion on existing, overburdened networks. Currently, researchers are primarily investigating traffic rerouting systems integrated with artificial intelligence to distribute vehicles and optimize road usage with minimal costs. Likewise, this paper proposed a dynamic rerouting system integrated with simple artificial intelligence that consists of near-future traffic flow prediction while also introducing a Key Performance Index (KPI) to accurately measure congestion levels and effectively select vehicles for rerouting with minimum cost. The proposed system first aggregates various traffic data to serve as the input of a Support Vector Machine (SVM) traffic flow prediction model, which estimates near-future traffic conditions. These conditions are then measured using a self-proposed KPI to label the congestion level of each road. A vehicle filtration process then selects the vehicles for rerouting based on the congestion level and the travel cost between the origin and destination. Finally, an alternative route is computed via a k Shortest Path (kSP) model by minimizing the travel time. Simulation results indicate that the proposed vehicle rerouting system is capable of reducing the total travel time of a given traffic network by an astonishing 33.1 %.}, archive = {J_EAAI}, author = {Liangyu Tay and Joanne Mun-Yee Lim and Shiuan-Ni Liang and Kah Keong Chua and Yong Haur Tay}, doi = {10.1016/j.engappai.2025.111022}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111022}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent traffic rerouting with dynamic vehicle selection for congestion mitigation}, volume = {154}, year = {2025}, }
• (2025). Multi-criteria consensus sorting model with flexible linguistic preferences based on fuzzy information granulation from the perspective of preference disaggregation. EAAI, 154, 111021. (www)

  @article{ , abstract = {Multi-criteria group sorting (MCGS) that considers linguistic preferences involves multiple individuals evaluating alternatives and assigning them to pre-determined ordered categories based on specific criteria. Nevertheless, due to the limited availability of class information and the constrained cognitive capacity of decision-makers (DMs), it becomes challenging for DMs to furnish explicit preference information to reach consensus. Besides, decision parameters such as consensus threshold and class thresholds are also too hard to be assumed in advance. Therefore, this paper studies the preference disaggregation problem in MCGS, in which DMs are allowed to provide their pairwise comparisons in flexible linguistic expressions (FLEs) as preference information. To more fully utilize the preferences, semantic granulation is introduced. First, an individual consistency recognition model is proposed to identify inconsistent preferences and provide modification directions to the corresponding individuals. Next, semantic granulation and maximum entropy are combined into a granulation-driven information transformation model to convert the preference information based on FLEs into triangular fuzzy numbers (TFNs). Subsequently, in the consensus-driven preference disaggregation model, decision parameters and sorting results can be obtained at the premise of consensus by adjusting weights. Ultimately, to substantiate the effectiveness of the proposal, two numerical applications concerning the sorting of government venture capitals and information system suppliers are presented, along with comparative analysis, sensitivity analysis, and simulation analysis.}, archive = {J_EAAI}, author = {Shiji Zhang and Shitao Zhang and Hao Tian and Muhammet Deveci and Xiaodi Liu}, doi = {10.1016/j.engappai.2025.111021}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111021}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-criteria consensus sorting model with flexible linguistic preferences based on fuzzy information granulation from the perspective of preference disaggregation}, volume = {154}, year = {2025}, }
• (2025). Mechanical reliability of high-power modules via simulation-based machine learning. EAAI, 154, 111019. (www)

  @article{ , abstract = {The thermal cycle reliability and process-induced warpage related to high-power module packaging integrated with the type of insulated gate bipolar transistor modules that have direct bonded copper are critical factors remarkably influencing the longevity and performance of power electronics. Despite the robustness of traditional finite element analysis (FEA) methods, they have profound limitations, including high computational demands and the necessity for extensive experimental validation. This study presents an integrated approach that combines FEA with machine learning (ML) to improve predictive reliability. The ML model integrated support vector algorithms is employed to trained base on FEA-generated stress, strain, and warpage data to predict outcomes with better accuracy and mitigated computational overhead. Hyperparameter optimization through grid search improves the model's predictive capabilities. The proposed ML-FEA framework, validated through comparison with experimental results, effectively forecasts failure probabilities and process-induced warpage under diverse thermal cycling scenarios. This integrated approach improves predictive accuracy and offers a scalable solution for the reliability assessment and design optimization of high-power electronic modules. Accordingly, it provides a foundational reference for mitigating warpage and enhancing overall reliability. Due to predictive accuracy, R2 value is employed to check the ML prediction. In this study, R2 value for ML in training data is more than 0.95. In addition, for computational time decreasing, the time from 604,800 s to lower than 60 s is mitigated.}, archive = {J_EAAI}, author = {Chang-Chun Lee and Wei-Cheng Tsai and Jui-Chang Chuang}, doi = {10.1016/j.engappai.2025.111019}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111019}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mechanical reliability of high-power modules via simulation-based machine learning}, volume = {154}, year = {2025}, }
• (2025). A serverless satellite edge computing infrastructure to detect lake shoreline change based on contrastive attention siamese transformer. EAAI, 154, 111015. (www)

  @article{ , abstract = {Supporting food security, industrial activities, environmental well-being, and ecological balance, water resources are absolutely vital for life on Earth. Still, issues such as water shortage, pollution, and melting glaciers point to the need for sustainable water management techniques. Monitoring changes in the water surface depends on satellite images and remote sensing technologies, but conventional satellite systems suffer with data latency and bandwidth inefficiency. Through data processing closer to the source, latency reduction, and performance enhancement, edge computing has transformed these technologies. Still under study, though, more precise image analysis techniques—especially those based on deep learning models—remain a subject of inquiry. This work contributes to the changing world by proposing a novel serverless satellite edge computing infrastructure with an emphasis on lake shoreline dilatation, a crucial factor in water management. We developed transformer-based Siamese deep learning models as the foundation of the image analysis method. We specifically adjusted the transformer-based Siamese network by including contrastive attention in the fusion module of the model. The suggested model beats state-of-the-art models in overall accuracy by at least 1.75 % by assessing it using actual satellite photos of various lakes all around the planet. Better accuracy in identifying lake shoreline dilatation aids in our understanding of water movement and long-term responsible water resource management in addition to the suggested distributed serverless satellite edge computing infrastructure. This study marks a significant first step towards creatively tackling the problems with water availability and quality, therefore promoting a whole strategy to protect this priceless resource for next generations as well as present ones.}, archive = {J_EAAI}, author = {Mahdi Amini Sedeh and Saeed Sharifian}, doi = {10.1016/j.engappai.2025.111015}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111015}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A serverless satellite edge computing infrastructure to detect lake shoreline change based on contrastive attention siamese transformer}, volume = {154}, year = {2025}, }
• (2025). Multi-objective optimization of additive manufacturing process parameters of nitinol alloys for elastocaloric cooling. EAAI, 154, 111012. (www)

  @article{ , abstract = {Nitinol shape memory alloys are recognized as green and environmentally friendly solid-state refrigerants, and their preparation through additive manufacturing technology demonstrates significant application potential. However, minimizing energy consumption while enhancing material utilization and manufacturing efficiency remains a primary goal and challenge for researchers. This study proposes the application of intelligent optimization algorithms to optimize process parameters, thereby minimizing energy consumption, improving material utilization, and reducing dilution rates. The objective is to achieve full utilization of energy and materials, thus enhancing additive manufacturing efficiency. Compared to traditional experimental methods, this approach eliminates limitations related to test conditions and parameters, allowing for more effective optimization of additive manufacturing process parameters. Designing and conducting experiments on key process parameters—such as laser powers, scanning speeds, powder feeding rates, and shielding gas flow rates—over a broader range can provide accurate data for intelligent optimization algorithms. The results indicate that the established genetic algorithm-optimized back propagation neural network prediction model exhibits high prediction accuracy for energy consumption, powder utilization, and dilution rates. The optimal process parameters were determined using a Pareto front for a multi-objective optimization problem. This multi-objective optimization resulted in reduction of energy consumption and dilution rates by 34.1 % and 27.2 %, respectively, while increasing the metal powder utilization rate by 139.8 % compared to the average of the experimental results. This study contributes to the reduction of energy and material consumption in the additive manufacturing process of Nitinol shape memory alloys and promotes the development of green cooling technology utilizing these alloys as solid-state refrigerants.}, archive = {J_EAAI}, author = {Shuyao Wang and Yongjun Shi and Wenjia Cheng and Kaijun Fan and Qin Wang and Ying Li and Quan Li and Huayang Sun}, doi = {10.1016/j.engappai.2025.111012}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111012}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective optimization of additive manufacturing process parameters of nitinol alloys for elastocaloric cooling}, volume = {154}, year = {2025}, }
• (2025). Physically interpretable discrete latent representations for the design of advanced mechanical metamaterials in complex geometries. EAAI, 154, 111011. (www)

  @article{ , abstract = {Recent advancements in artificial intelligence (AI)-based design strategies for mechanical metamaterials have significantly improved the ability to create customizable architectures across nano-to macro-scale dimensions. However, challenges persist in expanding the flexibility of design space and generating multiple microstructures within complex geometries. This study introduces an innovative approach that combines topology optimization (TO) with a deep generative model for mechanical metamaterials, enabling the generation of multiple microstructure compositions within arbitrarily shaped domains. The model leverages discrete latent vectors to achieve high prediction accuracy for mechanical properties, with R 2 = 0.99 on in-distribution sets and R 2 = 0.97 on out-of-distribution sets, surpassing the performance of traditional machine learning methods. Detailed analysis reveals strong correlations between these latent vectors and the mechanical, structural, and geometrical characteristics of microstructures, offering new insights into their physical interpretation. The proposed approach significantly enhances the ability to represent complex geometries with high resolution, enabling intricate designs while achieving results 7–8 times faster than conventional methods and a 10 % reduction in compliance without increasing the volume fraction. This innovative framework has broad potential applications in aerospace, biomedical engineering, and robotics, representing a pivotal advancement in understanding the complex latent design spaces of mechanical metamaterials.}, archive = {J_EAAI}, author = {Hansom Choi and Youngjoon Hong and Namjung Kim}, doi = {10.1016/j.engappai.2025.111011}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111011}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physically interpretable discrete latent representations for the design of advanced mechanical metamaterials in complex geometries}, volume = {154}, year = {2025}, }
• (2025). A gradient descent matrix factorization for microtask crowdsourcing. EAAI, 154, 111003. (www)

  @article{ , abstract = {Conventional crowdsourcing algorithms assign all tasks to all workers and collect their final answers. However, a major issue with these baseline approaches is their inability to identify and eliminate answers from low-quality or non-genuine workers. To tackle this challenge, novel microtasking algorithms have been proposed where workers' expertise is estimated to allocate tasks accordingly. These algorithms rely on a worker-task matrix, but a common problem is the potential sparsity of this matrix due to not all workers answering allocated tasks and a high ratio of tasks to workers. To address this issue, this paper introduces a novel optimization-based matrix factorization approach using Gradient Descent known as GRADi. GRADi aims to predict missing answers from different workers while factorizing the worker-task matrix. During this process, GRADi incorporates worker similarity information to enhance factorization accuracy. The effectiveness of GRADi is evaluated using datasets from Amazon Mechanical Turk and similar platforms, assessing accuracy and Root Mean Square Error. Comparative analyses against other matrix factorization algorithms, including both optimization and non-optimization techniques, as well as existing microtasking algorithms, demonstrate that GRADi consistently outperforms these methods. It shows promising results in improving microtasking outcomes by better predicting worker responses and handling the inherent challenges of sparse worker-task matrices.}, archive = {J_EAAI}, author = {Alireza Moayedikia}, doi = {10.1016/j.engappai.2025.111003}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111003}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A gradient descent matrix factorization for microtask crowdsourcing}, volume = {154}, year = {2025}, }
• (2025). A customized multi-class pavement distress segmentation method for routine repair monitoring. EAAI, 154, 111002. (www)

  @article{ , abstract = {This study investigates the efficacy of intelligent detection methods for monitoring the quality of urban road asphalt pavement repairs, and focuses on addressing the challenges of identifying overlapping target pixels and differentiating between fine-scale distress and large-scale block patches, tasks that heavily rely on spatial semantics. A segmentation dataset comprising 13 classes and 5633 labels was constructed. Five semantic segmentation models including Deeplab V3+, SCTNet (Single-Branch Convolutional Neural Network with Transformer Semantic Information), FastFCN (Fast Fully Convolutional Network), MobileNet V3, and SegNext were constructed and evaluated. Despite its relatively smaller size, SCTNet exhibited the highest processing speed, while MobileNetV3 exhibited the smallest size and lowest accuracy. The SegNeXt model demonstrated superior performance in both segmentation accuracy and model complexity, making it chosen as the baseline model. Three training strategies were explored: classifying pavement as background, employing multiscale input, and integrating a weighted loss function. The model that combined all three strategies (referred to as SegNeXt_IPWM) demonstrated the most promising results. Compared to the baseline model, SegNeXt_IPWM achieved significant enhancements, with a 2.08 % increase in mIoU (mean Intersection over Union) and a 1.42 % improvement in mean F-score . Notable improvements were observed across all categories except for marking loss, with particularly substantial gains in loose (+7.75 % IoU), block patch (+6.62 % IoU), and linear crack (+4.87 % IoU) detection. Additionally, SegNeXt_IPWM demonstrated superior generalization capabilities, especially in scenarios involving background target interference, underscoring its potential for robust performance in applications.}, archive = {J_EAAI}, author = {Hui Wang and Junyang Kang and Shuichao Zhang and Yanmao Xiang and Jun Li}, doi = {10.1016/j.engappai.2025.111002}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111002}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A customized multi-class pavement distress segmentation method for routine repair monitoring}, volume = {154}, year = {2025}, }
• (2025). Hybrid deep learning model to predict the ultimate tensile strength of friction stir welded joints. EAAI, 154, 111001. (www)

  @article{ , abstract = {Accurate prediction of tensile strength in friction stir welded joints is essential to improve the reliability of welded structures. Traditional Artificial Neural Networks (ANNs) often struggle to capture the complex relationships between welding parameters, leading to limited predictive accuracy. To address this challenge, a hybrid model was developed, combining Long Short-Term Memory (LSTM) and dense layers to leverage the ability of LSTM to learn sequential patterns and dependencies between inputs such as rotational speed, feed rate, and tool tilt angle. This hybrid model was compared to a traditional ANN model using a dataset containing welding parameters and tensile strength results. Performance indicators, including correlation coefficient and Mean Square Error (MSE), showed that the LSTM + Dense hybrid model achieved a higher correlation coefficient (0.9803) and lower MSE on the validation sets, indicating improved predictive ability. Analyses of error distribution and convergence of training and validation losses also showed that the hybrid model had higher consistency and reliability with reduced overfitting, even when dealing with outliers. The LSTM + Dense model was shown to significantly outperform the traditional ANN, making it a powerful tool for predicting tensile strength in welding applications with complex and nonlinear data. Although the accuracy and reliability of the proposed hybrid model have improved, limitations remain, including the need for larger datasets to enhance generalization and potential challenges in real-time implementation.}, archive = {J_EAAI}, author = {Akshansh Mishra and Abdelkader Miloudi and Eyob Messele Sefene and Chaaben Arroussi and Ismail Chekalil and Bassam Gamal Nasser Muthanna}, doi = {10.1016/j.engappai.2025.111001}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111001}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid deep learning model to predict the ultimate tensile strength of friction stir welded joints}, volume = {154}, year = {2025}, }
• (2025). A multimodal automatic generation and annotation framework for prohibited and restricted goods in online transactions. EAAI, 154, 111000. (www)

  @article{ , abstract = {Transactions of prohibited and restricted goods on e-commerce platforms threaten consumer safety and hinder Internet economy growth. However, the lack of such datasets hinders intelligent identification and effective regulation. To address this issue, a large-scale multimodal dataset of prohibited and restricted goods is constructed, comprising 18,446 images and 36,892 texts. Nevertheless, this is insufficient due to the diverse forms and deep concealment of prohibited and restricted products in online transactions. Therefore, we propose a multimodal automatic generation and annotation framework for prohibited and restricted goods in online transactions. This framework consists of an image generation module, a text description module, and an image annotation module. The image generation module is utilized to generate more diverse images for prohibited and restricted goods. The text description module can generate text descriptions for the generated images. The image annotation module is employed to annotate the generated images for various visual tasks. This framework achieved a 24.71 user rating for image generation, a 92.72 % mean average precision for image annotation, and an 81.67 % semantic score for image description. Experimental results demonstrate the effectiveness and accuracy of the proposed automatic annotation framework, which can enhance the effective supervision of prohibited and restricted goods.}, archive = {J_EAAI}, author = {Wenjin Liu and Shudong Zhang and Haoming Liu and Jingyu Zou}, doi = {10.1016/j.engappai.2025.111000}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {111000}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multimodal automatic generation and annotation framework for prohibited and restricted goods in online transactions}, volume = {154}, year = {2025}, }
• (2025). A novel framework for efficient dominance-based rough set approximations using K-dimensional (K-d) tree partitioning and adaptive recalculations techniques. EAAI, 154, 110993. (www)

  @article{ , abstract = {In dynamic datasets, both entries and their possible values frequently change. This makes it difficult to update the necessary calculations for analysis. The Dominance-based Rough Set Approach (DRSA) effectively analyzes preference-based data. Still, it can be slow because it requires recalculating important values each time the data are updated. This study proposes an optimized K-dimensional (K-D) tree algorithm that integrates dynamic weighting and lazy recalculations to reduce computational cost. The K-D tree effectively organizes data, dynamic weighting prioritizes recalculations for essential nodes, and lazy recalculation postpones unnecessary computation. The proposed approach significantly reduces the execution time and memory consumption while maintaining accuracy. An ablation study was conducted to separately evaluate the impact of each component. This demonstrates the efficiency improvement achieved using the proposed approach. Compared to conventional DRSA, our optimized method reduced computational time and memory usage by 99 % across ten publicly available University of California, Irvine (UCI) datasets, which provided an efficient and scalable solution for real-time data processing in dynamic environments.}, archive = {J_EAAI}, author = {Uzma Nawaz and Zubair Saeed and Kamran Atif}, doi = {10.1016/j.engappai.2025.110993}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110993}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel framework for efficient dominance-based rough set approximations using K-dimensional (K-d) tree partitioning and adaptive recalculations techniques}, volume = {154}, year = {2025}, }
• (2025). Blockchain-enabled healthcare supply chain management: Identification and analysis of barriers and solutions based on improved zero-sum hesitant fuzzy game theory. EAAI, 154, 110991. (www)

  @article{ , abstract = {Blockchain technology has emerged as a transformative approach in the health sector, enhancing efficiency, transparency, and security in Healthcare Supply Chain Management (HSCM). It addresses critical issues such as data privacy, traceability, and fraud reduction, providing a secure and reliable platform. However, significant barriers to its implementation must be overcome to ensure effective healthcare supply chain operations. This study proposes a two-stage decision-making model for identifying barriers and optimizing blockchain adoption solutions in HSCM under uncertainty. The first stage employs the Hesitant Fuzzy Best-Worst Method (HFBWM) to prioritize barriers. Compared to traditional methods such as Analytic Hierarchy Process (AHP), HFBWM achieves high accuracy with fewer pairwise comparisons. In the second stage, the Improved Zero-Sum Hesitant Fuzzy Game Theory (IZSHFG) model, based on the Weighted Sum Operator (WSO) under Hesitant Fuzzy Sets (HFSs), determines the optimal combination of strategies for blockchain application in HSCM. The challenges are modeled as one player and the solutions as another, with the decision matrix established using WSO under HFS. The obtained results indicate the worst-case scenario involves the simultaneous occurrence of four critical barriers: “Lack of Sufficient Knowledge about Blockchain in HSCM” (0.011217), “Lack of Access to Skilled Technical Personnel” (0.025457), “High Maintenance and Support Costs” (0.056076), and “Security Risks of Patients' Data” (0.069367). These findings highlight the need for targeted strategies to address these barriers, ensuring blockchain's successful integration into HSCM.}, archive = {J_EAAI}, author = {Seyed Behnam Razavian and Erfan Babaee Tirkolaee and Vladimir Simic and Sadia Samar Ali and Ömer Faruk Görçün}, doi = {10.1016/j.engappai.2025.110991}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110991}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Blockchain-enabled healthcare supply chain management: Identification and analysis of barriers and solutions based on improved zero-sum hesitant fuzzy game theory}, volume = {154}, year = {2025}, }
• (2025). Artificial neural network aided computing for two dimensional magnetohydrodynamic peristaltic movement of nanofluid with heat and mass transfer. EAAI, 154, 110990. (www)

  @article{ , abstract = {This study investigates an innovative Artificial Intelligence ( AI ) computing solver employing a multilayer perceptron, specifically the Levenberg Marquardt scheme implemented in Backpropagated Neural Network (LMB-ANN), to tackle the novel mathematical model of Magnetohydrodynamic (MHD) peristaltic movement of non-Newtonian nanofluid. This examination considers the effects of Hall current, mixed convection, Joule heating, and viscous dissipation. The Buongiorno nanofluid model is utilized to investigate the influences of thermophoresis and Brownian motion. The main purpose of this work is to construct an Artificial Neural Network (ANN) capable of accurately predicting dimensionless flow parameters across diverse flow profiles. The governing equations of this problem are simplified using negligible Reynolds number and long wavelength approximations, and the resulting system is numerically solved employing the BVP4c algorithm based on the finite difference scheme in MATLAB. A dataset is created using numerical calculations for the proposed LMB-ANN model, incorporating twelve different scenarios across various profiles such as axial velocity, nanoparticle concentration, wall stresses, mass transfer, nanofluid temperature, and thermal transport. This dataset is designed to analyze the peristaltic motion of the Reiner Philippoff nanofluid model, with the flow parameters adjusted accordingly. The data is split into three sections: 15 % for testing, 70 % for validation, and 15 % for training. These sections are used within the LMB-ANN to compare numerical solutions with the predicted results from Levenberg Marquardt Neural Networks (LM-NNs). The accuracy and effectiveness of the LMB-ANN are assessed through error histograms, performance evaluations, regression analysis, correlation indices, and fitness curves, based on the Mean Squared Error (MSE), ranging from 10 − 11 to 10 − 6 . ANN-predicted results are further validated through tables and graphs for heat and mass transfer, axial velocity, concentration profile, and nanofluid's temperature. Based on the results of the present examination, the concentration profile improves as the Brownian motion parameter and concentration Grashof number increases. Absolute error between the ANN-predicted results and the reference solution for these parameters varies from 10 − 7 → 10 − 4 .}, archive = {J_EAAI}, author = {H. Alahmadi and J. Iqbal and F.M. Abbasi}, doi = {10.1016/j.engappai.2025.110990}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110990}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial neural network aided computing for two dimensional magnetohydrodynamic peristaltic movement of nanofluid with heat and mass transfer}, volume = {154}, year = {2025}, }
• (2025). Aczel–Alsina operations‐based linguistic q-rung orthopair fuzzy aggregation operators and their application to site selection of electric vehicle charging station. EAAI, 154, 110989. (www)

  @article{ , abstract = {Air pollution and climate change caused by fossil fuels adversely affect human health. The transport sector's reliance on fossil fuels intensifies air pollution and climate change. Electric vehicles (EVs) offer a cleaner alternative, reducing emissions and environmental impact significantly. The availability of convenient charging stations is the first priority in supporting the widespread adoption of EVs. The process of site selection for electric vehicle charging stations (EVCSs) plays a crucial role in advancing the EV industry, and this evaluation task typically falls under the domain of Multi-Attribute Group Decision-Making (MAGDM). In this paper, an applicable MAGDM model is designed to evaluate the EVCSs locations. To achieve this, we utilize the concept of linguistic q-rung orthopair fuzzy sets (Lq-ROFS) to represent decision-maker (DM) assessments, which offers improved precision in modeling uncertainty compared to traditional fuzzy sets. Subsequently, we introduce novel Aczel–Alsina operations for Lq-ROFS, including addition, multiplication, scalar multiplication and power operation. This paper discusses that these operations not only satisfy the closure but also satisfy the exchange law and distribution law. Based on these defined operations, we develop various Lq-ROF aggregation operators, including Lq-ROF Aczel–Alsina weighted averaging (Lq-ROFAAWA) operator, Lq-ROF Aczel–Alsina ordered weighted averaging (Lq-ROFAAOWA) operator, Lq-ROF Aczel–Alsina weighted geometric (Lq-ROF AAWG) operator and Lq-ROF Aczel–Alsina ordered weighted geometric (Lq-ROFAAOWG) operator. Each operator is designed with distinct characteristics to address diverse decision-making scenarios effectively. Moreover, we propose a new model dependent on these operators to solve the MAGDM problem. Finally, the applicability of the established approach is demonstrated through a case study related to site selection for EVCSs. The sensitivity and comparative analyses of the model indicate that it has strong effectiveness, performs better than other existing methods, and is largely applicable to a broad range of theoretical and practical problems.}, archive = {J_EAAI}, author = {Shahid Hussain Gurmani and Muhammad Jabir Khan and Weiping Ding and Rana Muhammad Zulqarnain}, doi = {10.1016/j.engappai.2025.110989}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110989}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Aczel–Alsina operations‐based linguistic q-rung orthopair fuzzy aggregation operators and their application to site selection of electric vehicle charging station}, volume = {154}, year = {2025}, }
• (2025). Multi-echelon inventory optimization of waste electrical and electronic equipment closed-loop supply chain based on reinforcement learning under carbon tax policy. EAAI, 154, 110987. (www)

  @article{ , abstract = {In response to environmental challenges posed by waste electrical and electronic equipment (WEEE), the WEEE closed-loop supply chain (CLSC) has emerged as a crucial means to promote circular economy through the recycling and reuse of WEEE, which not only reduces waste emissions but also improves the efficiency of resource utilization. In practice, both economic and environmental benefits must be considered in sustainable manufacturing within the CLSC to ensure the sustainable development of enterprises. Therefore, a multi-echelon, multi-period inventory model for the CLSC under carbon tax policy is developed in this paper, which innovatively introduces the carbon footprint to assess environmental impact and integrates the impacts of collection planning and the uncertainty of recycling quantity and product demand on the system, aiming to minimize total enterprise costs. To address the uncertainties in the model, the Proximal Policy Optimization (PPO) algorithm is employed to train a reinforcement learning (RL) agent. This agent enables enterprises to dynamically adjust internal strategies such as collection and production, as well as external procurement strategies, based on inventory levels and market conditions. By internalizing carbon emission costs through the carbon tax rate, the RL agent optimizes total costs while achieving a balance between economic and environmental benefits. Numerical experiments demonstrate that the PPO algorithm outperforms the traditional inventory management policy in terms of both cost control and carbon footprint reduction. Moreover, the moderate carbon tax policy on the CLSC appropriately increases the cost of enterprises while significantly reducing their carbon footprint and promoting sustainable development.}, archive = {J_EAAI}, author = {Jinlong Wang and Shangzhuo Zhou and Min Li and Guanyu Ren and Xianquan Ren and Xiaoyun Xiong and Yuanyuan Zhang}, doi = {10.1016/j.engappai.2025.110987}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110987}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-echelon inventory optimization of waste electrical and electronic equipment closed-loop supply chain based on reinforcement learning under carbon tax policy}, volume = {154}, year = {2025}, }
• (2025). Mechanism of knowledge management system self-learning case and knowledge matching based on bidirectional dimension reduction. EAAI, 154, 110985. (www)

  @article{ , abstract = {Purpose Knowledge overload in enterprises generates inefficient knowledge applications, which leads to confusion and inefficiency in decision-making and problem-solving. It can be effectively mitigated by implementing case knowledge matching process. Therefore, the current study aims to reduce redundancy within knowledge management systems and enhance the efficiency and effectiveness of case knowledge matching. Design/methodology/approach This study first applies information gain to retain important attributes to address the redundancy issue in knowledge management systems self-learning case knowledge. Then, spectral clustering is used to create a double-index structure that can identify and eliminate inconsistent case knowledge. The research measures the matching degree using view similarity in a reduced-dimensional space, resulting in improved matching outcomes for each problem. And then compare bidirectional dimension reduction method with neighborhood rough sets, inductive retrieval, and information gain on various University of California Irvine datasets. Findings Experimental analysis demonstrated that with properly calibrated model parameters, our approach effectively reduces redundancy in knowledge management systems' self-learning case knowledge while achieving high efficiency and effectiveness in knowledge matching. Originality/value This research introduces the concept of inconsistent case knowledge and employs information gain for bidirectional dimension reduction, significantly compressing the matching space. Consequently, the case knowledge base can be utilized more effectively to meet the comprehensive needs of knowledge subjects better.}, archive = {J_EAAI}, author = {Xiaoqian Zhou and Ziao Cao and Sajjad Alam and Longfei He}, doi = {10.1016/j.engappai.2025.110985}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110985}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mechanism of knowledge management system self-learning case and knowledge matching based on bidirectional dimension reduction}, volume = {154}, year = {2025}, }
• (2025). An explainable artificial intelligence – Human collaborative model for investigating patent novelty. EAAI, 154, 110984. (www)

  @article{ , abstract = {With the accumulation of technology-related big data, including the patent database, existing studies have proposed a framework for patent analysis using natural language processing models. Artificial intelligence (AI) applications require human experience and insight based on the understanding of complex environments and uncertainties and model predictive performance. However, existing research has focused on applying big data and developing automated processes. Actual user understanding and the consideration of model usability are insufficient. Studies must consider the human–machine cooperation-based approach in developing the AI model. This study proposes a collaborative approach through which the explainable AI (XAI) model, a self-explaining deep neural network for text classification, communicates with users. The proposed XAI model provides users with an explanation for the model prediction along with the prediction results for patent evaluation. Users provide feedback based on the model predictions and their explanations. The source XAI model is refined via relearning by reflecting on user feedback. This study experiments to assess model improvement using the human collaboration method. As for the human collaborative method, this study considers the process of human intervention independent of the XAI model's results as well as the method of human participation based on the explanation presented by the XAI model. The experimental results verified the XAI model performance, showing the highest accuracy (0.890) and F1 score (0.916), such that the model can be applied efficiently to patent evaluation. The XAI–human collaboration model presented in this study can also be expanded and applied to technology intelligence tasks. However, the collaborative approach in this study has complete trust in human advice from technical experts; thus, subsequent collaborative XAI models could be improved by communicating bidirectionally with human resources as a complementary relationship.}, archive = {J_EAAI}, author = {Hyejin Jang and Byungun Yoon}, doi = {10.1016/j.engappai.2025.110984}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110984}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An explainable artificial intelligence – Human collaborative model for investigating patent novelty}, volume = {154}, year = {2025}, }
• (2025). Defense against false data injection attacks on the electric vehicle charging stations data markets. EAAI, 154, 110983. (www)

  @article{ , abstract = {Data-driven technology depends on high-quality training data. Although many research institutions advocate for data sharing, private data owners are often reluctant to share their data considering the privacy concerns related to potential data breaches. As a result, the availability of data limits the application of data-driven technologies in energy systems. To enhance the availability of data, we have constructed a data market model for forecasting the power demand of electric vehicle charging stations (EVCSs), enabling data transactions within this market to improve the accuracy of forecasts. Since the data market relies on communication networks to collect data, this makes it vulnerable to malicious false data injection attacks (FDIAs) during transmission, exposing the data market to serious security risks. To ensure the safe operation of this data market, this article proposes a defense method based on Wasserstein Generative Adversarial Network that combines Transformer and Convolutional Neural Network(TCWGAN). This method effectively reduces the impact of FDIAs and has a strong defense against the injection of false data, achieving an accuracy of 95.09 %.}, archive = {J_EAAI}, author = {Huqun Mu and Aiping Pang and Congmei Jiang and Wen Yang and Qianchuan Zhao}, doi = {10.1016/j.engappai.2025.110983}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110983}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Defense against false data injection attacks on the electric vehicle charging stations data markets}, volume = {154}, year = {2025}, }
• (2025). Short-term electricity-load forecasting by deep learning: A comprehensive survey. EAAI, 154, 110980. (www)

  @article{ , abstract = {Short-Term Electricity-Load Forecasting (STELF) refers to the prediction of the immediate demand (in the next few hours to several days) for the power system. Various external factors, such as weather changes and the emergence of new electricity consumption scenarios, can impact electricity demand, causing load data to fluctuate and become non-linear, which increases the complexity and difficulty of STELF. Over the past decade, deep learning, as a key component of implemented artificial intelligence, has been widely applied to STELF, enabling accurate modeling and prediction of electricity demand. This paper provides a comprehensive survey on deep-learning-based STELF over the past ten years. It examines the entire forecasting process, including data pre-processing, feature extraction, deep-learning modeling and optimization, and results evaluation. This paper also identifies key research challenges and potential directions for further investigation in artificial intelligence applications to STELF.}, archive = {J_EAAI}, author = {Qi Dong and Rubing Huang and Chenhui Cui and Dave Towey and Ling Zhou and Jinyu Tian and Jianzhou Wang}, doi = {10.1016/j.engappai.2025.110980}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110980}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Short-term electricity-load forecasting by deep learning: A comprehensive survey}, volume = {154}, year = {2025}, }
• (2025). An edge-guided defect segmentation network for in-service aerospace engine blades. EAAI, 154, 110974. (www)

  @article{ , abstract = {Currently, 80 % of in-service aerospace engine blade defect detection relies on manual visual assessment. Operators use a borescope to capture images of the blade surface and make judgments based on their experience and expertise. However, this method is costly and time-consuming. With the widespread application of artificial intelligence across various fields, its strong capabilities in automated defect detection are becoming increasingly evident. To meet the demand for efficient defect detection in aero-engine blades, we have constructed a dataset based on videos collected from real inspection scenarios, ensuring alignment with actual defect types. Based on this dataset, we analyze existing defect detection methods for in-service aero-engine blades and propose an improved edge-guided and channel-enhanced network using the "Transformer" architecture. Our method leverages global edge information from "Segment Anything (SAM)" to guide learning, while the channel shuffling module boosts feature capture. Experimental results show an mean intersection over union (mIoU) of 88.13 % and a detection speed of 30.6 frames per second (FPS) on a single graphics processing unit (GPU), meeting real-world efficiency needs. The code will be publicly available at the link: https://github.com/Newbiejy/EGCIENet_In-service-blade-defect-detection .}, archive = {J_EAAI}, author = {Xianming Yang and Kechen Song and Shaoning Liu and Fuqi Sun and Yiming Zheng and Jun Li and Yunhui Yan}, doi = {10.1016/j.engappai.2025.110974}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110974}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An edge-guided defect segmentation network for in-service aerospace engine blades}, volume = {154}, year = {2025}, }
• (2025). An intelligent hybrid machine learning framework for compressive strength prediction of alkali-activated binders based on fly ash characteristics. EAAI, 154, 110971. (www)

  @article{ , abstract = {The inconsistent physicochemical properties of coal combustion fly ash limit its reliable utilization in alkali-activated binders, despite its potential as a sustainable precursor. To overcome this challenge, this work proposes a hybrid machine learning framework that incorporates dataset optimization for the prediction of alkali-activated binder performance based on the characteristics of fly ash (e.g., chemical composition, particles state, and leaching capacity). The framework is comprised of three key segments: data-optimization, data-preparation module, as well as the training module. The framework addresses data scarcity through synthetic sample generation via attention-enhanced generative adversarial networks, followed by anomaly removal using isolation forest algorithms. Subsequently, an optimized database related to compressive strength was creation to analyzed the performances of six models, in which the transformer model shows the best ability, with testing determination coefficient of the transformer model increased from 0.89 to 0.97 following the implementation of the framework. The generalization of the model was evaluated via microstructural analysis and previous calculated model.}, archive = {J_EAAI}, author = {Ke-yu Chen and Zhen-ming Li and Zhi-tai Zhu and Shu-yang Zhang and Shun Li and Jin Xia}, doi = {10.1016/j.engappai.2025.110971}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110971}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An intelligent hybrid machine learning framework for compressive strength prediction of alkali-activated binders based on fly ash characteristics}, volume = {154}, year = {2025}, }
• (2025). Comprehensive performance evaluation of valuable medical equipment based on cloud modelling and combined weighting methodologies. EAAI, 154, 110969. (www)

  @article{ , abstract = {The construction of the performance evaluation index system for valuable medical equipment is the basis for measuring the use of medical equipment in hospitals. It is crucial to the management and evaluation of equipment. This study aimed to develop a comprehensive evaluation model that thoroughly assesses the operational status, utilization efficiency, and service quality of valuable medical equipment in hospitals. The performance evaluation index system for valuable medical equipment was constructed using four dimensions. The subjective weights were determined using the Analytical Hierarchy Process (AHP), while the objective weights were calculated using the Entropy Weight Method (EWM). The combined weights of the index system were derived by integrating both subjective and objective weights through game theory. The Delphi method was employed to establish a standard cloud model, which was subsequently integrated with the combined weights to construct a comprehensive evaluation cloud model. This research evaluated the performance of nine newly acquired pieces of valuable medical equipment that were operational in a hospital after 2020, thereby validating the reliability of the proposed model. The outcomes indicate that the model effectively addresses the problem of data uncertainty in fuzzy evaluations while alleviating the limitations associated with single weighting methods. The performance evaluation model for medical equipment proposed in this study provided an innovative and effective strategy for assessing valuable medical equipment in hospitals, thereby enhancing the scientific and effective management of medical equipment.}, archive = {J_EAAI}, author = {Xingtong Zhang and Saifeng Fang and Yongchun Jin and Ying Huang and Shucheng Wang and Jie Wang and Yunhua Xu}, doi = {10.1016/j.engappai.2025.110969}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110969}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comprehensive performance evaluation of valuable medical equipment based on cloud modelling and combined weighting methodologies}, volume = {154}, year = {2025}, }
• (2025). Real-time fuzzy logic-based direct power control for wind energy systems. EAAI, 154, 110968. (www)

  @article{ , abstract = {This paper presents a fuzzy logic-based direct power control (F-DPC) strategy for wind turbines using double-fed induction generators (DFIGs) to improve power quality and system performance. In contrast to conventional direct power control (C-DPC), the proposed approach minimizes power ripple, improves response time and significantly reduces total harmonic distortion (THD). Simulation results show that F-DPC reduces THD from 7.06 % to 1.53 % in super-synchronous mode, from 4.88 % to 0.92 % in synchronous mode, and from 4.67 % to 1.18 % in sub-synchronous mode, achieving an average improvement of 78.08 %. In addition, F-DPC effectively reduces power ripple by 72.29 % for active power and 70.93 % for reactive power, as well as overshoot and steady-state error ensuring a more stable and efficient power conversion process. To assess its robustness, the performance of F-DPC is further evaluated under parametric variations, including a 100 % increase in winding resistances and a 20 % reduction in inductances. The results confirm that F-DPC maintains stable power regulation, reduced fluctuations and improved dynamic response, outperforming C-DPC in terms of power ripple, overshoot and steady-state error. To validate its real-time feasibility, the F-DPC strategy is implemented and tested on an OPAL-RT OP4512 real-time simulator using the RT-LAB platform. The real-time simulation results closely match the MATLAB/Simulink results, confirming that F-DPC maintains efficient power control under various operating conditions. These results highlight the practical scalability of F-DPC for real wind energy systems and demonstrate its potential to improve grid integration, power quality and overall system reliability.}, archive = {J_EAAI}, author = {Karim Fathi Sayeh and Salah Tamalouzt and Djamel Ziane and Nabil Benyahia and Sofia Lalouni Belaid and Youcef Belkhier}, doi = {10.1016/j.engappai.2025.110968}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110968}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time fuzzy logic-based direct power control for wind energy systems}, volume = {154}, year = {2025}, }
• (2025). Sensitive components of temperature-induced track deformation on cable-stayed bridge impacting dynamic response of high-speed train based on deep learning. EAAI, 154, 110967. (www)

  @article{ , abstract = {To explore the impact of different components of track deformation caused by temperature variations of cable-stayed bridges (CSBs) and slab tracks on the dynamic response of high-speed trains (HSTs), this paper established an analytical framework based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) and deep learning. Initially, a finite element model for the slab track-CSB and a multi-body dynamics model for HST-CSB were established to calculate the temperature-induced deformation of the slab tracks on the CSB and the corresponding dynamic response of the HST. Subsequently, the ICEEMDAN and Convolutional Neural Networks (CNN)-Bi-Long Short-term Memory (BiLSTM)-Convolutional Block Attention Module (CBAM), were fused to obtain the integrated ICEEMDAN-CNN-BiLSTM-CBAM (ICBC) model. The ICEEMDAN can effectively decompose the signal of track deformation by frequency, while BiLSTM had superior performance in capturing the spatial features of input data. The model could establish the mapping relationship between track deformation components and vehicle dynamic responses. The main conclusions obtained are as follows: When both the CSB and slab track experienced a simultaneous temperature rise, it resulted in an intricate combination of short-wave track deformations, long-wave bridge deformations, and localized deformations induced by girder ends. At the mid-span location, since the span length is as large as 450m, the wavelength of the deformation induced by temperature variations has also increased. By extracting the weight vectors of the output neural network model, the contribution of each Intrinsic Mode Functions to the vehicle dynamic response was obtained. The dynamic response of HSTs was most sensitive to the long-wave and large-amplitude components in the temperature deformation of slab track on CSB. This study could contribute to the understanding of the spatial deformation components of slab tracks on CSB and provide scientific predictions for the dynamic response of HSTs.}, archive = {J_EAAI}, author = {Qian Zhang and Xiaopei Cai and Jijun Wang and Weibin Liu and Yilin Zhong and Moyan Zhang and Ji Wang}, doi = {10.1016/j.engappai.2025.110967}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110967}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sensitive components of temperature-induced track deformation on cable-stayed bridge impacting dynamic response of high-speed train based on deep learning}, volume = {154}, year = {2025}, }
• (2025). Enhanced deep learning framework for real-time pain assessment using multi-modal fusion of facial features and video streams. EAAI, 154, 110966. (www)

  @article{ , abstract = {This paper introduces an enhanced deep learning multimodal framework for real-time pain assessment that addresses the limitations of the single-mode approaches. Specifically, time-related disadvantages and its poor temporal analysis, noted in recent studies. The proposed framework fuses VGG16 (Visual Geometry Group 16-layer network), a pre-trained convolutional neural network (CNN), with a bidirectional long short-term memory network (Bi-LSTM) augmented with attention mechanisms. VGG16 has a proven ability to capture fine-grained spatial information in facial expressions to extract powerful spatial features, whereas the Bi-LSTM augmented with attention mechanisms because of its computational efficiency and concisely captures temporal features. Moreover, adaptive boosting has used seeking generalization across a broad base of settings and Gradient-weighted Class Activation Mapping (Grad-CAM) to provide comprehensible and interpretable visual explanations of the decision-making processes of the framework. Extensive evaluations were conducted on diverse popular benchmark datasets: BioVid, UNBC-McMaster, and MIntPAIN. Our framework achieves an AUC (the Area under the Receiver Operating Characteristic Curve) of 0.99 and 99% cross-validation accuracy, corresponding to a 12.5% improvement in accuracy and a 20% reduction in false positives relative to state-of-the-art convolutional neural network–long short-term memory (CNN–LSTM) models. Real-time deployment on an NVIDIA Jetson Nano yields 120 ms (ms) inference per frame with less than 1% accuracy loss, demonstrating its feasibility for edge-based telemedicine. Combined with explainability via Grad-CAM and privacy-preserving vision-only inputs, this framework offers an ethical, scalable solution for nonverbal pain detection.}, archive = {J_EAAI}, author = {Hany El-Ghaish and Mohamed Yousry Al-Basiouny and Mahmoud A.M. Alshewimy}, doi = {10.1016/j.engappai.2025.110966}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110966}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced deep learning framework for real-time pain assessment using multi-modal fusion of facial features and video streams}, volume = {154}, year = {2025}, }
• (2025). A channel-independent network based on wavelet enhancement for long-term time series forecasting. EAAI, 154, 110964. (www)

  @article{ , abstract = {Long-term time series forecasting is critical for applications like financial market analysis, energy consumption prediction, and industrial optimization. However, this task is challenging due to the long prediction horizon, high data volatility, and complex multi-variable dependencies. While Transformer-based models have shown promising performance by focusing on encoder–decoder structures and attention mechanisms, they primarily capture correlations within individual samples, neglecting inter-sample dependencies. Additionally, these models mainly rely on attention mechanisms and seldom explore alternative approaches. To address these limitations, we propose a novel single-encoder framework that adopts a channel-independent, patch-based processing mode. Specifically, we apply multi-level discrete Wavelet transforms to each channel’s sequence and introduce a Wavelet external attention mechanism to capture the correlation of sequences between multiple samples in the Wavelet domain. Additionally, we incorporate patch graph convolution network and Mamba to model local spatial dependencies within patches and long-term dependencies between patches. Extensive experiments on multiple real-world datasets demonstrate that our proposed approach outperforms recent and popular approaches, including Transformer-based, convolutional neural networks-based, multilayer perceptron-based, and cross-domain learning models. Our approach achieves improvements of 5.7%, 16.8%, 17.1%, and 13.2%, respectively, demonstrating its effectiveness in improving forecasting accuracy and capturing broader temporal dependencies. This offers a robust and scalable solution for critical applications in finance, energy, healthcare, and industrial automation.}, archive = {J_EAAI}, author = {Zhigen Huang and Fan Zhang and Yepeng Liu}, doi = {10.1016/j.engappai.2025.110964}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110964}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A channel-independent network based on wavelet enhancement for long-term time series forecasting}, volume = {154}, year = {2025}, }
• (2025). A survey on machine learning methods for food safety risk assessment: Approaches, challenges, and future outlook. EAAI, 154, 110960. (www)

  @article{ , abstract = {Food safety is essential for protecting health and supply chain management. Food engineering plays a foundational role in ensuring food safety by developing innovative processes and technologies for quality control, contamination monitoring, and risk mitigation. Risk assessment is an effective means to ensure food safety, while machine learning (ML) is crucial in facilitating this process. It improves the accuracy of food quality inspection and the speed of risk assessments through rapid learning and processing of data. This survey provides a comprehensive analysis of commonly used supervised and unsupervised learning methods for food safety risk assessment, highlighting advancements, challenges, and future directions. Supervised learning methods, such as Bayesian network (BN), support vector machine (SVM), artificial neural network (ANN), etc., are successfully applied in prediction of food safety risks and improves the prediction accuracy and efficiency. Unsupervised learning methods, such as k-means, hierarchical cluster analysis (HCA), autoencoder (AE), etc., perform well for unlabeled or high-dimensional food anomaly data. The review also addresses key challenges in the food field, such as class imbalance, the emergence of new and unexpected risks, and the integration of multi-source heterogeneous data, including regulatory data, e-commerce sentiment, and public opinion. The utilization of large language models (LLMs), few-shot learning (FSL), and knowledge graphs together offers promising solutions to key challenges in food safety risk assessment. This comprehensive survey emphasizes the transformative potential of ML in enhancing the field of food safety risk assessment and management.}, archive = {J_EAAI}, author = {Zhiyao Zhao and Jiaxin Dong and Bojian Qi and Nuo Duan and He Qian}, doi = {10.1016/j.engappai.2025.110960}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110960}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A survey on machine learning methods for food safety risk assessment: Approaches, challenges, and future outlook}, volume = {154}, year = {2025}, }
• (2025). Forced convective micropolar nanofluidic transportation with cattaneo-christov heat and mass flux model: Levenberg-marquardt backpropagation neural network approach. EAAI, 154, 110947. (www)

  @article{ , abstract = {Micropolar fluids are characterized by their rotational effect and microstructures which exhibit unique heat transfer properties. In various engineering applications where, precise control and prediction of heat flow rates are vital i.e. in cooling technologies, material processing, and energy systems. The current study accounts for the forced convective electrically conducting Cattaneo-Christov model for thermal and solutal flux relaxation time over an expanding surface equipped with porous matrix that provides more realistic representation of thermo-mass transport mechanism. In addition to it, the role of cross diffusion due to Brownian motion, and thermophoresis along with irregular heat emission/release enriches the flow mechanism by incorporating the analysis of velocity slip conditions. The model is transformed into dimensionless form for the assumptions of similarity rules and then the system is solved numerically. Further, Levenberg-Marquardt backpropagation algorithm in training “artificial neural network (ANNs)” to predict the heat flow rate utilizing distinct factors. The anticipated algorithm is efficient and accurate in handling non-linear optimization problems and it is employed to train the ANN, enhancing its predictive capability for complex heat transfer phenomena. However, the important outcomes of the study are reported as; the cross-diffusion lead to the impact of Brownian and thermophoresis are useful in accelerating the thermal transport properties. Further, the heat transport phenomenon along with the heat transfer rate enhances for the augmentation of the thermal radiation.}, archive = {J_EAAI}, author = {Rupa Baithalu and S.R. Mishra and Subhajit Panda and Nehad Ali Shah}, doi = {10.1016/j.engappai.2025.110947}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110947}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forced convective micropolar nanofluidic transportation with cattaneo-christov heat and mass flux model: Levenberg-marquardt backpropagation neural network approach}, volume = {154}, year = {2025}, }
• (2025). A novel framework for selection of renewable energy source based on three-way decision making model under pythagorean fuzzy credibility numbers. EAAI, 154, 110941. (www)

  @article{ , abstract = {An international company is planning a renewable energy project and needs to select the most suitable renewable energy source. Selecting the best option is challenging due to the complexity and uncertainty involved in evaluating multiple criteria. Traditional methods often fall short in handling such complexities, making the decision-making process difficult and risky, because they do not account for this degree of credibility when handling Pythagorean fuzzy numbers. Without a credibility degree, traditional methods cannot effectively represent the level of confidence or uncertainty associated with each criterion, making it challenging to make accurate and reliable decisions. To address this, a novel three-way decision-making model using Pythagorean fuzzy credibility information is proposed. The model introduces Pythagorean fuzzy credibility numbers (PFCNs) and utilizes Dombi aggregation operators (DAOs) to process and combine complex data. PFCNs allow for the representation of uncertain and imprecise information, while DAOs facilitate the aggregation of multiple experts' information into a single aggregated decision framework. The process begins by defining the basic concepts of PFCNs and explaining how DAOs work with these numbers. The proposed model applies this framework to select the best renewable energy source. For this, we collect the three expert opinions on renewable energy sources, and the aggregation operator is used to combine these inputs into a single, consolidated result. To improve the decision-making process, the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) is used to calculate conditional probabilities, which are then applied in the three-way decision-making model to determine the final results. This approach helps rank the renewable energy sources effectively. A sensitivity analysis is then conducted to assess how changes in the Dombi parameter affect the results, ensuring the robustness and stability of the model. Additionally, the proposed decision-making model is compared with existing multi-criteria decision-making techniques to evaluate and validate its effectiveness. This comparison shows how the model improves decision-making in renewable energy selection by providing a more accurate and reliable framework. Overall, this novel model enhances decision-making accuracy by addressing the challenges of complex, uncertain data. It has significant potential for improving renewable energy adoption and offers valuable insights into sustainable energy development.}, archive = {J_EAAI}, author = {Ihsan Ullah and Saleem Abdullah and Marya Nawaz}, doi = {10.1016/j.engappai.2025.110941}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110941}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel framework for selection of renewable energy source based on three-way decision making model under pythagorean fuzzy credibility numbers}, volume = {154}, year = {2025}, }
• (2025). An aspect-level sentiment graph convolutional network model based on transformer and frequency domain. EAAI, 154, 110938. (www)

  @article{ , abstract = {Aspect-based sentiment analysis (ABSA) has long been a crucial research direction in Natural Language Processing (NLP). Its significance lies in unraveling finer-grained sentiments, establishing relationships among different emotional elements, and predicting and analyzing sentiments for specific aspects within text sequences. In recent years, the use of syntactic dependencies and dependency trees to construct neural networks for sentiment analysis tasks has proven to be effective. Simultaneously, using pre-trained large models has significantly improved performance across various text sequence tasks. However, effectively leveraging syntactic dependency information and modeling relevant information in text sequences pose higher demands on the capabilities of network models. Reducing noise generated during this process and enhancing prediction accuracy are crucial challenges. In this paper, we propose an ABSA network model, called Bidirectional Encoder Representations from Transformers-Spectral Graph Convolutional Network (BSGCN), which processes text sequences using Bidirectional Encoder Representations from Transformers (BERT). Incorporating relevant information through a combination of self-attention and aspect-attention, the model employs graph convolutional network (GCN) to enhance information propagation between related nodes, thereby improving the model’s predictive capabilities. Notably, we introduce a spectral layer constructed through a Fast Fourier Transform (FFT) with random weights to reduce noise generated during the data processing. Experimental results based on datasets demonstrate that our proposed model achieves state-of-the-art performance.}, archive = {J_EAAI}, author = {Xiaoyang Wang and Wenfeng Liu}, doi = {10.1016/j.engappai.2025.110938}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110938}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An aspect-level sentiment graph convolutional network model based on transformer and frequency domain}, volume = {154}, year = {2025}, }
• (2025). Enhanced cross-dimensional transformer for long-term wellhead pressure forecasting during hydraulic fracturing. EAAI, 154, 110935. (www)

  @article{ , abstract = {Observational analysis of wellhead pressure variations is crucial for detecting fracturing fluid leakage and assessing wellbore integrity during fracturing operations. Accurately predicting multi-time step pressure changes is essential for enhancing oil and gas extraction efficiency while ensuring wellhead safety. This paper proposes the Enhanced Cross-Dimensional Transformer (ECformer), a time series model that predicts long-term wellhead pressure by capturing time-variable dependencies, the cross-dimensional information, in fracturing operation data. ECformer employs Patch Generation to segment local data regions, enabling the Cross-Dimensional Enhancement Structure to capture latent time-variable relationships. Patch Shuffle and Multi-Scale Fusion further refine temporal extraction and feature-level analysis at the patch scale, thereby enhancing prediction accuracy. Experimental results show that ECformer outperforms the newer Near Time Transformer (NTformer) and Patch Time Series Transformer (PatchTST) in long-term wellhead pressure prediction, reducing the Mean Squared Error (MSE) by 3.5% and 33.4% within 90 s and by 6.6% and 38.5% within 120 s about the average of multiple data sets. ECformer shows higher accuracy in long-step predictions, providing greater potential for future proactive risk prevention.}, archive = {J_EAAI}, author = {Tao Zhang and Yuan Zhong and Jing Zhou and Ping Li and Jie Gong}, doi = {10.1016/j.engappai.2025.110935}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110935}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced cross-dimensional transformer for long-term wellhead pressure forecasting during hydraulic fracturing}, volume = {154}, year = {2025}, }
• (2025). Data-driven forecasting and its role in enhanced decision-making. EAAI, 154, 110934. (www)

  @article{ , abstract = {Decision-making is a crucial process for any organization, since it involves the selection of the most effective action from a variety of options. In this context, data plays an important role in driving decisions. Analyzing data allows us to extract patterns that enable better decision-making for achieving specific goals. However, to make the right decisions to control the behavior of a system, it is necessary to take into account different factors, which can be challenging. Indeed, in dynamic systems, numerous variables change over time, and understanding the future state of these systems can be crucial for controlling the system. Predicting future states based on historical data is known as time series forecasting, which can be divided into univariate and multivariate forecasting, with the latter being particularly relevant due to its consideration of multiple variables. Deep Learning methods enhance decision-making by identifying patterns in complex datasets. As data complexity grows, techniques like Automated Machine Learning optimize model performance. The present study introduces a novel methodology that integrates multivariate time series forecasting into decision-making frameworks. We used Automated-Machine Learning to develop a predictive model for forecasting future system states, aiding optimal decision-making. The study compares machine learning models based on performance metrics and computational cost across various domains, including weather monitoring, power consumption, hospital electricity monitoring, and exchange rates. We also analyzed the importance of the hyperparameters in identifying key factors affecting model performance. The obtained results show that Neural Architecture Search method can improve state predictor design by reducing computational resources and enhancing performance.}, archive = {J_EAAI}, author = {Francesco Zito and Vincenzo Cutello and Mario Pavone}, doi = {10.1016/j.engappai.2025.110934}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110934}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven forecasting and its role in enhanced decision-making}, volume = {154}, year = {2025}, }
• (2025). Harnessing artificial neural networks for improved control of wind turbines based on brushless doubly fed induction generator. EAAI, 154, 110925. (www)

  @article{ , abstract = {Objectives Optimizing the performance of wind turbines equipped with brushless doubly fed induction generators (BDFIGs) remains a key challenge in enhancing energy efficiency and ensuring the stability of wind power systems. Conventional approaches based on proportional-integral (PI) controllers optimized by heuristic algorithms exhibit limitations in terms of response speed and accuracy. This paper aims to overcome these limitations by proposing an artificial neural network (ANN)-based approach for wind turbine power control. Methods This study relies on detailed wind turbine modeling using Python and MATLAB/Simulink. A wind turbine database was created to train an advanced neural architecture composed of five layers of neurons (10, 16, 32, 64, and 2). The proposed ANN controller was evaluated by comparing it to a PI controller optimized by the particle swarm algorithm (PSO). Novelty The developed neural architecture introduces an adaptive optimization model enabling more precise and rapid control of the wind system. Unlike conventional approaches, this method eliminates setpoint overshoot and improves dynamic response. Findings The proposed ANN controller exhibits a significantly faster dynamic response (0.02 s–0.06 s) compared to the PSO-optimized PI controller (0.15 s–0.30 s). Static error is also reduced (0.1 %–1.58 % vs. 2.58 %–3.79 % for the PI). What's more, the RNA model completely eliminates setpoint overshoot, which can reach 734.72 W with a PI controller. Finally, the average response dynamic obtained (0.043 s) far exceeds that of the work by Meroufel et al. (4.83 s); and the proposed neural structure is more extensive than that of sayeh et al., in 2025.}, archive = {J_EAAI}, author = {Ulrich Ngnassi Nguelcheu and Ndjiya Ngasop and Eric Duckler Kenmoe Fankem and Aslain Brisco Ngnassi Djami and Golam Guidkaya and Joseph Yves Effa}, doi = {10.1016/j.engappai.2025.110925}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110925}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Harnessing artificial neural networks for improved control of wind turbines based on brushless doubly fed induction generator}, volume = {154}, year = {2025}, }
• (2025). A novel adaptive spatial–temporal cross-graph convolutional fusion learning network for skeleton-based abnormal gait recognition. EAAI, 154, 110922. (www)

  @article{ , abstract = {Developing graph-based abnormal gait classification models with high generalization has been a challenging problem in gait analysis. In this study, a novel adaptive spatial–temporal cross-graph convolutional fusion learning network is proposed to accurately recognize skeleton-based abnormal gait patterns. In the proposed model, with an adaptive fusion adjacency matrix including self-adaptive adjacency matrices and cross-adaptive adjacency matrices, a joint–bone gait graph convolutional fusion learning algorithm is constructed to capture spatial gait abnormality features hidden in skeleton data. A temporal convolution network is then adopted to explore temporal dependencies of gait abnormality embedded in the spatial feature space. This could discover the most discriminative spatial–temporal gait abnormality representations containing richer information about interaction coupling across joints and bones for high-generalization. The skeleton data of mimic abnormal gait from 57 participants were collected to evaluate the feasibility of our model. The experimental results based on the leave-one-subject-out (LOSO) cross-validation scheme show that our proposed model reaches the optimal performance with the highest accuracy of 99.43%, and significantly outcompetes several recent state-of-the-art models. Our model can feasibly take advantage of the adaptive fusion adjacency matrix to greatly enhance the aggregation degree of joints and bones. This helps to learn excellent gait abnormality representations containing richer interaction information for high generalization while keeping a low learning complexity. Our findings hopefully provide a powerful technical solution for abnormal gait recognition in practical clinical application.}, archive = {J_EAAI}, author = {Liang Wang and Xiaoyan Wu and Bin Wu and Jianning Wu}, doi = {10.1016/j.engappai.2025.110922}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110922}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel adaptive spatial–temporal cross-graph convolutional fusion learning network for skeleton-based abnormal gait recognition}, volume = {154}, year = {2025}, }
• (2025). Stochastic reliability optimization of a controlled memristor-based van der pol circuit using a new intelligent algorithm. EAAI, 154, 110921. (www)

  @article{ , abstract = {Enhancing the reliability of structural dynamical systems under random loads via controllers is essential for maintaining system stability, resilience, performance, and safety. The challenge lies in optimally designing the controller while determining reliability probability. To address this issue, our study develops a new intelligent algorithm for optimal reliability analysis on a controlled memristor-based Van der Pol circuit under stochastic excitation. This algorithm integrates Gaussian Radial Basis Function Neural Network with Genetic Algorithm, the reliability function with control parameters serves as objective function, while the reliability probability equation acts as constraint conditions, meanwhile the fitness function is derived from the neural network’s solution to the governing equation. Our algorithm achieves optimal reliability results and overcomes the challenge of simultaneously optimizing an implicit objective and solving the reliability governing equation. The sensitivity of key parameters such as population size, maximum iteration to reliability performance are discussed, respectively. The effectiveness of the proposed algorithm is numerically compared with Monte-Carlo simulation and finite difference method. The control strategy in this work establishes a theoretical foundation for systems exhibiting random vibration and probabilistic fatigue in structural engineering, and holds promising potential for real-world application.}, archive = {J_EAAI}, author = {Wei Li and Mingzhi Lin and Junfeng Zhao and Drazan Kozak}, doi = {10.1016/j.engappai.2025.110921}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110921}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Stochastic reliability optimization of a controlled memristor-based van der pol circuit using a new intelligent algorithm}, volume = {154}, year = {2025}, }
• (2025). Adaptive convolutional network pruning through pixel-level cross-correlation and channel independence for enhanced model compression. EAAI, 154, 110920. (www)

  @article{ , abstract = {In deep learning model optimization, the majority of pruning techniques employ predefined criteria to guide filter elimination. Although effective in identifying redundant filters, these methods often overlook intrinsic linear redundancy embedded within the network during early training stages, thus not fully exploiting the latent capacity of the model. To address this limitation, we propose an innovative structured pruning approach grounded in a Pixel-Level Cross-Correlation(PCC) loss term. By incorporating this data-driven loss function in the pre-training phase, our method promotes the formation of more robust intrinsic connections among feature maps. Leveraging PCC of output features, the network discerns finer inter-pixel relationships. This approach enhances feature map granularity, refining local feature representation and overall map quality. During the pruning phase, we employ a channel screening strategy based on Channel Independence (CI), which allows the classification of channels by importance and preserves adequate linear redundancy. Following pruning, the model is retrained and fine-tuned. Compared to state-of-the-art approaches, our method achieves substantial compression while retaining performance. For instance, on the Canadian Institute For Advanced Research 10 (CIFAR-10) dataset, the Visual Geometry Group 16-layer network (VGG-16) model demonstrates a 0.87% improvement in pre-training accuracy (94.59%) and achieves 93.40% accuracy post-pruning, with pruning rates of 92.75% for parameters(Params) and 84.61% for Floating Point Operations (FLOPs). Similarly, on the ImageNet(LSVRC-2012) dataset, our method reduces Params and FLOPs of Residual Network-50(ResNet-50) by 60.24% and 58.40%, respectively. This demonstrates a significant reduction in model size and computational complexity while maintaining high performance.}, archive = {J_EAAI}, author = {Guangyao Li and Haijian Shao and Xing Deng and Yingtao Jiang}, doi = {10.1016/j.engappai.2025.110920}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110920}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive convolutional network pruning through pixel-level cross-correlation and channel independence for enhanced model compression}, volume = {154}, year = {2025}, }
• (2025). A lightweight deep learning framework for wild berry detection in complex natural environments. EAAI, 154, 110918. (www)

  @article{ , abstract = {Wild berries (WildB) play a crucial role in Nordic forest ecosystems’ ecological and economic balance. However, research on wild berry detection has remained scarce in recent years. Driven by recent advancements in Artificial Intelligence (AI), particularly in deep learning-based computer vision applications, we propose a detection model based on the You Only Look Once version 11n (YOLOv11n) architecture (WildB-YOLO). WildB-YOLO integrates multiple innovations: the Frog Feature Pyramid Network (FrogFPN) improves multi-scale feature fusion, facilitating the detection of objects at various scales. The Scale-Aware Context Module (SACM) enhances contextual modeling, improving target discrimination. The Weighted Exponential Moving Average Loss (WEMA Loss) mitigates class imbalance. Additionally, Soft Non-Maximum Suppression (Soft-NMS) refines bounding box selection and reduces false positives, enhancing overall detection performance. The model employs Layer Adaptive Magnitude-Based Pruning (LAMP) to enhance efficiency further, achieving lightweight optimization while maintaining high detection precision. Experimental results demonstrate that WildB-YOLO achieves a mean Average Precision (mAP) of 59.5% at Intersection over Union (IoU) thresholds ranging from 50% to 95% (mAP50-95), outperforming the original YOLOv11n by 1.9%. Furthermore, WildB-YOLO’s optimized model size is 1.5MB, with 2.5G floating point operations per second (FLOPs) and 2.6M parameters, representing reductions of 71.1%, 60.3%, and 76.9%, respectively, compared to the baseline model. This reduction in complexity facilitates deployment on resource-constrained devices, significantly enhancing the applicability of AI-driven berry detection in practical field scenarios. This study pioneers a dedicated solution for wild berry detection and contributes novel strategies for small object detection in complex natural environments. WildB-YOLO is open-source at: https://github.com/zxr0826/WildB-YOLO .}, archive = {J_EAAI}, author = {Xiaorong Zhang and Fei Li and XuTing Hu and Juan Fang}, doi = {10.1016/j.engappai.2025.110918}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110918}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight deep learning framework for wild berry detection in complex natural environments}, volume = {154}, year = {2025}, }
• (2025). SARA: Span-aware framework with relation-augmented grid tagging for conversational aspect-based sentiment quadruple analysis. EAAI, 154, 110915. (www)

  @article{ , abstract = {Conversational aspect-based sentiment quadruple analysis (DiaASQ) is an emerging research topic in the field of aspect-based sentiment analysis (ABSA), which aims to extract sentiment quadruples (i.e., target , aspect , opinion , and sentiment polarity ) from a given dialogue. In DiaASQ, numerous span quadruples can be observed, where the constituent elements of these quadruples are often dispersed across a single utterance or span multiple utterances within a conversation, posing significant challenges to previous methods. In addition, existing methods typically focus on linking the target term with the opinion term, while overlooking the direct relationship between the aspect term and the opinion term, leading to inaccurate extraction of quadruples. Therefore, this paper proposes a Span-Aware framework with Relation-Augmented grid tagging scheme (SARA) to enhance DiaASQ. Specifically, a novel span-aware memory block is devised to enhance the model’s ability to obtain the long-distance dependency among elements within span quadruples. Furthermore, a new relation-augmented grid tagging scheme is introduced to strengthen the pairing of aspect terms and opinion terms, thereby enhancing the performance of quadruple extraction. Extensive experimental results on two benchmark datasets validate the effectiveness of the proposed approach.}, archive = {J_EAAI}, author = {Xiaoyong Liu and Miao Hu and Chunlin Xu and Zhiguo Du}, doi = {10.1016/j.engappai.2025.110915}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110915}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SARA: Span-aware framework with relation-augmented grid tagging for conversational aspect-based sentiment quadruple analysis}, volume = {154}, year = {2025}, }
• (2025). Time series representation learning via cross-domain predictive and contextual contrasting: Application to fault detection. EAAI, 154, 110914. (www)

  @article{ , abstract = {Data-driven methods for fault detection increasingly rely on large historical datasets, yet annotations are costly and time-consuming. As a result, learning approaches that minimize the need for extensive labeling, such as self-supervised learning (SSL), are becoming more popular. Contrastive learning, a subset of SSL, has shown promise in fields like computer vision and natural language processing (NLP), yet its application in fault detection is not fully explored. In this paper, we introduce Cross-Domain Predictive and Contextual Contrasting (CDPCC), a novel contrastive learning framework that integrates temporal and spectral information to capture informative time-frequency features from time series data. CDPCC consists of two key components: cross-domain predictive contrasting, which predicts future embeddings across time and frequency domains, and cross-domain contextual contrasting, which aligns time- and frequency-based representations in a shared latent space. We evaluate CDPCC on fault detection tasks using both simulated and industrial datasets. Our results show that a linear classifier trained on features learned by CDPCC performs comparably to fully supervised models. Moreover, CDPCC proves highly effective in scenarios with limited labeled data, achieving superior performance with only 50% of the labeled data compared to fully supervised training on the entire dataset. The source code is publicly available at https://github.com/iy641/CDPCC.git .}, archive = {J_EAAI}, author = {Ibrahim Yousef and Sirish L. Shah and R. Bhushan Gopaluni}, doi = {10.1016/j.engappai.2025.110914}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110914}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Time series representation learning via cross-domain predictive and contextual contrasting: Application to fault detection}, volume = {154}, year = {2025}, }
• (2025). PDFusion: A domain-adaptive incremental learning model based on physical-data fusion for lithium-ion battery state estimation. EAAI, 154, 110913. (www)

  @article{ , abstract = {Accurate estimation of battery state is crucial for ensuring the safe, stable, and efficient operation of lithium-ion batteries. State of Charge (SOC) and State of Energy (SOE) are critical parameters for assessing battery health, but accurately estimating them remains challenging due to the nonlinear, non-stationary and strong coupling characteristics of complex battery charging and discharging processes. To address these issues, a novel domain-adaptive incremental learning model driven by both physical and data is proposed. To reduce state noise and covariance, a novel Kalman Filtering method is used for primary trend prediction. However, physics-based models fail to estimate the seasonal components that contain time-frequency patterns. To overcome the limitation, a data-driven model with Time-frequency Interactive Attention (TIA) is proposed to accurately capture the temporal relationships and effectively compensate for peak errors. To make the model operate across different temperature conditions, a domain-adaptive incremental learning strategy is employed. The results on Lithium iron phosphate (LFP) and Nickel Cobalt Manganese (NCM) batteries show that the proposed model outperforms current state-of-the-art (SOTAs), with the average Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) of 0.048 and 0.021 for LFP under two operating conditions, and 0.060 and 0.023 for NCM. Under Beijing Dynamic Stress test (BJDST) conditions, RMSE and MAE are reduced by 22.69% and 28.56% respectively. Under US06 Highway Driving Schedule (US06) conditions, these metrics are reduced by 41.02% and 49.09%, respectively. The algorithm exhibits high robustness to temperature, enabling precise estimation of the lithium-ion battery states.}, archive = {J_EAAI}, author = {Yufei Xie and Wenlin Wang and Guohua Wu and Haichuan Zhang}, doi = {10.1016/j.engappai.2025.110913}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110913}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {PDFusion: A domain-adaptive incremental learning model based on physical-data fusion for lithium-ion battery state estimation}, volume = {154}, year = {2025}, }
• (2025). Multi-scale low-frequency enhanced spectral neural operator for reducing low-frequency error in partial differential equations solving. EAAI, 154, 110912. (www)

  @article{ , abstract = {Designing universal artificial intelligence (AI) solver for partial differential equations (PDEs) is an open problem and a significant challenge in science and engineering. AI-inspired data-driven solvers, such as neural operators, have achieved great success in PDE solving. However, insufficient low-frequency learning ability and inability to utilize physical prior knowledge remain an obstacle for the PDEs solver which designed by neural operator. To tackle this challenge, we drew inspiration from the multigrid method and developed the multi-scale low-frequency enhanced spectral neural operator. Our approach broadens the neural operator’s learnable range in the frequency domain by folding the frequency spectrum, thereby reducing low-frequency error through a carefully designed residual structure. Additionally, to adapt to the varying spectral distributions of different PDEs, we propose a neural operator correction strategy based on the correspondence between the PDE spectrum distribution pattern and the neural operator learning pattern in the low-frequency region, which we summarized, to correct the results by utilizing the prior knowledge of the PDE. Extensive experiments on benchmark fluid datasets, including the Darcy equation, the Navier–Stokes equation, and their variants, demonstrate that our model achieves a 26.7% reduction in low-frequency error and a 25.6% improvement in accuracy, outperforming traditional neural operators and verifying the effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Fengrui Jing and Chuchu Zhai and Peizhi Zhao and Xue Li and Peifu Han and Hongzhen Ding and Yunlong Dong and Long Hao and Xu Tian and Tao Song}, doi = {10.1016/j.engappai.2025.110912}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110912}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale low-frequency enhanced spectral neural operator for reducing low-frequency error in partial differential equations solving}, volume = {154}, year = {2025}, }
• (2025). A dual two-stage attention-based model for interpretable hard landing prediction from flight data. EAAI, 154, 110911. (www)

  @article{ , abstract = {Hard landings are a significant safety concern in aviation, with potential consequences ranging from poor passenger experiences to serious injuries or fatalities. Predicting and explaining hard landing events are equally important for enhancing flight safety, the former makes it possible to give proactive warnings, while the latter helps pilots identify the reasons and refine their skills. However, existing studies generally lack a comprehensive consideration for the fine-grained characteristics of flight data containing both inter-temporal and inter-parametric relationships, resulting in suboptimal prediction performance. In addition, most of existing approaches aim at improving the prediction performance but fail to provide interpretability for the causes of hard landing. To address the above problems, we propose DUTSAM , a DU al T wo- S tage A ttention-based interpretable M odel for hard landing prediction from quick access recorder (QAR) data. The model consists of dual parallel modules, each of which combines a convolutional feature encoder and a two-stage attention mechanism. The two encoders capture fine-grained characteristics by encoding multivariate data from temporal domain and parametric domain respectively. After that, the dual two-stage attention mechanism captures the inter-temporal and inter-parametric correlations in reverse order to predict hard landing and provide interpretation from both temporal and parametric perspectives. Experimental results on a real QAR dataset with 37,920 flights show that DUTSAM achieves better prediction performance compared with other state-of-the-art baselines in terms of Precision, Recall, and F1-score. Additionally, case study demonstrates that DUTSAM can uncover key flight parameters and moments strongly correlated to the hard landing events.}, archive = {J_EAAI}, author = {Jiaxing Shang and Xiaoquan Li and Ruixiang Zhang and Linjiang Zheng and Xu Li and Riquan Zhang and Xinbin Zhao and Fan Li and Hong Sun}, doi = {10.1016/j.engappai.2025.110911}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110911}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual two-stage attention-based model for interpretable hard landing prediction from flight data}, volume = {154}, year = {2025}, }
• (2025). Symmetric positive definite convolutional network for surrogate modeling and optimization of modular structures. EAAI, 154, 110906. (www)

  @article{ , abstract = {While modular structures offer great construction efficiency, scalability, safety, and reusability in engineering and architectural applications, their wide-spread adoption is hindered by the perceived material inefficiency and low design flexibility. Finding an optimal design within a modular system is a significant challenge, mostly because of associated computational complexity. Existing methods of accelerating combinatorial optimization with machine learning rely on heuristics and are often not transferrable between varying domain shapes, boundary conditions, and external loads. In this work, we present two key contributions to address this issue: (i) a deep neural network (DNN)-based surrogate model that accelerates the evaluation of mechanical responses by predicting reduced-order stiffness matrices, and (ii) a stochastic gradient optimization method that leverages the surrogate’s capability to compute sensitivities of the structure’s response to changes in module types. Our model combines convolutional layers with a physics-guided approach, ensuring that the output stiffness matrices are symmetric positive definite, consistent with the structure’s reduced-order representation via Schur’s complement. A distinguishing feature of our approach is its intrinsic independence from the specific domain shape, boundary conditions, and applied loads, allowing for broader applicability once the DNN-based surrogate is trained on a specific module set. We validate our method by optimizing multiple modular layout plans differing in size and loading conditions and demonstrate its efficacy by comparing its performance against the standard density-based topology optimization method. We achieve a computational speed-up of up to 1000x compared to the full-scale simulation, with a fast converging optimization for different domain sizes. This work lays the foundation for more flexible, efficient, and scalable modular design processes.}, archive = {J_EAAI}, author = {Liya Gaynutdinova and Martin Doškář and Ivana Pultarová and Ondřej Rokoš}, doi = {10.1016/j.engappai.2025.110906}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110906}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Symmetric positive definite convolutional network for surrogate modeling and optimization of modular structures}, volume = {154}, year = {2025}, }
• (2025). Dual prototypes contrastive learning based semi-supervised segmentation method for intelligent medical applications. EAAI, 154, 110905. (www)

  @article{ , abstract = {In medical intelligence applications, the labeling of medical data is crucial and expensive, so it becomes urgent to explore labeling-efficient ways to train applications. Semi-supervised techniques for medical image segmentation have demonstrated potential, effectively training models using scarce labeled data alongside a wealth of unlabeled data. Therefore, semi-supervised medical image segmentation is a key issue in engineering applications of medical intelligence. Consistency constraints based on prototype alignment provide an intuitively sensible way to discover valuable insights from unlabeled data that can motivate segmentation performance. In this work, we propose a Dual prototypes Contrastive Network to motivate semi-supervised medical segmentation accuracy by imposing image-level global prototype and pixel-level local prototype constraints. First, we introduce a Background-Separation Global Prototype Contrastive Learning technique that utilizes the natural mutual exclusivity of foreground and background to separate the inter-class distances and encourage the segmentation network to obtain segmentation results that are more complete and do not contain background regions. Second, we design a Cross-Consistent Local Prototype Contrastive Learning techniques to extend the perturbation consistency of the two networks to the prototype’s localized response to the feature map, thereby shaping a more stable intra-class prototype space and producing accurate and robust pixel-level predictions. Finally, we comprehensively evaluate our method on mainstream semi-supervised medical image segmentation benchmarks and settings, and experimental results show that our proposed method outperforms current state-of-the-art methods. Specifically, our method achieves a Dice Coefficient score of 91.8 on the Automatic Cardiac Diagnosis Challenge dataset using only 10% labeled data training, 1.1% ahead of the second best method. Code is available at https://github.com/yuelily2024/DPC .}, archive = {J_EAAI}, author = {Tianai Yue and Rongtao Xu and Jingqian Wu and Wenjie Yang and Shide Du and Changwei Wang}, doi = {10.1016/j.engappai.2025.110905}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110905}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual prototypes contrastive learning based semi-supervised segmentation method for intelligent medical applications}, volume = {154}, year = {2025}, }
• (2025). Autonomous dynamic formation for maritime target tracking using multi-agent reinforcement learning. EAAI, 154, 110904. (www)

  @article{ , abstract = {In various maritime missions such as escort and roundup, dynamic formation target tracking plays a crucial role. Most existing dynamic formation methods require user intervention before formation changes, resulting in poor flexibility and low automation. And they do not consider variations in the abilities of individual members. To address the above issue, we propose an autonomous dynamic formation planning method based on multi-agent reinforcement learning, integrating formation configuration into the strategy. This method can automatically adjust the formation based on the current state of the formation, providing greater flexibility and adaptability. Simultaneously, a staged reward function is devised for the training process to guide agents in progressively learning dynamic formation tasks. Finally, we validate the effectiveness and generalization of our proposed method through various experiments.}, archive = {J_EAAI}, author = {Hua Wang and Jiaxin Li and Hao Tao and Junnan Liu and Chaochao Li and Ke Wang and Mingliang Xu}, doi = {10.1016/j.engappai.2025.110904}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110904}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Autonomous dynamic formation for maritime target tracking using multi-agent reinforcement learning}, volume = {154}, year = {2025}, }
• (2025). Possibilistic c-means clustering approach based on a novel weighted-kernel distance for imbalanced images with minority targets in sparsely distribution. EAAI, 154, 110902. (www)

  @article{ , abstract = {Possibilistic c-means clustering (PCM), a classic partition clustering algorithm, is an important data mining technique for unsupervised image segmentation in the field of artificial intelligence. The typicalities (memberships) of PCM have better description of local information and excellent noise resistance due to its absolute attribute. However, it still faces the challenges of segmenting color images with multiple characteristics, such as feature imbalance, cluster-size imbalance, noise attack, especially sparsely distribution of minority targets in the feature space. Therefore, this paper introduces a weighted-kernel distance-based PCM (WK-PCM) algorithm. Firstly, a weighted-kernel distance (WK-distance) is defined by combining the absolute attribute of typicalities and the Gaussian kernel function to enhance the intra-class compactness of sparse targets. Meanwhile, the feature weighting scheme in the WK-distance is expected to avoid center offset caused by imbalanced features. Secondly, to overcome the issue of center overlapping caused by insufficient interclass relationships of typicalities, the cutset theory is introduced based on the WK-distance to select partial objects. Then the typicalities of these selected objects are suppressed to increase inter-class separateness and avoid the issue of coincident clustering (also called center overlapping). Finally, an improved WK-PCM image segmentation algorithm (LWK-PCM) based on local spatial information acquired through bilateral filtering is proposed for imbalanced color images with noise corruption. Experiments conducted on synthetic datasets and imbalanced color images indicate that the proposed WK-PCM and LWK-PCM algorithms get excellent clustering performance compared to the relevant clustering algorithms.}, archive = {J_EAAI}, author = {Haiyan Yu and Yuting Wu and Haocong Zheng and Qianqian Luo and Lu Zhang}, doi = {10.1016/j.engappai.2025.110902}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110902}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Possibilistic c-means clustering approach based on a novel weighted-kernel distance for imbalanced images with minority targets in sparsely distribution}, volume = {154}, year = {2025}, }
• (2025). Four-dimension efficient pixel-frequency transformer for light field spatial and angular super-resolution. EAAI, 154, 110901. (www)

  @article{ , abstract = {Super-resolution plays an important role in optical engineering, which can offer high-quality image information for vision applications. The light field camera can capture the intensity and direction information of light, providing abundant scene information. However, it faces the trade-off problem of spatial and angular resolution. Benefiting from global context modeling by self-attention mechanism, transformer-based artificial intelligence methods for the light field (LF) super-resolution (SR) provide superior performance. However, in these methods, the four-dimensional (4D) LF is reshaped into a large one-dimensional (1D) sequence to be processed, which leads to a significant increase in computation and memory and raises difficulties of 4D structure reconstruction for LF spatial and angular SR. This paper presents a 4D efficient pixel-frequency transformer (4DEPFT) for spatial and angular SR, which models 4D correlation in the pixel and frequency domains based on the proposed inner-outer transformer with 4D position encoding. Specifically, two parallel branches separately handle LF images and their corresponding frequency maps in 4DEPFT. In the pixel branch, sub-patch-based inner-outer pixel transformers are proposed to extract 4D pixel features inside and outside local windows, which deal with pixels with small and large disparity shifts separately. In the frequency branch, a multi-band-based inner-outer frequency transformer is proposed to reconstruct 4D frequency features from different frequency bands. Finally, an attention-based pixel and frequency feature fusion module is presented to acquire the high-resolution LF. Experimental results on real-world and simulated public datasets indicate that 4DEPFT outperforms other methods with less computational cost and graphics processing unit (GPU) consumption.}, archive = {J_EAAI}, author = {Sizhe Wang and Hao Sheng and Ruixuan Cong and Da Yang and Rongshan Chen and Zhenglong Cui}, doi = {10.1016/j.engappai.2025.110901}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110901}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Four-dimension efficient pixel-frequency transformer for light field spatial and angular super-resolution}, volume = {154}, year = {2025}, }
• (2025). A feature matching-based method for few-shot multivariate time series anomaly detection with symmetric patch mask siam transformer. EAAI, 154, 110894. (www)

  @article{ , abstract = {Accurate anomaly detection of industrial system operating status based on multivariate time series data is an important means to ensure the stable operation of the system. However if there is insufficient training data for the objects to be detected, it is difficult for existing deep learning methods to learn a clear outline of the normal pattern of the data under unsupervised conditions, leading to the failure of anomaly detection. This paper proposes a feature matching-based method for few-shot multivariate time series anomaly detection with a symmetric patch mask Siam Transformer (SPMST). Using only a small number of normal samples from the target domain, SPMST realizes the rapid deployment of the universal representation model pre-trained on multiple public datasets to the target domain without the need for retraining or parameter adjustment for more categories. First, two augmented views of the original data are obtained by adding a symmetric patch mask to the augmented aligned multisource data. The Transformer model is then pre-trained with reconstruction and contrastive learning tasks to acquire robust latent representations. Second, the feature support set of the target domain is obtained based on the pre-trained representation model and the proposed clustering-based support set reduction strategy, avoiding excessive consumption of computing resources. Finally, the anomaly score is calculated by combining the feature matching loss, reconstruction loss, and contrastive loss. The experimental results show that SPMST, under few-shot conditions, is not weaker than 21 state-of-the-art baselines trained with a large amount of data on 5 representative cyber–physical system datasets.}, archive = {J_EAAI}, author = {Jiahao Yu and Xin Gao and Taizhi Wang and Heping Lu and Baofeng Li and Feng Zhai and Bing Xue and Zhihang Meng}, doi = {10.1016/j.engappai.2025.110894}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110894}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A feature matching-based method for few-shot multivariate time series anomaly detection with symmetric patch mask siam transformer}, volume = {154}, year = {2025}, }
• (2025). Prediction of heart failure using voting ensemble learning models and novel data normalization techniques. EAAI, 154, 110888. (www)

  @article{ , abstract = {Early diagnosis and prediction of heart failure are crucial for improving the quality of life and extending the lifespan of patients. However, due to the complexity of the disease and the variability of risk factors among individuals, developing accurate prediction models remains a significant challenge. Machine learning techniques have been widely applied in heart failure prediction, but their performance is highly dependent on the quality of data preprocessing before model training. In this paper, we propose two novel data normalization methods named Squash-Norm and Squash-Norm-Sigmoid and apply them to 6 single machine learning models and soft-voting ensemble models. Experiments were conducted on a clinical dataset, with 918 samples with 11 features, compiled from 5 sub-datasets (Cleveland, Hungarian, Switzerland, Long Beach, and Stalog Data Set). Results show that these new normalization methods significantly improve the performance of the models compared to the Min–Max scaler. When applying the Squash-Norm and Squash-Norm-Sigmoid normalization techniques, the metrics such as Accuracy, Recall, F1-Score, and Balanced Accuracy for all 6 single models showed significant improvement compared to using the Min–Max scaler. Notably, in the case of ensemble models, these metrics also demonstrated substantial enhancements when applying the two normalization methods compared to Min–Max scaler. Specifically, the single models utilizing Squash-Norm and Squash-Norm-Sigmoid achieved the highest Accuracy of 94.0217%, while the ensemble model reached a peak Accuracy of 95.6522% with both normalization techniques. This paper introduces two entirely new data normalization methods combined with ensemble models, proven to be effective in predicting heart failure. These methods not only enhance prediction accuracy but also hold great potential for broader applications in various machine learning tasks.}, archive = {J_EAAI}, author = {Kien Cao-Van and Le Gia Minh and Tran Cao Minh and Ha Minh Tan}, doi = {10.1016/j.engappai.2025.110888}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110888}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction of heart failure using voting ensemble learning models and novel data normalization techniques}, volume = {154}, year = {2025}, }
• (2025). Hybrid gaussian quantum particle swarm optimization and adaptive genetic algorithm for flexible job-shop scheduling problem. EAAI, 154, 110882. (www)

  @article{ , abstract = {Flexible job-shop scheduling problem (FJSP) is a critical challenge in both modern manufacturing and artificial intelligence (AI), characterized by its non-deterministic polynomial-time hard nature. Despite advancements in optimization algorithms, balancing solution quality, computational efficiency, and convergence speed remains challenging, especially for large-scale FJSP instances. Although existing methods are effective in certain scenarios, they often struggle to balance global exploration and local refinement within the solution space. This study proposes a hybrid Gaussian quantum particle swarm optimization and adaptive genetic algorithm (HGQPSO-AGA) to address these limitations. HGQPSO-AGA leverages the global search capability of Gaussian quantum particle swarm optimization (GQPSO) inspired by quantum mechanics with the local refinement strength of adaptive genetic algorithm (AGA). To further enhance performance, the algorithm employs a chaotic encoding scheme, a multi-subpopulation selection mechanism, and adaptive crossover and mutation operators. The performance of HGQPSO-AGA is evaluated on three widely used FJSP benchmark datasets (Kacem, Brandimarte, and Dauzere-Peres) and validated through an industrial case study. The results demonstrate that HGQPSO-AGA outperforms comparison algorithms significantly in terms of solution quality, computational efficiency, and convergence speed. Specifically, HGQPSO-AGA achieves average makespan reductions of 2.40, 7.84, and 110.49 for Kacem, Brandimarte, and Dauzere-Peres datasets, respectively, and 5.84 in the industrial case. These findings indicate that HGQPSO-AGA provides a robust and efficient solution to FJSP, with wide-ranging implications for optimizing complex scheduling problems in manufacturing and AI-driven applications.}, archive = {J_EAAI}, author = {Yuanxing Xu and Mengjian Zhang and Deguang Wang and Ming Yang and Chengbin Liang}, doi = {10.1016/j.engappai.2025.110882}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110882}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid gaussian quantum particle swarm optimization and adaptive genetic algorithm for flexible job-shop scheduling problem}, volume = {154}, year = {2025}, }
• (2025). Sample less meta-learning fault diagnosis based on ordered time–frequency features. EAAI, 154, 110881. (www)

  @article{ , abstract = {To address the challenges of insufficient fault samples and incomplete feature acquisition that compromise diagnostic accuracy, this study proposes a meta-learning framework for fault diagnosis integrating temporally-ordered time–frequency feature representation and a self-attention-enhanced multi-scale network. The proposed methodology systematically extracts time–frequency features from raw vibration signals and organizes them into discriminative two-dimensional (2D) image representations through structured temporal sequencing. A neural architecture combining spatial self-attention mechanisms with multi-scale convolutional feature extraction is developed, enhanced by meta-learning strategies to optimize parameter initialization. The model undergoes primary training on publicly available benchmark datasets to establish generalized feature representations, followed by task-specific fine-tuning and evaluation using targeted diagnostic datasets. Comprehensive experimental validation demonstrates the efficacy of the proposed approach, with the ordered time–frequency feature extraction achieving superior precision of 98.28%. The trained network exhibits exceptional few-shot learning capabilities, and when only a single fault sample is available, it attains a maximum diagnostic accuracy of 81.69%, which significantly outperforms conventional methods. Comparative analyses reveal enhanced adaptability and generalization capacity across diverse operational conditions, confirming the framework’s robustness in addressing data scarcity challenges inherent in industrial fault diagnosis scenarios.}, archive = {J_EAAI}, author = {Cheng Wang and Neng Wang and Lili Deng}, doi = {10.1016/j.engappai.2025.110881}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110881}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sample less meta-learning fault diagnosis based on ordered time–frequency features}, volume = {154}, year = {2025}, }
• (2025). U-bilateral attention gate nested U-transformers for medical image segmentation. EAAI, 154, 110880. (www)

  @article{ , abstract = {Automatic medical image segmentation facilitates accurate computer-aided diagnosis and computer-assisted surgery. Current segmentation approaches, predominantly based on convolutional neural networks, often struggle with capturing long-range or global features effectively. We propose a compact yet powerful deep learning model, the U-bilateral attention gate (U-BAG) nested U-transformers (UnUFormer), specifically designed for abdominal image segmentation. In this model, we introduce the U-bilateral attention gate module, which combines an encoder–decoder U-structure with a bilateral attention gate. Additionally, we develop a novel convolutional transformer module. This module replaces the conventional multihead self-attention (token mixer) with one-dimensional convolution for more efficient feature fusion. This convolutional transformer module, working in tandem with skip connections at multiple stages, adeptly identifies both global and local features. Crucially, it also reduces the computational overhead associated with multihead self-attention. We conducted evaluations of our proposed method on two publicly available clinical datasets. The experimental results demonstrate that UnUFormer significantly outperforms other network methods. It notably improves the average dice similarity coefficient and average symmetric surface distance in abdominal multiorgan segmentation from (94.5%, 0.49) to (95.2%, 0.36).}, archive = {J_EAAI}, author = {Wenkang Fan and Haichao Peng and Rui Li and Yong Peng and Jie Luo and Xiongbiao Luo}, doi = {10.1016/j.engappai.2025.110880}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110880}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {U-bilateral attention gate nested U-transformers for medical image segmentation}, volume = {154}, year = {2025}, }
• (2025). Improving the accuracy of hand sign recognition by chaotic swarm algorithm-based feature selection applied to fused surface electromyography and force myography signals. EAAI, 154, 110878. (www)

  @article{ , abstract = {This study addresses the challenge of distinguishing ambiguous American Sign Language (ASL) gestures by integrating surface electromyography (sEMG) and force myography (FMG) signals. A Chaotic Salp Swarm Algorithm (CSSA) with six chaotic maps was employed for feature selection, followed by k-Nearest Neighbors (kNN) classification. The objective of this approach was to enhance the classification accuracy of 20 American Sign Language (ASL) hand signs using individual and combined sEMG and FMG signals. The experiment involved the collection of data from ten participants, with each participant repeating the ASL postures ten times. The results of the study demonstrated that the CSSA effectively reduced the number of features required for classification while slightly improving the classification accuracy. The fusion of sEMG and FMG achieved an average accuracy of 88.5%, surpassing the accuracy of individual sEMG (81%) and FMG (76.75%) signals. Moreover, after systematically eliminating eight ambiguous gestures, the model attained 100% accuracy. These findings underscore the potential of integrating sEMG and FMG signals with CSSA-based feature selection to enhance ASL gesture recognition. Furthermore, the study identifies specific ambiguous postures that must be excluded to achieve perfect classification.}, archive = {J_EAAI}, author = {Hiba Hellara and Rim Barioul and Salwa Sahnoun and Ahmed Fakhfakh and Olfa Kanoun}, doi = {10.1016/j.engappai.2025.110878}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110878}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving the accuracy of hand sign recognition by chaotic swarm algorithm-based feature selection applied to fused surface electromyography and force myography signals}, volume = {154}, year = {2025}, }
• (2025). Anti-vibration hammer defect detection based on structural knowledge representation. EAAI, 154, 110875. (www)

  @article{ , abstract = {Anti-vibration hammer defects pose a significant risk to the safe operation of power transmission lines. Addressing the issues posed by the various manifestations of identical category defects and the similarities among different defects, this paper introduces a defect detection algorithm for anti-vibration hammers based on structural knowledge representation. Initially, a Structural Knowledge Enhancement (StKE) Module is proposed to conduct a statistical analysis of the aspect ratios of different defects in anti-vibration hammers, effectively extracting the structural features of the anti-vibration hammers and their corresponding defects. Subsequently, a Structural Knowledge Representation (StKR) Module is introduced, which bolsters the model’s ability to precisely locate defects that disrupt structural symmetry. The model incorporates the Coordinate Attention (CA) mechanism to acquire contextual information, thereby enhancing detection accuracy. Experimental results demonstrate that the improved model achieves a 6.1% increase in mean detection precision over the baseline model, and a notable improvement in detection accuracy compared to other advanced algorithms.}, archive = {J_EAAI}, author = {Zhenbing Zhao and Guangxue Guo and Yitian Pan and Ke Zhang and Yongjie Zhai and Wenqing Zhao}, doi = {10.1016/j.engappai.2025.110875}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110875}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Anti-vibration hammer defect detection based on structural knowledge representation}, volume = {154}, year = {2025}, }
• (2025). Multi-occupant tracking with radar and wearable devices for enhanced accuracy in indoor environments. EAAI, 154, 110872. (www)

  @article{ , abstract = {This work explores the integration of millimetre-wave (mmWave) radar and a minimal configuration of ultra-wideband (UWB) devices for enhanced multi-occupant tracking in real domestic environments. Using a low-cost, non-intrusive, and rapidly deployable device setup, our approach addresses key challenges in multi-occupant tracking, including individual identification and ease of installation. While mmWave radar precisely detects occupant presence, it lacks individual recognition and exhibits limited sensitivity. This limitations are addressed by incorporating a minimal configuration of UWB (wearable tags and ambient anchors), enabling individual identification through signal strength measurements. Several data autoencoder models, such as long short-term memories (LSTMs), Convolutional Neural Networks (CNN) or Transformers, were evaluated. Experiments conducted in two real-world domestic settings, each with up to three inhabitants, demonstrate the effectiveness of combining mmWave and UWB technologies for indoor multi-occupant tracking. Our results show that ConvLSTM achieves the best performance with a mean squared error (MSE) between 0,0142 and 0,0433 in single and multi-occupation, respectively. These findings suggest promising applications for accurate inhabitant tracking in ambient assisted living and other smart environment contexts.}, archive = {J_EAAI}, author = {Aurora Polo-Rodríguez and Miguel Ángel Anguita-Molina and Ignacio Rojas-Ruiz and Javier Medina-Quero}, doi = {10.1016/j.engappai.2025.110872}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110872}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-occupant tracking with radar and wearable devices for enhanced accuracy in indoor environments}, volume = {154}, year = {2025}, }
• (2025). A novel adaptive noise model selection framework for blind denoising of scanning electron microscopy images. EAAI, 154, 110871. (www)

  @article{ , abstract = {Scanning Electron Microscopy (SEM) is crucial for imaging micro-nano particles in material science, wherein faster dwell times increase noise, while slower dwell times increase the risk of sample damage. The dwell time is defined as the time required for an electron beam to remain focused per pixel of the surface. Rapid dwell times and high-quality maintenance are essential in certain instances; however, this introduces noise, necessitating denoising. There is uncertainty in the noise information from SEM images, which makes it more difficult to apply traditional or well-known supervised denoising techniques. Additionally, self-or unsupervised methods require processing time and computation, which makes us choose supervised and blind methods, but still has problems with prior noise information. To address this issue, an Adaptive Blind Denoising framework that identifies and adapts to local noise patterns without prior knowledge is proposed. The framework comprises three modules: a Dataset Synthesizer that utilizes the power set theory to generate noise combinations, a denoising network that combines Shifted Windows Transformer, Convolutional Neural Network in U-shaped network (UNet), and a Noise Model Ranker (NMRanker) that adapts the optimal noise patterns. It was validated on SEM images using metrics such as the peak signal-to-noise ratio (PSNR), Structural Similarity Index Measure (SSIM), keypoint detection (necessary for further three-dimensional reconstruction), and visual assessment. The framework demonstrated improved denoising by accurately adapting to noise patterns from noisy input, with Speckle and Poisson noise combinations exhibiting optimal performance. This approach enhances SEM imaging denoising and can be validated in other fields.}, archive = {J_EAAI}, author = {Sheikh Shah Mohammad Motiur Rahman and Michel Salomon and Sounkalo Dembélé}, doi = {10.1016/j.engappai.2025.110871}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110871}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel adaptive noise model selection framework for blind denoising of scanning electron microscopy images}, volume = {154}, year = {2025}, }
• (2025). Three-dimensional ultrasonic self supervised segmentation. EAAI, 154, 110870. (www)

  @article{ , abstract = {This study introduces a novel self-supervised learning approach for volumetric segmentation of defect indications captured by phased array ultrasonic testing data from Carbon Fiber Reinforced Polymers. By employing this self-supervised method, defect segmentation is achieved automatically without the need for labelled training data or examples of defects. The approach has been tested using artificially induced defects, including back-drilled holes and Polytetrafluoroethylene inserts, to mimic different defect responses. Additionally, it has been evaluated on stepped geometries with varying thickness, demonstrating impressive generalization across various test scenarios. Minimal preprocessing requirements are needed, with no removal of geometric features or Time-Compensated Gain necessary for applying the methodology. The model's performance was evaluated for defect detection, in-plane and through-thickness localisation, as well as defect sizing. All defects were consistently detected with thresholding and different processing steps able to remove false positive indications for a 100 % detection accuracy. Defect sizing aligns with the industrial standard 6 dB amplitude drop method, with a Mean Absolute Error (MAE) of 1.41 mm (mm). In-plane and through-thickness localisation yielded comparable results, with MAEs of 0.37 and 0.26 mm, respectively. Visualisations are provided to illustrate how this approach can be utilised to generate digital twins of components.}, archive = {J_EAAI}, author = {Shaun McKnight and Vedran Tunukovic and Amine Hifi and S. Gareth Pierce and Ehsan Mohseni and Charles N. MacLeod and Tom O'Hare}, doi = {10.1016/j.engappai.2025.110870}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110870}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-dimensional ultrasonic self supervised segmentation}, volume = {154}, year = {2025}, }
• (2025). Parallel bayesian optimization using satisficing thompson sampling for fast charging design of lithium-ion batteries. EAAI, 154, 110868. (www)

  @article{ , abstract = {The rapid growth of the battery industry has increased the demand for fast charging strategies, which face two key optimization challenges: (i) the large space of the candidate charging strategy compared to the limited experiment budget, and (ii) the limited understanding of various degradation mechanisms. Existing optimization methods can be broadly classified into model-based and data-driven approaches. Model-based approaches are constrained by the accuracy of degradation models, while classical data-driven methods, such as Bayesian optimization (BO), may require many iterations to find the optimal solution. In this paper, we propose a data-driven approach named Blahut-Arimoto satisficing Thompson sampling-based parallel Bayesian optimization (BLASTS-PBO) for fast charging optimization. BLASTS-PBO shifts the target from the optimal solution to a satisficing solution that balances the informational requirements and optimality, which is particularly beneficial for time-sensitive optimization problems. We employ Gaussian processes as the surrogate model to capture the relationship between different charging strategies and utilize the BLASTS algorithm to compute the satisficing strategy. We also introduce the parallel technique to further improve time efficiency. Theoretical bounds on Bayesian cumulative regret confirm the method’s efficacy, and experiments using a porous electrode theory-based battery simulator demonstrate that BLASTS-PBO outperforms both sequential counterparts and parallel BO with traditional Thompson sampling in both synchronous and asynchronous settings. The results underscore BLASTS-PBO’s practical implication in accelerating the identification of effective charging strategies, offering a valuable solution for time-sensitive battery optimization problems.}, archive = {J_EAAI}, author = {Xiaobin Song and Benben Jiang}, doi = {10.1016/j.engappai.2025.110868}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110868}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Parallel bayesian optimization using satisficing thompson sampling for fast charging design of lithium-ion batteries}, volume = {154}, year = {2025}, }
• (2025). Multilevel parallel attention knowledge distillation for multimodal biometric recognition. EAAI, 154, 110865. (www)

  @article{ , abstract = {Multimodal Biometric Recognition (MBR) is a biometric technology designed to meet stringent security requirements in high-stakes scenarios. During the last decade, researchers have proposed various approaches to improve MBR, such as combining redundant biometric features and designing sophisticated recognition networks. However, the increasing complexity of these networks and the integration of additional features have substantially increased computational demands and prolonged training cycles, thereby hindering deployment on resource-constrained embedded systems. To address the aforementioned computational and deployment constraints, this paper proposes a Multilevel Parallel Attention Knowledge Distillation (MPAD) framework to compress MBR models. Unlike traditional knowledge distillation frameworks that only transfer unimodal features, our approach transfers attention activation maps from a complex teacher network to a lightweight student network through hierarchical layer alignment. Specifically, instead of guiding the student network’s learning of each modality with isolated teacher’s features, MPAD implements multilevel parallel attention fusion by combining spatial and channel attention mechanisms. This dual-attention design enables the model to simultaneously prioritize critical spatial regions and channel-wise dependencies, thereby enhancing the representational fidelity of distilled knowledge. Furthermore, a linear attention module generates batch-level activation matrices, which enforce similarity constraints between the feature extraction processes of the student and teacher networks. Extensive experiments on two real-world multimodal biometric datasets demonstrate that student models trained with our approach outperform those trained with other advanced knowledge distillation methods.}, archive = {J_EAAI}, author = {Kaikui Lu and Ruizhi Wu and Ergude Bao}, doi = {10.1016/j.engappai.2025.110865}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110865}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multilevel parallel attention knowledge distillation for multimodal biometric recognition}, volume = {154}, year = {2025}, }
• (2025). Deep learning-driven channel estimation for intelligent reflecting surfaces aided networks: A comprehensive survey. EAAI, 154, 110861. (www)

  @article{ , abstract = {Intelligent reflecting surfaces (IRS) technology has demonstrated considerable potential in enhancing wireless communication by improving signal quality and extending coverage. However, IRS-assisted systems face unique issues in channel estimation caused by their passive nature and the complexity of the channel environment. Deep learning-driven methods provide powerful tools to address complexities such as non-linearities and the high dimensionality inherent in these systems. This paper offers an extensive survey of existing channel estimation techniques in IRS-assisted systems, laying a foundation for future research. To achieve this, a comprehensive literature search was conducted across eight reputable databases and search engines, including IEEE Xplore, Google Scholar, and Scopus etc. After applying rigorous inclusion criteria, 57 key articles were identified as highly relevant, forming the basis of this review. The survey covers traditional methods, such as least squares (LS), minimum mean squared error (MMSE), and linear MMSE (LMMSE), and contrasts them with advanced approaches, including matrix decomposition, compressed sensing, and deep learning techniques. The survey then systematically categorizes the selected studies into three groups: discriminative (supervised learning), generative (unsupervised learning), and hybrid learning. This study reveals that convolutional neural networks (CNNs) are well-suited for resource-constrained or real-time applications, while transformers provide excellent adaptability and accuracy, albeit with higher computational demands. The survey concludes with insights into future research directions, emphasizing the need for improved estimation efficiency and robustness in next-generation wireless systems.}, archive = {J_EAAI}, author = {Jaya Singh and Kuldeep Singh and Dimpal Janu and Sandeep Kumar and Ghanshyam Singh}, doi = {10.1016/j.engappai.2025.110861}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110861}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning-driven channel estimation for intelligent reflecting surfaces aided networks: A comprehensive survey}, volume = {154}, year = {2025}, }
• (2025). Electric power optimization in solar trough plants with deep learning-based model predictive control. EAAI, 154, 110832. (www)

  @article{ , abstract = {On-site computational capabilities limit the optimization of electricity production in commercial plants. This work addresses the implementation of a virtual operator for electric power maximization in parabolic-trough collector plants by manipulating the flow rate circulating through the pipes with a neural network-based procedure. A model predictive control (MPC) strategy is proposed using nonlinear models to predict the system’s response. The effect of including diverse parts of the plant in the prediction model and using different objective functions (the model of the pump and the pipes and the difference between optimizing thermal and electric power) is analyzed. First, two control layers are implemented: one for obtaining the temperature setpoints and one for computing the flow rate. Since the nonlinear MPC cannot be computed in real-time for medium and large plants, an artificial neural network is trained to learn the optimal solution and lower the computational burden, reducing a 2-layer MPC strategy into only one neural controller that internally decides the operating point. The simulation results obtain a mean time reduction of 99.99995%, while the electricity production for the studied cases is only reduced by around 1%, making the controller implementable in real-time in actual plants.}, archive = {J_EAAI}, author = {Sara Ruiz-Moreno and Antonio J. Gallego and José Ramón D. Frejo and Eduardo F. Camacho}, doi = {10.1016/j.engappai.2025.110832}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110832}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Electric power optimization in solar trough plants with deep learning-based model predictive control}, volume = {154}, year = {2025}, }
• (2025). Big-data-driven vessel destination prediction for smart port management. EAAI, 154, 110829. (www)

  @article{ , abstract = {The accurate prediction of vessel destinations is crucial for enhancing maritime traffic efficiency, optimizing port management, and improving regional economic analysis. However, destination information in Automatic Identification System (AIS) data is often missing or inaccurate, which undermines the reliability of maritime analytic. Traditional vessel destination prediction methods primarily focus on measuring trajectory similarities, which results in high computational complexity. This study develops a deep learning approach to vessel destination prediction by transforming the problem into an image classification task. Rasterized images of historical destination ports and vessel trajectories are generated, incorporating AIS data within a fixed spatial context. A multi-scale residual convolutional network is constructed to extract relevant trajectory and port distribution features. To enhance the representation of trajectory endpoints, which are critical for predicting the destination port, a multi-attention mechanism is introduced. This mechanism increases the learning weight assigned to endpoint features, improving prediction accuracy. Finally, a classification network predicts the destination port based on the extracted features. The performance of the proposed method is evaluated using AIS data from the Denmark Strait. Experimental results demonstrate that the model outperforms existing methods, highlighting its potential for applications in smart port management and maritime traffic optimization.}, archive = {J_EAAI}, author = {Jin Chen and Qiang Zhang and Maohan Liang and Chang Peng and Chen Chen}, doi = {10.1016/j.engappai.2025.110829}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110829}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Big-data-driven vessel destination prediction for smart port management}, volume = {154}, year = {2025}, }
• (2025). Leveraging pretrained embeddings and self-attention based convolutional neural networks for early prediction of ventilator-associated pneumonia from clinical notes. EAAI, 154, 110827. (www)

  @article{ , abstract = {Patients in intensive care unit (ICU) encounter a high risk of developing ventilator-associated pneumonia (VAP); a serious healthcare-associated infection. Timely and early prediction of VAP can significantly improve patient outcomes by facilitating prompt treatment. The present research introduces a powerful deep learning framework that leverages pretrained biomedical word embeddings and a convolutional neural network (CNN) architecture, enriched with an attention mechanism, for VAP prediction from clinical notes within the Medical Information Mart for Intensive Care (MIMIC) III dataset. Conventional machine learning models with structured data are suffering in term of accuracy and generalizability. To address these challenges, our approach combines BioWordVec embeddings—pretrained embeddings designed to capture the complex semantics in biomedical text, with a CNN model that applies variable filter sizes for comprehensive feature extraction from unstructured clinical notes, thus capturing a broad range of n-gram features. Our model’s performance has been significantly enhanced by the addition of an attention mechanism, which enables it to concentrate on the most pertinent parts of the text. A hyperparameter tuning was performed to find the best set of hyperparameters for the model which leaded to significant performance in identifying VAP with a precision and recall rates of 99.26%, outperforming baseline models. To tackle our model’s interpretability, we visualized the contribution of features in the prediction process based on attention weights which enables us to identify the most relevant words for VAP prediction, this improves our approach’s transparency making it helpful for healthcare practitioners in medical decision-making. To assess the robustness and generalizability of our approach, we performed in inverse cross validation and closely mirrored its performance on the test set proving that the model generalizes well to unseen data. This research represents a significant advancement in the early diagnosis of VAP, extracting insightful information from unstructured clinical text and using it to generate extremely accurate binary classifications which improve patient care and lower the incidence of infections.}, archive = {J_EAAI}, author = {Zineb El M’hamdi and Mohamed Lazaar and Oussama Mahboub}, doi = {10.1016/j.engappai.2025.110827}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110827}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Leveraging pretrained embeddings and self-attention based convolutional neural networks for early prediction of ventilator-associated pneumonia from clinical notes}, volume = {154}, year = {2025}, }
• (2025). Anomaly detection for non-stationary rotating machinery based on signal transform and memory-guided multi-scale feature reconstruction. EAAI, 154, 110824. (www)

  @article{ , abstract = {Unexpected faults in rotating machinery pose a severe threat to safe operation. Condition monitoring can help recognize the abnormal state of machinery, thereby preventing downtime and improving reliability. However, due to the scarcity of abnormal data and the neglect of non-stationary conditions for rotating machinery, existing anomaly detection methods are difficult to distinguish the differences between normal and abnormal samples in real-world scenarios. To overcome these issues and improve condition monitoring performance, a novel anomaly detection method based on the signal transform and memory-guided multi-scale feature reconstruction is proposed. First, wavelet transform is utilized to reveal the complex features of signals in time-frequency diagrams and enhance the difference between normal and abnormal samples under non-stationary conditions. Second, the autoencoder-based anomaly detection network is designed to enlarge the discrimination of anomaly scores between normal and abnormal samples. Specifically, a compound scaling feature extraction network is innovatively embedded into the network to improve the diagram feature extraction ability. Furthermore, memory modules are incorporated at multiple levels to adjust the reconstruction ability of the autoencoder and restrain the uncontrollable generalizability of the decoder. Finally, experiment results demonstrate the effectiveness and superiority of the proposed method.}, archive = {J_EAAI}, author = {Ruonan Lu and Da Zheng and Qinmin Yang and Weiwei Cao and Chengyuan Zhu}, doi = {10.1016/j.engappai.2025.110824}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110824}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Anomaly detection for non-stationary rotating machinery based on signal transform and memory-guided multi-scale feature reconstruction}, volume = {154}, year = {2025}, }
• (2025). Interpreting frequency evolution in ventricular fibrillation using embeddings and deep learning methods. EAAI, 154, 110822. (www)

  @article{ , abstract = {Recently, the necessity for advanced tools to scrutinize ventricular fibrillation (VF) has been highlighted. Despite progress in the field, applying deep learning techniques and manifold interpretations in clinical settings remains underexplored. This study aims to evaluate the effectiveness of low-dimensional embeddings for distinguishing VF. We analyzed VF from three clinical conditions: patients during cardiopulmonary bypass, dogs administered with different drugs, and implantable cardioverter defibrillator devices with varying offset characteristics. We employed several algorithms, including uniform manifold approximation and projection embeddings, temporal convolutional networks, fully connected networks, and Kolmogorov–Arnold networks. Our experiments revealed that VF dynamics can be categorized based on frequency evolution, and the result can be interpreted based on clinical knowledge. However, each dataset has unique characteristics, leading to variations in the best-performing method. These differences may arise because some VF types are more easily identifiable. Our findings prove that longer signals differentiate VF types more clearly as the frequency evolution becomes clearer over extended periods. Across the same dataset, methods showed only slight differences in performance. Notably, for one dataset, two different drugs in dogs showed similar frequency patterns. For the rest of the datasets and methods, accuracy ranged between 0.68 and 0.86, precision ranged from 0.69 to 0.84, recall ranged from 0.68 to 0.84, and F1 scores ranged from 0.68 to 0.84. We conclude that low-dimensional embeddings are an effective method for characterizing VF types, and these methods can support ongoing research that aims to clarify the mechanisms of VF.}, archive = {J_EAAI}, author = {Dafne Lozano-Paredes and Juan José Sánchez-Muñoz and Luis Bote-Curiel and Francisco M. Melgarejo-Meseguer and Antonio Gil-Izquierdo and F. Javier Gimeno-Blanes and José Luis Rojo-Álvarez}, doi = {10.1016/j.engappai.2025.110822}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110822}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpreting frequency evolution in ventricular fibrillation using embeddings and deep learning methods}, volume = {154}, year = {2025}, }
• (2025). Polar edge distance loss in edge-aware plug-and-play scheme for semantic segmentation. EAAI, 154, 110810. (www)

  @article{ , abstract = {Semantic segmentation is a classic and fundamental computer vision problem dedicated to assigning each pixel with its corresponding class. Some recent methods introduce edge-based information for improving the segmentation performance. However these methods are specific and limited to certain network architectures, and they cannot be transferred to other models or tasks. Therefore, we propose a universal edge supervision method, EPS (Edge-aware Plug-and-play Scheme), which can be easily integrated into any semantic segmentation model. EPS extracts Edge of Ground Truth (Edge GT) with a predefined thickness from the training data and copies the decoder head as an auxiliary task for edge supervision. We introduce a novel Polar Edge Distance (PED) loss, which has lower computational complexity, is easier to optimize, and improves model performance more than other boundary-based losses. We train various recent open-source semantic segmentation models, such as Segmentation Transformer (SegFormer) and Image Cascade Network (ICNet), on the relatively simple public dataset Cityscapes and the challenging public dataset ADE20K to validate the effectiveness of the EPS and PED loss. Specifically, they can further enhance the mIoU (mean Intersection over Union) of segmentation based on SOTA (state-of-the-art) models. Moreover, we find that our method shows more significant improvements when applied to single-part closed objects with larger pixel values. EPS and PED loss can be applied to semantic and instance segmentation, with potential for future exploration in edge segmentation and weakly-supervised semantic segmentation.}, archive = {J_EAAI}, author = {Jianye Yi and Xiaopin Zhong and Weixiang Liu and Zongze Wu and Yuanlong Deng}, doi = {10.1016/j.engappai.2025.110810}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110810}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Polar edge distance loss in edge-aware plug-and-play scheme for semantic segmentation}, volume = {154}, year = {2025}, }
• (2025). A fine-grained detection algorithm for identity and actions of weapon-holding targets in public safety. EAAI, 154, 110800. (www)

  @article{ , abstract = {Fine-grained detection of threat targets in public security can provide more precise and detailed security information.Compared with general security systems, it can achieve faster and more accurate responses, which is of great significance for ensuring public security. In the face of the problems of low detection accuracy, inaccurate classification, and complex scene in fine-grained threat detection, we propose a fine-grained threat detection model based on the improved RealTime Detection Transformers (RT-DETR) algorithm.First, we built a high-performance backbone network called the Dilated Feature Aggregation Network to efficiently extract multiscale features. We used dilated convolutions for spatial feature aggregation and reparameterized them as large kernel convolutions to efficiently capture fine-grained target features. Secondly, a new feature encoder was developed to enable the model to better capture local and global features in complex real-world scenarios.Meanwhile, we propose a new feature fusion module DyContext Fusion Module, which improves the efficiency and fine-grained feature representation of feature fusion by dynamic sampling, context-guided features, and convolutional fusion. Finally, we utilized Adaptive Classification Loss to balance positive and negative samples, increased the inter-class diversity, and improved the model’s classification accuracy. We have built a public safety dataset and conducted comparative experiments with state-of-the-art algorithms that are publicly available to verify the effectiveness of the proposed improvement method. We also tested the performance of this method in terms of fine-grained and action detection on the Stanford 40 actions dataset, and the results showed that its performance was good. This proves the possibility of using this method for fine-grained detection of public security threat targets. The relevant code is released at https://github.com/ZIXUAN98/FG-DETR .}, archive = {J_EAAI}, author = {Zibo Yu and Weichao Wu and Jianzhong Wang and Yu You and Shaobo Bian and Endi Wang}, doi = {10.1016/j.engappai.2025.110800}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110800}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A fine-grained detection algorithm for identity and actions of weapon-holding targets in public safety}, volume = {154}, year = {2025}, }
• (2025). Multi-task lane-free driving strategy for connected and automated vehicles: A multi-agent deep reinforcement learning approach. EAAI, 154, 110797. (www)

  @article{ , abstract = {Deep reinforcement learning has shown promise in various engineering applications, including vehicular traffic control. The non-stationary nature of traffic, especially in the lane-free environment with more degrees of freedom in vehicle behaviors, poses challenges for decision-making since a wrong action might lead to a catastrophic failure. In this paper, we propose a novel driving strategy for Connected and Automated Vehicles (CAVs) based on a competitive Multi-Agent Deep Deterministic Policy Gradient approach. The developed multi-agent deep reinforcement learning algorithm creates a dynamic and non-stationary scenario, mirroring real-world traffic complexities and making trained agents more robust. The algorithm’s reward function is strategically and uniquely formulated to cover multiple vehicle control tasks, including maintaining desired speeds, overtaking, collision avoidance, and merging and diverging maneuvers. Moreover, additional considerations for both lateral and longitudinal passenger comfort and safety criteria are taken into account. We employed inter-vehicle forces, known as nudging and repulsive forces, to manage the maneuvers of CAVs in a lane-free traffic environment. The proposed driving algorithm is trained and evaluated on lane-free roads using the Simulation of Urban Mobility platform. Experimental results demonstrate the algorithm’s efficacy in achieving various defined objectives. These include laterally sorting vehicles based on their desired speeds with minimal deviation from those speeds. Additionally, the jerk and acceleration values remain within acceptable ranges. Overall, these promising results highlight the potential of the proposed approach to enhance safety and efficiency in autonomous driving within lane-free traffic environments.}, archive = {J_EAAI}, author = {Mehran Berahman and Athanasia Karalakou and Majid Rostami-Shahrbabaki and Klaus Bogenberger}, doi = {10.1016/j.engappai.2025.110797}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110797}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-task lane-free driving strategy for connected and automated vehicles: A multi-agent deep reinforcement learning approach}, volume = {154}, year = {2025}, }
• (2025). Long short-term memory and kolmogorov arnold network theorem for epileptic seizure prediction. EAAI, 154, 110757. (www)

  @article{ , abstract = {Brain activity during epileptic seizure is frequently examined using Electroencephalogram (EEG) signals. Analyzing electrical signals sent by scalp neurons is necessary to predict epileptic seizures. The intricacy and unpredictability of EEG data frequently restrict the efficacy of existing deep learning approaches, despite their potential for seizure prediction. In this study, we developed two deep learning techniques: the Kolmogorov Arnold Network (KAN) Theorem as a novel approach and Long Short-Term Memory (LSTM) networks. The temporal sequences in the EEG data are analyzed using the LSTM technique, where the data point order is crucial. Then, by estimating temporal patterns, KAN is used to improve the deep network capabilities, allowing for precise epileptic seizure prediction. The results indicate that our LSTM model performs well and achieves 95.37% accuracy over the KAN model, which got 80.15% accuracy as a lower performance to predict the epileptic seizure. the findings highlight that the LSTM model gives robust power for epileptic seizure prediction in the clinical domain. Despite being novel, the KAN method performed worse in this trial than the conventional LSTM model. For seizure prediction on the time-series EEG dataset, LSTM performed better than KAN, producing more dependable and robust findings. However, future model development may benefit from KAN’s special abilities to handle fluctuations and capture nonlinear correlations in EEG signals. Future research could benefit from a hybrid model that combines KAN and LSTM, where KAN learns from spline functions to further improve model performance and LSTM captures long-term dependencies.}, archive = {J_EAAI}, author = {Mohsin Hasan and Xufeng Zhao and Wenjuan Wu and Jiafei Dai and Xudong Gu and Asia Noreen}, doi = {10.1016/j.engappai.2025.110757}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110757}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Long short-term memory and kolmogorov arnold network theorem for epileptic seizure prediction}, volume = {154}, year = {2025}, }
• (2025). Development of an industry-applicable two-phase flow regime identification approach using flow-induced vibration acceleration signals. EAAI, 154, 110751. (www)

  @article{ , abstract = {Accurate classification of two-phase flow regimes is essential for optimizing industrial pipeline operations and ensuring system safety. Flow-induced vibration (FIV) signals present a promising option for developing flow regime classifiers due to the non-intrusiveness and accessibility of force/acceleration sensors. However, conventional FIV measurement and classification approaches often suffer from the database’s limited representativeness and necessitate complex feature engineering, thereby impeding their practical application in industrial systems. This study proposes a novel methodology that integrates a specialized experimental setup with an end-to-end one-dimensional convolutional neural network (1D-CNN) to effectively classify two-phase flow regimes using FIV acceleration signals. Firstly, a more representative FIV acceleration database was constructed by a tightly fixed 90-degree elbow section, thereby ensuring that induced vibrations are predominantly attributable to two-phase flow rather than structural resonance. Leveraging this database, a multi-channel 1D-CNN model was developed, capable of autonomously extracting temporal patterns from high-frequency FIV signals without the necessity for extensive feature engineering. The model architecture incorporates an adaptive average pooling layer and is optimized through a Bayesian framework to enhance both computational efficiency and classification performance. Comprehensive evaluations using metrics such as precision, recall, accuracy, and F1-score demonstrate that the proposed 1D-CNN model surpasses feature-based methods. An average classification accuracy of 89.9% was achieved across all flow regimes. Notably, the model demonstrated superior capability in distinguishing bubbly and slug flow regimes, achieving precisions of 97.2% and 91.7% and recalls of 94.6% and 89.8%, respectively. Furthermore, visualization analysis of the proposed model further reveals that convolutional kernels with varying degrees of smoothness and symmetry effectively capture dominant frequency patterns. This work not only advances the accuracy and efficiency of two-phase flow regime classification but also establishes a robust foundation for future research aimed at broadening the applicability of deep learning models in complex industrial settings involving two-phase flows.}, archive = {J_EAAI}, author = {Hanyu Wang and Shuichiro Miwa and Wen Zhou and Takashi Hibiki and Koji Okamoto}, doi = {10.1016/j.engappai.2025.110751}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110751}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Development of an industry-applicable two-phase flow regime identification approach using flow-induced vibration acceleration signals}, volume = {154}, year = {2025}, }
• (2025). A metro rail corrugation detection framework based on car body vibration signals and unsupervised learning. EAAI, 153, 110976. (www)

  @article{ , abstract = {Onboard detection of rail corrugation traditionally relies on axle box acceleration data and requires labeled data for supervised learning. To address data acquisition and labeling challenges, this study proposes an unsupervised learning framework using car body vertical vibration acceleration signals. First, the vibration response and transmission characteristics of train components were analyzed. Then, car body acceleration signals were transformed into time-frequency spectrograms using the Synchrosqueezed Wave Packet Transform (SSWPT). An unsupervised learning framework based on momentum contrastive learning was trained on these spectrograms, followed by fine-tuning for tasks such as corrugation wavelength classification and amplitude assessment. The results confirm that rail corrugation wavelength and amplitude can be effectively characterized using the SSWPT spectrograms. The baseline model achieved a Top-1 accuracy of 97.70 % and a Top-5 accuracy of 98.92 %. Incorporating the Convolutional Block Attention Module (CBAM) and multi-crop augmentation further improved Top-1 and Top-5 accuracy by 0.93 % and 0.49 %, respectively. For binary classification of corrugation presence, wavelength classification accuracy ranged from 95 % to 100 %, while amplitude assessment exceeded 95 %. Field verification results indicate that the proposed detection framework offers a robust data-driven method for onboard detection of rail corrugation.}, archive = {J_EAAI}, author = {Yang Wang and Hong Xiao and Mahantesh M. Nadakatti and Zhihai Zhang and Yihao Chi and Xiubo Liu}, doi = {10.1016/j.engappai.2025.110976}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110976}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A metro rail corrugation detection framework based on car body vibration signals and unsupervised learning}, volume = {153}, year = {2025}, }
• (2025). Explainable deep learning model with the internet of medical devices for early lung abnormality detection. EAAI, 153, 110961. (www)

  @article{ , abstract = {The decision-making process in healthcare monitoring systems has extensively used deep learning and the Internet of Things (IoT). Among the new uses in today's procedures is disease prediction. A difficult challenge in computer-aided diagnosis (CAD) is lung cancer prediction, which is addressed in this paper's solution using deep learning and IoT. IoT medical devices send disease-related data to the server, as lung cancer is a hazardous medical condition that must be detected faster. Following processing, a multi-layer Convolutional Neural Network (ML-CNN) model is used to classify the medical data into benign and malignant groups. Enhanced Particle swarm optimization (EPSO) is also used to enhance learning capacity (accuracy and loss). Medical data from the Internet of Medical Things (IoMT), including sensor data and Computed Tomography (CT) scan results, is used in this stage. The information from sensors and IoMT devices' picture data is collected for this purpose, and classification operations are then performed. The suggested method's accuracy, precision, sensitivity, specificity, F-score, and computation time are compared to well-known current approaches such as the Support Vector Machine (SVM), Probabilistic Neural Network (PNN), and Convolutional Neural Network (CNN). Linear Imaging and Self-Scanning Sensor (LISS) and Lung Image Database Consortium (LIDC) datasets were the two lung datasets used for this performance evaluation. Trial results indicate that the recommended strategy may aid in the timely and accurate identification of lung cancer in radiologists compared to other techniques. The efficacy of the suggested ML-CNN was examined through Python analysis. The results showed that the accuracy was superior to the number of instances, the precision was superior to the number of cases, and the sensitivity was superior to several instances, the F-score was superior to the number of cases, the error rate was inferior to the number of cases, and the computation time was inferior to the total number of instances computed for the proposed work, even when accounting for prior knowledge. Previous efforts are outperformed by this method, as shown by the suggested ML-CNN architecture.}, archive = {J_EAAI}, author = {Xin Hou and Nisreen Innab and Saad Alahmari and Meshal Shutaywi and Sara A. Althubiti and Ali Ahmadian}, doi = {10.1016/j.engappai.2025.110961}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110961}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable deep learning model with the internet of medical devices for early lung abnormality detection}, volume = {153}, year = {2025}, }
• (2025). ASDS-you only look once version 8: A real-time segmentation method for cross-scale prefabricated laminated slab components. EAAI, 153, 110958. (www)

  @article{ , abstract = {Prefabricated laminated slabs (PLS) are widely used globally due to their convenience. However, this convenience often comes with challenges in quality control. Although factories currently conduct quality inspections of PLS component arrangements, these inspections mainly rely on manual visual detection methods, which are highly inefficient. This paper proposes an improved You Only Look Once version 8 (YOLOv8) instance segmentation network for PLS inspection. To address the difficulties in detecting PLS components, we introduced multilevel auxiliary information in tandem with the main branch, designed an additional small-target feature fusion layer and segmentation header, and enhanced the original YOLOv8. These improvements allow for the extraction and segmentation of cross-scale information, reducing information gradient loss. However, this approach generates excessive cross-scale information, requiring a balance between the fusion weights of large-scale and small-scale information. To achieve this, we introduced a multilevel feature fusion module Semantic and Detail Infusion (SDI) and a dynamic upsampling module (Dysample). Experimental results show that the proposed method achieved a mean average precision (mAP 50 ) of 93.9 % and a detection speed of 108.7 Frames Per Second. Additionally, to support future research and applications, our method provides code that allows for direct derivation of the coordinates of each component class relative to the floor slab. Thus, the proposed detection method holds significant practical application value.}, archive = {J_EAAI}, author = {Lin Li and Qing Jiang and Guanting Ye and Xun Chong and Xinyu Zhu}, doi = {10.1016/j.engappai.2025.110958}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110958}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ASDS-you only look once version 8: A real-time segmentation method for cross-scale prefabricated laminated slab components}, volume = {153}, year = {2025}, }
• (2025). Physics-informed dual guidance method using physical envelope harmonic distribution and transfer learning for few-shot gear fault classification. EAAI, 153, 110956. (www)

  @article{ , abstract = {Recent years have seen artificial intelligence algorithms gain considerable popularity in gear fault classification, yet their performance remains hindered by the scarcity of labeled fault data, often leading to suboptimal classification results. Several state-of-the-art studies have demonstrated that incorporating physical information can improve classification accuracy. However, the differences between simulated and measured signals pose a significant challenge in enhancing the performance of physics-informed methods. In order to fill this gap, this paper introduces a novel physics-informed dual guidance (PI-DG) method using physical envelope harmonic distribution (PEHD) and transfer learning (TL) for few-shot gear fault classification. Within the proposed method, we introduce a new concept of PEHD, which is defined as the distribution feature of the shaft frequency and its harmonics in envelope spectrum. A physics-informed parameter optimization model (PI-POM) is developed to minimize the difference between the simulation and measured signals in terms of PEHD, enabling the accurate identification of detailed parameters within the dynamic model. Subsequently, a TL guidance framework is established for the fine-tuning and adaptation of a Long Short-Term Memory-aided Different Kolmogorov-Arnold network (LSTM-DKAN), with the aim of improving classification accuracy. Validation on a constructed back-to-back gear test rig with induced crack and spalling faults demonstrates the PI-DG method's effectiveness in reducing physics-simulation discrepancies, exhibiting superior classification performance especially in few-shot cases.}, archive = {J_EAAI}, author = {Kun Yue and Liming Wang and Xiaoxi Ding and Wennian Yu and Zaigang Chen and Wenbin Huang}, doi = {10.1016/j.engappai.2025.110956}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110956}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed dual guidance method using physical envelope harmonic distribution and transfer learning for few-shot gear fault classification}, volume = {153}, year = {2025}, }
• (2025). High-temperature bond strength evaluation of concrete overlays with industrial by-products: Experimental and analytical approaches using machine learning. EAAI, 153, 110954. (www)

  @article{ , abstract = {This paper presents a comprehensive study on the bond strength of concrete overlays incorporating industrial by-products such as fly ash and ground granulated blast furnace slag under high-temperature conditions. The research involved 540 concrete specimens exposed to temperatures of 25 °C, 100 °C, 300 °C, 500 °C, and 700 °C and cured for 14, 28, 60, and 90 days. Results demonstrated the beneficial roles of fly ash incorporation in overlays in improving the thermal stability and bond strength between concrete substrates and repair material overlays at high temperatures. Utilizing artificial intelligence techniques, a set of advanced machine learning models was created to accurately predict bond strength properties, specifically compressive, tensile, and shear strengths. The models evaluated included gradient boosting, extreme gradient boosting, light gradient boosting, and categorical boosting. Additionally, a multilayer perceptron (MLP) neural network was used as a baseline for comparison. The categorical boosting model demonstrated superior performance, achieving a root mean square error of 2.570, mean absolute error of 1.792, and R 2 of 0.989 for predicting compressive strength in the test dataset. It also achieved similarly high accuracy for tensile and shear strength, with R 2 values of 0.927 and 0.942, respectively. The SHapley Additive exPlanations analysis highlighted the significant impact of the predictive models on temperature, curing time, density, and porosity. Monte Carlo simulations further ensured robust model predictions and insights. The findings, accessible through a user-friendly web application ( https://high-temperature-opc-ibpc-cs-ts-ss-prediction-by-javid-toufigh.streamlit.app/ ), provide valuable insights into optimizing material compositions and enhancing the durability of repaired concrete structures in extreme environments. This study demonstrates the efficiency and reliability of artificial intelligence in bond strength evaluation of heat-damaged concretes.}, archive = {J_EAAI}, author = {Alireza Javid and Erfan Javid and Vahab Toufigh}, doi = {10.1016/j.engappai.2025.110954}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110954}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {High-temperature bond strength evaluation of concrete overlays with industrial by-products: Experimental and analytical approaches using machine learning}, volume = {153}, year = {2025}, }
• (2025). Horizontal-to-tilted conversion of solar radiation data using machine learning algorithms. EAAI, 153, 110951. (www)

  @article{ , abstract = {Solar radiation is the main input of system design algorithms in solar energy engineering. Solar radiation is usually measured on horizontal surfaces. However, in majority of solar energy applications such as photovoltaics, surfaces are either fixed at certain angles or continuously track the sun for maximizing energy input. Therefore, converting solar radiation data from horizontal to tilted surfaces is essential. Conventionally, conversion of solar radiation from horizontal to tilted is carried out using analytical methods. As with many other disciplines in science and technology, machine learning has recently been successfully applied also to solar radiation modelling to solve various problems such as in-advance forecasting of solar radiation. In the present article, solar radiation collected on horizontal surface is converted to tilted surface by machine learning algorithms and compared to solar radiation measured at a tilted surface. Eight different machine learning algorithms have been presently used for the conversion of solar radiation data. Accuracy of the models has been assessed based on a total of seven statistical metrics commonly used in literature. Overall, extra trees algorithm led to the best results as indicated by the statistical metrics used, for example, the mean absolute error of 7.3219 and coefficient of determination 0.9964. Based on the results presently obtained, it is demonstrated that machine learning led to an improved prediction when compared to the analytical models. The present research highlights the crucial significance of such advanced techniques, emphasizing their potential to drive a paradigm shift in solar energy engineering.}, archive = {J_EAAI}, author = {Ali Naci Celik and Bahadır Sarman and Kemal Polat}, doi = {10.1016/j.engappai.2025.110951}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110951}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Horizontal-to-tilted conversion of solar radiation data using machine learning algorithms}, volume = {153}, year = {2025}, }
• (2025). Granger-guided reduced dual attention long short-term memory for travel demand forecasting during coronavirus disease 2019. EAAI, 153, 110950. (www)

  @article{ , abstract = {The spread of Coronavirus disease 2019 (COVID-19) is closely related to residents’ travel. Quantities of studies have explored the spread of the epidemic impacted by the travel patterns. However, the researches of changes in residents' travel demand during the evolution of the epidemic are still insufficient. In particular, the lag effect of the epidemic on residents' travel demand has not been fully studied. This paper proposes a novel model named Granger-guided reduced dual attention long short-term memory (GRDA-LSTM) to predict changes in residents' travel demand under the impact of COVID-19. The contribution to artificial intelligence lies in leveraging Granger causality tests to enhance sensitivity to trend changes, incorporating a dual attention mechanism to improve forecasting performance, and utilizing channel reduction to boost efficiency. Experiments have proved that GRDA-LSTM is superior to existing Granger causality-integrated deep learning models. It effectively handles the abruptness and uncertainty of epidemic data, improves prediction accuracy, and meets rapid prediction requirements. The contribution in practical engineering applications is that the research findings not only provide more scientific guidance for traffic management practices in response to future comparable public health crises, but also broaden the research perspective on the impact of long-term public health events on transportation systems.}, archive = {J_EAAI}, author = {Yanan Zhang and Zan Chen and Bo Feng and Xueliang Sui and Shen Zhang}, doi = {10.1016/j.engappai.2025.110950}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110950}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Granger-guided reduced dual attention long short-term memory for travel demand forecasting during coronavirus disease 2019}, volume = {153}, year = {2025}, }
• (2025). A bidirectional bi-objective graph search model for sustainable urban railway alignment optimization. EAAI, 153, 110943. (www)

  @article{ , abstract = {Designing railway alignments in building-dense urban areas is a challenging task, requiring consideration of both costs and impacts on existing buildings and the environment. Achieving a viable solution necessitates the application of computer-aided techniques for three-dimensional (3D) global path searches while simultaneously optimizing multiple objectives. To tackle this challenge, this study proposes a bidirectional bi-objective graph search model. This model efficiently searches the 3D space to generate high-quality railway alignment solutions that simultaneously consider both comprehensive costs (including railway construction, ecological, and affected building costs) and carbon emissions (covering emissions from railways and buildings). It provides valuable reference solutions for designers, enhancing the design efficiency. The model includes two main innovations: (1) the ability to quickly search the entire 3D space using a graph-based strategy, generating multiple alignment solutions that meet design constraints in a single optimization process, and (2) the ability to accurately and efficiently account for the impact of railway alignments on existing buildings during optimization. Testing the model on a real-world urban case demonstrates its capability to generate multiple alternative railway alignments within minutes. The Pareto balanced solution achieves an 18.91 % reduction in comprehensive costs and a 13.46 % decrease in carbon emissions compared to manual design. The estimation error of affected building areas is approximately 2 %–4 % along the approximately 40 km alignment. Overall, the significance of this study lies in exploring the application of efficient graph search algorithms in the multi-objective optimization design of railway alignments in urban areas, advancing ongoing research in this field.}, archive = {J_EAAI}, author = {Tianlong Zhang and Yan Gao and Shuangting Xu and Ting Deng and Qing He and Paul Schonfeld and Yang Zou and Dong Liang and Ping Wang}, doi = {10.1016/j.engappai.2025.110943}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110943}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A bidirectional bi-objective graph search model for sustainable urban railway alignment optimization}, volume = {153}, year = {2025}, }
• (2025). A preliminary study of digital twin for nuclear reactor dynamics: A synergy of machine learning and model predictive control. EAAI, 153, 110940. (www)

  @article{ , abstract = {The operational stability of small modular reactor(SMR) in unmanned environments is a critical factor restricting the application of nuclear power. In these conditions, reactor faces the issue of sparse communication data, making it difficult to accurately grasp reactor's operational state. Digital twin for reactor is an effective solution to this problem. However, the complexity of reactor prevents straightforward synchronization of digital model's state with sensor measurements from physical system. Therefore, this paper proposes a method to achieve digital twin of reactor, comprising a data-driven state synchronization method and an online calibration method based on model predictive control (MPC). This approach combines hybrid dilated convolution(HDC) and long short-term memory (LSTM) to enable initial state synchronization. Online calibration utilizes point-kinetics model and core thermal-hydraulic model as predictive models.Controllable characteristic calibration parameters are selected as control variables, and a set of control variables that minimize model calculation bias is determined using the particle swarm optimization algorithm(PSO). These control variables are used to predict the transient changes in reactor. Simulation results indicate that the maximum mean absolute error (MAE) of state synchronization remains below 1.5. The error from online calibration also meets engineering requirements, and a lyapunov stability analysis further verifies the system's stability.Thus, the method has potential application value in other industrial processes with complex dynamic behaviors and can be extended to systems with similar characteristics.}, archive = {J_EAAI}, author = {Yunze Xue and Bowen Zhang and Ke Su and Yizhuo Li and Haixu Zhu and Honglin Pan}, doi = {10.1016/j.engappai.2025.110940}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110940}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A preliminary study of digital twin for nuclear reactor dynamics: A synergy of machine learning and model predictive control}, volume = {153}, year = {2025}, }
• (2025). A small object detection algorithm for mine environment. EAAI, 153, 110936. (www)

  @article{ , abstract = {The detection of protective equipment carried by underground mine operators is a crucial measure for preventing safety accidents and safeguarding personal life and property. However, current challenges include low object detection accuracy and difficulty detecting small objects, we propose a small object detection algorithm based on the improved You Only Look Once Version 8 (YOLOv8) for the mine environment. To minimize the semantic gap between features at different levels and enhance the feature fusion effect, the Asymptotic Feature Pyramid Network-Four (AFPN-F) has been designed to replace the Neck component of YOLOv8, enabling the detection model to better adapt to semantic information across varying levels. To enhance the model's sensitivity to small objects in the mine environment, a superficial feature output layer has been added to the model. This addition helps to prevent the loss of small-sized objects, which may contain limited feature information, during successive convolution operations. To address the significant differences in the scales of various objects in the mine, the More Focused Intersection over Union Loss (Focaler-IoU) is introduced as a loss function. This modification is intended to improve the handling of different types of regression samples, enhance training accuracy, and ensure that the model is more focused on small objects in the mine environment. The experimental results show that the proposed model outperforms other mainstream models. Compared to the baseline model YOLOv8, achieving an improvement of 3.7 percent in mean Average Precision (mAP), the number of parameters has been reduced by 30 percent, resulting in a model size of only 5 Megabytes. This study provides an effective solution for detecting small objects in underground mines.}, archive = {J_EAAI}, author = {Dong Liu and Xin Zhao and Weiqiang Fan}, doi = {10.1016/j.engappai.2025.110936}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110936}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A small object detection algorithm for mine environment}, volume = {153}, year = {2025}, }
• (2025). Artificial intelligence approaches in predicting the mechanical properties of natural fiber-reinforced concrete: A comprehensive review. EAAI, 153, 110933. (www)

  @article{ , abstract = {Implementing artificial intelligence (AI) techniques in predicting the mechanical properties of natural fiber-reinforced concrete (NFRC) has emerged as a transformative approach, offering significant advancements over traditional modeling methods. Construction materials science increasingly leverages advanced technologies to enhance composites like natural fiber-reinforced concrete (NFRC). However, the varied nature of natural fibers and their interactions within concrete matrices present significant challenges in accurately predicting the NFRC's mechanical properties. This comprehensive review highlights the innovative use of artificial intelligence (AI) techniques to address challenges in predicting material reliability. It explores the application of various AI methodologies, including machine learning (ML) techniques such as artificial neural networks (ANN) and support vector machines (SVM), along with pattern recognition (PR) and deep learning (DL). These approaches are utilized to tackle the challenges posed by the variability of natural fibers and their interactions within concrete matrices. These techniques have proven highly effective in accurately predicting the mechanical behavior of NFRC, marking significant advancements in the field. This review paper aims to summarize techniques for applying the AI methods mentioned in the NFRC. Firstly, we present a general introduction to AI and NFRC, highlighting the significance of AI in predicting the mechanical properties of NFRC. After that, a comparison between ML, PR, and DL in the field is discussed, and a review of recent applications of AI in the field is provided. Further, the advantages of employing such algorithmic methods are discussed in detail. Finally, future directions for employing ML, PR, and DL are presented, and their limitations are discussed.}, archive = {J_EAAI}, author = {Mohammed Mohammed and Jawad K. Oleiwi and Aeshah M. Mohammed and Azlin F. Osman and Tijjani Adam and Bashir O. Betar and Subash C.B. Gopinath}, doi = {10.1016/j.engappai.2025.110933}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110933}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence approaches in predicting the mechanical properties of natural fiber-reinforced concrete: A comprehensive review}, volume = {153}, year = {2025}, }
• (2025). Parallel multi-layer sensor fusion for pipe leak detection using multi-sensors and machine learning. EAAI, 153, 110923. (www)

  @article{ , abstract = {Effective pipe leak detection is critical for maintaining the structural integrity and efficiency of water distribution systems and preventing damage such as sinkholes. Traditional leak detection methods often rely on single sensors, overlooking the advantages of multi-sensor configurations that capture diverse leak-related phenomena. To address this limitation, the study proposes an innovative machine learning-based sensor fusion approach called Parallel Multi-Layer Sensor Fusion (PMLSF), which leverages Convolutional Neural Networks (CNN) and Few-Shot Learning (FSL) to enhance leak detection. PMLSF integrates data from multiple sensors, including hydrophone, acoustic emission, and vibration sensors. The comparative analysis demonstrates that the PMLSF with multi-sensor systems substantially outperforms the CNN-based FSL (CNN-FSL) approach with single-sensor systems, achieving a leak detection accuracy of 97.1 % and leak location classification accuracy between 95.5 % and 97.4 %. Additionally, the study investigates the use of the acoustic emission sensor combined with CNN-FSL for early detection of material failure in pipes, demonstrated by a Pencil Test that achieved 92.3 % accuracy in detecting pencil breakage on the pipe. These results indicate that combination of CNN-FSL for the acoustic emission sensor and PMLSF offers a comprehensive solution for detecting and localizing existing leaks while predicting potential failures, thus laying a robust foundation for the development of reliable and efficient water distribution monitoring systems.}, archive = {J_EAAI}, author = {Nicholas Satterlee and Xiaowei Zuo and Chang-Whan Lee and Choon-Wook Park and John S. Kang}, doi = {10.1016/j.engappai.2025.110923}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110923}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Parallel multi-layer sensor fusion for pipe leak detection using multi-sensors and machine learning}, volume = {153}, year = {2025}, }
• (2025). An integrated deep learning method and optimization algorithm framework for energy-saving steel billet heating. EAAI, 153, 110916. (www)

  @article{ , abstract = {Accurate prediction and optimization of steel billet temperature are crucial for ensuring billet quality and reducing the energy consumption of reheating furnaces in the steel industry. However, the complex heat transfer processes and varying operating conditions within the furnace make real-time application of traditional mechanism-based models computationally prohibitive. To address this challenge, a data-driven steel billet temperature prediction model is proposed based on Bidirectional Long Short-Term Memory networks. To enhance its performance, the model integrates an early stopping strategy and Tree-structured Parzen Estimator Bayesian optimization for hyperparameter tuning, which improves prediction accuracy while preventing overfitting. The proposed model achieves rapid inference within milliseconds while maintaining high accuracy, significantly reducing computational complexity and making real-time applications feasible. Furthermore, to maximize energy savings while ensuring heating quality, the prediction model is integrated with Adaptive Particle Swarm Optimization algorithm to establish a furnace temperature optimization framework. Experimental results demonstrated that the prediction model achieved a high determination coefficient (R 2 ) of 0.9756 and exhibited millisecond-level inference speed, with an average improvement of 10.24 percent in R 2 compared to the three benchmark models. The optimized furnace temperature strategies in Case 1 reduce fuel consumption by 2480.3 cubic meters per hour, achieving a 3.54 percent reduction in energy usage. Based on the current fuel prices in the steel plant investigated in this study, these savings correspond to a cost reduction of 97.9 dollars per hour in Case 1.}, archive = {J_EAAI}, author = {Wenchao Ji and Guojun Li and Chunguang Zhao and Zhi Yi and Linyang Wei and Shuangcheng Sun}, doi = {10.1016/j.engappai.2025.110916}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110916}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated deep learning method and optimization algorithm framework for energy-saving steel billet heating}, volume = {153}, year = {2025}, }
• (2025). Model reconstruction and update method for dynamic prediction of loader loading resistance using deep incremental learning. EAAI, 153, 110910. (www)

  @article{ , abstract = {Addressing the challenges of high computational demand and slow speed in Discrete Element Method-Multibody Dynamics (DEM-MBD) co-simulation models, this paper proposes a model reconstruction and update method using deep incremental learning. First, a digital model based on DEM-MBD co-simulation is developed through calibration and verification of experimental data from loader loading operations. Subsequently, feature parameters for DEM-MBD model reconstruction are obtained via analysis of loading resistance-related parameters. Second, an improved Transformer-based deep learning model is trained using simulation data, incorporating knowledge distillation (KD) and domain adaptation transfer (DA) to construct the deep incremental learning model Transformer-KD-DA, aimed at achieving reconstruction and update of the DEM-MBD co-simulation model. Finally, the model performance is verified using simulation data from various loading trajectories. The results demonstrate that the Transformer-KD-DA proposed in this study effectively reconstructs and updates the DEM-MBD co-simulation model, enabling dynamic prediction of loader loading resistance under different loading trajectories.}, archive = {J_EAAI}, author = {Binyun Wu and Liang Hou and Shaojie Wang and Xiangjian Bu and Cheng Xiang}, doi = {10.1016/j.engappai.2025.110910}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110910}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Model reconstruction and update method for dynamic prediction of loader loading resistance using deep incremental learning}, volume = {153}, year = {2025}, }
• (2025). Ensemble insights: Unlocking the recombination losses in perovskite solar cells using stacked classifier. EAAI, 153, 110909. (www)

  @article{ , abstract = {Recombination losses in Perovskite solar cells (PSCs) have been triggered/witnessed to improve power conversion efficiency (PCE) through experimental and machine learning (ML) techniques in recent years. ML approaches possess significant potential for enhancing the PSCs efficiency by addressing the recombination losses at band to band (BB), Grain Boundaries (GB), and interface compared to the experimental setup. To maximize this potential, a versatile and optimal ML model is needed, to unveil these recombination losses within PSCs effectively. Herein, diverse ML models implemented, underwent through the best features, and seven-fold cross-validation to ensure generalizability and robustness to address the recombination losses. Furthermore, an optimal hyperparameter was determined through Optuna. Additionally, for the model interpretation, Shapley additive explanations (SHAP) is employed. Among them, the stacked classifier emerged as an exceptionally efficient classifier, achieving a remarkable accuracy of 0.94. In addition, the best model (stacked classifier) is validated with experimental data, revealing a strong agreement between the experimental outcomes and the prediction of the ML model showing an optimal route to address the recombination losses in PSCs. Finally, the web server is developed based on a stacked classifier to provide a user-friendly interface, and easy navigation, enabling the practical application for addressing the recombination losses in PSCs. Thus, the proposed approach improved accuracy by 11 % and correctly identified recombination losses with a 15.48 % improvement at GB and 15.79 % at interface compared to previously reported methodologies.}, archive = {J_EAAI}, author = {Basir Akbar and Kil To Chong and Hilal Tayara}, doi = {10.1016/j.engappai.2025.110909}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110909}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ensemble insights: Unlocking the recombination losses in perovskite solar cells using stacked classifier}, volume = {153}, year = {2025}, }
• (2025). Touching the limit of rolling multilayer perceptron for efficient two-dimensional medical image segmentation. EAAI, 153, 110899. (www)

  @article{ , abstract = {Significant progress has been made in medical image segmentation using deep learning techniques, with the U-shaped architecture being a classic choice. However, effectively capturing and integrating both local features and remote dependencies remains a key challenge for improving deep learning-based segmentation methods. In this paper, we propose a flexible Rolling Multilayer Perceptron (Rolling-MLP) module to address this issue. Building upon this concept, we present the Rolling-Unet network, which combines the strengths of Multilayer Perceptrons (MLPs) with Convolutional Neural Networks (CNNs) to efficiently extract and fuse local features and remote dependencies. Furthermore, to explore the potential of Rolling-MLP for two-dimensional medical image segmentation, we propose Rolling-MLP configurations with distinct receptive field shapes (linear and area-shaped) and summarize the influence of Rolling-MLP's key parameters on the shape of receptive fields. We conducted extensive experiments on four datasets, surpassing a variety of state-of-the-art methods in accuracy. Moreover, Rolling-MLP is far ahead in Central Processing Unit (CPU) inference speed, indicating its potential in medical cyber-physical systems engineering applications. This paper demonstrates the strong comprehensive ability of Rolling-MLP in two-dimensional medical image segmentation tasks, providing a novel approach for constructing medical image segmentation networks, alternative to CNNs and Transformers.}, archive = {J_EAAI}, author = {Yutong Liu and Haijiang Zhu and Ning An and Li Xu}, doi = {10.1016/j.engappai.2025.110899}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110899}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Touching the limit of rolling multilayer perceptron for efficient two-dimensional medical image segmentation}, volume = {153}, year = {2025}, }
• (2025). Scalable and transferable graph neural networks for predicting temperature evolution in laser powder bed fusion. EAAI, 153, 110898. (www)

  @article{ , abstract = {Predicting temperature distributions in laser powder bed fusion (L-PBF) processes is essential for mitigating thermal distortions and ensuring the structural integrity of manufactured parts. Traditional finite element analysis (FEA) methods, while accurate, are computationally intensive and struggle to scale to larger domains. To address these limitations, this study proposes novel predictive models based on Graph Neural Networks (GNNs) to simulate thermal dynamics in L-PBF processes. The models leverage high-fidelity FEA data from small-scale domains to generalize effectively to larger domains with minimal retraining. For single-laser setups, the GNN achieves a Mean Absolute Percentage Error (MAPE) of 3.77 %, while significantly reducing computational costs. For instance, a thermomechanical simulation for a 2 mm square domain typically takes about 4 h, whereas the single-laser model predicts thermal distributions almost instantly. When calibrated for larger domains, the models significantly enhance predictive performance, showing notable improvements for square domains of 3 mm and 4 mm. Additionally, the models show a decreasing trend in Root Mean Square Error when tuned to larger domains, suggesting the potential for becoming geometry-agnostic. The interaction of multiple lasers complicates heat transfer, necessitating larger model architectures and advanced feature engineering. Using hyperparameters from Gaussian process-based Bayesian optimization, the best multi-laser surrogate model demonstrates a 46.4 % improvement in MAPE over the baseline model. By providing scalable and efficient predictive tools alongside FEA, this work paves the way for thermal modeling in L-PBF.}, archive = {J_EAAI}, author = {Riddhiman Raut and Amit Kumar Ball and Amrita Basak}, doi = {10.1016/j.engappai.2025.110898}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110898}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Scalable and transferable graph neural networks for predicting temperature evolution in laser powder bed fusion}, volume = {153}, year = {2025}, }
• (2025). An intelligent fault diagnosis for rotating machine under strong noise based on cross-attention-driven spatial-temporal feature fusion and duplexing time sequence convolution optimization. EAAI, 153, 110897. (www)

  @article{ , abstract = {In complex working conditions, noise is still the major problem affecting the accuracy of fault identification with deep learning. To solve this problem, an intelligent fault identification method for rotating machine is proposed, which is based on improved bidirectional temporal convolutional network (Bi-TCN) and cross-attention mechanism with spatiotemporal features. Firstly, a new signal pretreatment strategy is proposed. Therefore, noise is reduced and fault information highlighted through the cross-correlation function of difference signal and original signal. Secondly, to construct a matrix containing dual-domain information, cross-correlation function is subjected to feature mode decomposition (FMD) and fast Fourier transform. Thirdly, to address the difficulty of classic Bi-TCN in converging rapidly under strong noise, the convolution kernel of the first temporal convolutional neural network in Bi-TCN is improved into a large convolution kernel. Fourthly, the constructed dual-domain matrix is respectively input into bidirectional gated recurrent unit (Bi-GRU) and the improved Bi-TCN. The temporal and spatial features are extracted. Meanwhile, dot product operation between different vectors is improved into one of the same vector to capture the long-distance dependency within the same vector more deeply. Finally, extracted temporal and spatial features are inputted into the cross-attention mechanism and feature information is combined to identify faults. Results indicate that in the identification of rotor-stator rub-impact faults, the proposed method can achieve an identification accuracy of 99.9 %; under two different bearing datasets, the accuracy reached 99.66 % and 99.84 %; under strong noise, the identification accuracy of 3 datasets can still reach to 89.92 %, 86.03 %, and 97.95 % respectively.}, archive = {J_EAAI}, author = {Mingyue Yu and Yongpeng Li and Guanglei Meng and Yunbo Wang}, doi = {10.1016/j.engappai.2025.110897}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110897}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An intelligent fault diagnosis for rotating machine under strong noise based on cross-attention-driven spatial-temporal feature fusion and duplexing time sequence convolution optimization}, volume = {153}, year = {2025}, }
• (2025). Leveraging machine learning for analyzing structural behavior of a composite wind turbine blade. EAAI, 153, 110895. (www)

  @article{ , abstract = {The main objective of this paper is to develop models to quickly investigate the mechanical behaviors of an industrial case of a composite wind turbine blade based on probable variations in the mechanical properties of the constituent materials purchased from different suppliers. The wind turbine blade behaviors, including flexural natural frequencies and stiffnesses in both longitudinal and transverse directions, are predicted using four machine learning models. To construct each model, the blade's corresponding behavior is initially characterized using an appropriate low-fidelity model based on possible variations in mechanical properties. Subsequently, the required databases for training and evaluating each improved high-order machine learning model are created. After well-training the models and evaluating their performances through various error criteria, the sensitivity improved model is examined. The information presented in this paper substantially resolves one of the significant challenges in large-scale wind turbine industries.}, archive = {J_EAAI}, author = {Roham Rafiee and Ali Shahcheraghi and Amir Shayestehmanesh}, doi = {10.1016/j.engappai.2025.110895}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110895}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Leveraging machine learning for analyzing structural behavior of a composite wind turbine blade}, volume = {153}, year = {2025}, }
• (2025). Automated skills assessment in open surgery: A scoping review. EAAI, 153, 110893. (www)

  @article{ , abstract = {Surgical skills proficiency lowers the incidence of adverse clinical outcomes during surgeries. Artificial intelligence (AI) has been applied for surgical skills assessment, especially in the field of minimally invasive surgeries (MIS). This paves the way for integrating AI for skills assessment in open surgeries as well. An overview of its applications can inform the scientific community and facilitate further developments. In this scoping review, we present the open surgeries and clinical settings where AI-based skill assessment has been applied, the kind of surgical data acquired for the AI-based algorithms, and the types of AI-based models used for automated skills assessment. A total of 40 articles were identified and included. Majority of the articles focused on macrosurgical suturing (45 %, n = 18). Most of the studies acquired data by capturing surgeon's hands (50 %, n = 20). About 35 % utilized deep learning algorithms, specifically convolutional neural networks (CNN) ( n = 14). The assessment input for the automation algorithms were predominantly hand movement. Around 37.5 % ( n = 15) of the studies assessed algorithm performance using classification accuracy. In the review, we compare conventional methods such as statistical modeling and custom algorithms with the emerging AI-based approaches. We also explore the utilization of object detection and temporal information for surgical skills assessment. We highlight the progress in automated skills assessment during open surgery with advancements in sensor technology, and AI algorithms with high prediction accuracies. Further developments in data acquisition and processing methods are essential to facilitate clinical implementation of such technologies.}, archive = {J_EAAI}, author = {Hawa Hamza and Dehlela Shabir and Omar Aboumarzouk and Abdulla Al-Ansari and Khaled Shaban and Nikhil V. Navkar}, doi = {10.1016/j.engappai.2025.110893}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110893}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated skills assessment in open surgery: A scoping review}, volume = {153}, year = {2025}, }
• (2025). A cusp catastrophe theory based multistage group decision-making method: Addressing scenario mutations in emergencies and heterogeneous risk perceptions of experts. EAAI, 153, 110892. (www)

  @article{ , abstract = {Emergency response is a risk decision-making problem in essence, with decision makers (DMs) typically relying on their perceptions of risks to guide their decisions. However, few researchers in the field of emergency decision-making (EDM) have considered differences in risk perceptions (RPs) among DMs, nor have they given much attention to the mutation nature of risk situations as emergencies evolve. In light of this, a cusp catastrophe theory based multistage group EDM method is proposed for addressing heterogeneous risk perceptions (HRPs) of DMs and scenario mutations in emergencies. First, a quantitative model of HRPs, considering risk perception probability influenced by noise disturbance and varying levels of risk perception intensity of DMs, is proposed to calculate the RPs values of DMs for various alternatives. Second, an integrated model that incorporates risk factors coupling and DMs' different perception abilities is developed, allowing for the determination of risk factors' weights and experts' weights. These weights are then used to aggregate HRPs from multiple DMs under multiple risk factors. Next, a cusp catastrophe (CC) model aims to analyze the mutation characteristics and calculate the risk potential values for each stage, thus the stage weights are used to calculate the comprehensive RPs values of DMs on alternatives. Finally, a case study and comparative analysis are carried out to demonstrate the proposed method's effectiveness.}, archive = {J_EAAI}, author = {Huifang Nie and Jianjun Zhu and Huagang Tong and Zhiying Wang}, doi = {10.1016/j.engappai.2025.110892}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110892}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cusp catastrophe theory based multistage group decision-making method: Addressing scenario mutations in emergencies and heterogeneous risk perceptions of experts}, volume = {153}, year = {2025}, }
• (2025). Hydropower projects investment estimation artificial intelligence model by similarity measure-based few-shot machine learning. EAAI, 153, 110891. (www)

  @article{ , abstract = {Hydropower is an important and clean renewable energy source. The construction of hydropower projects involves huge investments, and it's important to estimate the investment accurately in the design and planning stage. Machine learning can construct investment estimation models based on a large number of samples. However, in the hydropower industry, given that Engineering Procurement Construction (hereafter EPC) is an emerging investment and construction model, relatively few samples are available for investment estimation of EPC hydropower projects. This study aims to propose a few-shot machine learning method for hydropower project investment estimation. First, a comprehensive similarity index is proposed to select similar projects; after that, a hydropower project investment estimation artificial intelligence model is constructed by integrating an improved support vector machine and swarm intelligence optimization algorithm. In this study, the investment data of 36 EPC hydropower projects are collected. Experiments show that the proposed method can construct a hydropower project investment model using a small number of samples, and the model prediction accuracy is 97.1 %, which is better than the traditional machine learning method. The proposed method helps to provide support for the comparison of investment and construction programs for EPC hydropower projects, thus reducing the investment cost.}, archive = {J_EAAI}, author = {Mengnan Shi and Xinyu Qu and Hongtao Li and Qiang Yao and Jun Zeng}, doi = {10.1016/j.engappai.2025.110891}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110891}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hydropower projects investment estimation artificial intelligence model by similarity measure-based few-shot machine learning}, volume = {153}, year = {2025}, }
• (2025). Proactive approach to enhancing safety management using deep learning classifiers for construction safety documentation. EAAI, 153, 110889. (www)

  @article{ , abstract = {Construction accidents remain a major global concern, highlighting the need for proactive safety measures during the design phase. Design for Safety (DfS) reports, which aim to identify and mitigate hazards during the design stage, play a crucial role in preventing accidents. However, their manual classification in South Korea is time-consuming and error-prone, reducing their effectiveness. Thus, this study presents a deep learning-based classification framework to automatically classify DfS reports based on accident types, facility types, and work types. Three deep learning architectures, Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM model, were developed and trained on a comprehensive dataset of 1513 DfS reports collected from government databases and construction companies' internal records. An extensive model evaluation was conducted using key performance metrics, including accuracy, precision, recall, F1 score, and Area Under the Receiver Operating Characteristic Curve (AUC ROC). The results demonstrated that the CNN model achieved the highest performance across all tasks, with an accuracy of 71 % for accident type classification, 46 % for facility type classification, and 64 % for work type classification. Additionally, CNN outperformed LSTM and CNN-LSTM in terms of precision, recall, and F1 score across all categories, indicating its superior capability in extracting spatially correlated textual patterns. Beyond model performance, this study contributes to construction safety by introducing an automated, data-driven approach to DfS report analysis, which facilitates proactive hazard identification, enhances risk prioritization, and optimizes resource allocation during the design phase.}, archive = {J_EAAI}, author = {Louis Kumi and Jaewook Jeong and Jaemin Jeong}, doi = {10.1016/j.engappai.2025.110889}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110889}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Proactive approach to enhancing safety management using deep learning classifiers for construction safety documentation}, volume = {153}, year = {2025}, }
• (2025). An enhanced T-spherical fuzzy generalized multi-criteria decisioning model for evaluating the effectiveness of digital transformation in the sustainability of agri-food systems. EAAI, 153, 110887. (www)

  @article{ , abstract = {The rapid advancement and widespread integration of artificial intelligence (AI) have provided substantial technical support for digital transformation (DT), positioning it as a key enabler of sustainable development across various industries. However, research on the role of DT in achieving sustainability within the agri-food sector remains limited. To address this gap, this study proposes a hybrid decision-making approach to assess the impact of DT on sustainability in the agri-food sector, particularly under conditions of uncertainty. The proposed framework integrates a modified generalized TODIM (an acronym in Portuguese for Interactive and Multicriteria Decision Making), a T-spherical fuzzy weighted Heronian mean operator, and Cronbach's coefficient to enhance decision-making reliability. The model evaluates DT's effectiveness in fostering sustainable agri-food systems by aggregating expert judgments through the Relative Closeness Coefficient (RCC) method, ensuring comprehensive factor interaction analysis. Additionally, a weighted Minkowski distance Heronian aggregation operator is introduced to prioritize organizational performance in sustainability efforts. To validate the proposed approach, an empirical case study illustrates its application in evaluating DT-driven sustainability within the agri-food sector. The findings highlight traceability and visibility as critical factors enhancing production efficiency. Sensitivity and comparative analyses further confirm the robustness and reliability of the proposed decision-support framework. This study contributes to the literature by offering a novel methodological approach for assessing DT's role in sustainable agri-food systems, providing valuable insights for both academia and industry stakeholders.}, archive = {J_EAAI}, author = {Weizhong Wang and Zhengyan Yang and Yushuo Cao and Muhammet Deveci and Huai-Wei Lo and Dursun Delen}, doi = {10.1016/j.engappai.2025.110887}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110887}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An enhanced T-spherical fuzzy generalized multi-criteria decisioning model for evaluating the effectiveness of digital transformation in the sustainability of agri-food systems}, volume = {153}, year = {2025}, }
• (2025). Multi-condition pipeline leak diagnosis based on acoustic image fusion and whale-optimized evolutionary convolutional neural network. EAAI, 153, 110886. (www)

  @article{ , abstract = {This study proposes a multi-condition pipeline leak diagnosis method that utilizes pixel-level image fusion and an evolutionary convolutional neural network enhanced by the whale optimization algorithm. Firstly, the traditional one-dimensional leakage time series are transformed into two-dimensional acoustic images, which contain richer information. The signal characteristics, as well as the advantages and disadvantages of different image conversion methods, are compared and analyzed. Secondly, given that a single acoustic image conversion method struggles to achieve effective diagnostics across various working conditions, this study employs two pixel-level image fusion techniques: the principal component transformation method (PCA) and the Intensity-Hue-Saturation (IHS) method, to conduct a comparative study. The results indicate that the diagnostic performance of the fusion methods surpasses that of the unfused acoustic image conversion method, with the IHS method proving to be more suitable for pipeline leakage detection scenarios. In addition, to enhance the accuracy and robustness of pipeline leak detection under varying conditions, a Whale Optimized Evolutionary Convolutional Neural Network model (WOA-ECNN) has been developed for leak diagnosis research. The model was tested under five working conditions (two types of no leakage, 1 mm leakage, 3 mm leakage, and 5 mm leakage). The results demonstrate that the proposed pixel-level fusion method can accurately identify and differentiate between leakage and non-leakage states. Furthermore, multiple performance indicators show superior results compared to the single acoustic image conversion method. In addition, the diagnostic accuracy of the ECNN model, enhanced by whale optimization (WOA-ECNN), is significantly higher than that of the traditional CNN model.}, archive = {J_EAAI}, author = {Yuan Yuan and Xiwang Cui and Xiaojuan Han and Yan Gao and Fangcheng Lu and Xianhong Liu}, doi = {10.1016/j.engappai.2025.110886}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110886}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-condition pipeline leak diagnosis based on acoustic image fusion and whale-optimized evolutionary convolutional neural network}, volume = {153}, year = {2025}, }
• (2025). Classification of non-standard mechanical parts with graph convolutional networks. EAAI, 153, 110879. (www)

  @article{ , abstract = {The structure and functionality of non-standard mechanical parts (NSMPs) are intricate with a gradual increase in proportion in the complex customized products. The classification for standard mechanical parts usually struggles to effectively encompass the diversity and complexity of NSMPs. The existing standard classification system seems to be insufficient and lacks flexibility. The research classifies NSMPs into four categories according to processing methods, aiming to efficiently allocate batches of NSMPs to the workshops with suitable processing capabilities. First, the attributed graph was constructed from boundary representations, in which dimensionless geometric parameters with geometric invariance serve as associated features. Secondly, we trained the improved graph convolutional network, named NSMP-Net, which uses nonlinear projections for representing implicit relationships among surfaces and curves. Specific attention mechanisms are used to readout node hidden features, reducing redundant features' influence on graph-level embeddings. Finally, the experimental results demonstrate that NSMP-Net achieved 91.3 % accuracy in classification tasks, outperforming other models with higher consistency after spatial and dimensional transformation, demonstrating great potential in classification.}, archive = {J_EAAI}, author = {Zirui Li and Xiaomeng Tong and Peng Shi and Maolin Cai and Fei Han}, doi = {10.1016/j.engappai.2025.110879}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110879}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Classification of non-standard mechanical parts with graph convolutional networks}, volume = {153}, year = {2025}, }
• (2025). A data-driven decision support system for selecting managers based on their performance evaluations over time. EAAI, 153, 110874. (www)

  @article{ , abstract = {One of the pillars of organizational success is having capable and efficient managers; therefore, selecting competent people for managerial positions can play a significant role in the future of the organization. This paper proposes a method for selecting managers within the organization using data-driven methods based on their performance-evaluation information over time. To do this, optimal function and suitable algorithm for aggregating the performance evaluation data are determined, and the individual's performance in each vacant management position is considered based on mean absolute error (MAE), mean squared error (MSE), and root mean square error (RMSE) measures. The proposed approach has been implemented in the Yazd University as a case study, which showed reliable results. Three managerial positions have been considered in this case study, i.e., 1) budget and planning manager, 2) financial and administrative affairs manager, and 3) student affairs manager. Among several rival aggregations and predicting algorithms, the data averaging aggregation method with gradient boosting machine algorithm showed the best results. Finally, the selected individuals have been assigned to each management position. The experts validated the output of the proposed model, and their satisfaction scores for the three managerial positions were 100 %, 80 %, and 80 %, respectively. The achievements of this research include identifying the best algorithm for predicting the performance of personnel in managerial positions, identifying proper people in the organization, and using the performance evaluation information to select capable managers.}, archive = {J_EAAI}, author = {Sepideh Saboori-Taft and Majid Shakhsi-Niaei and Mohammad-Sadegh Zamani}, doi = {10.1016/j.engappai.2025.110874}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110874}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A data-driven decision support system for selecting managers based on their performance evaluations over time}, volume = {153}, year = {2025}, }
• (2025). Use deep transfer learning for efficient time-series updating of subsurface flow surrogate model. EAAI, 153, 110873. (www)

  @article{ , abstract = {In subsurface flow, history matching is a complex high-dimensional sampling problem demanding numerous computationally intensive numerical simulations, which result in high costs. To mitigate this, various deep learning-based surrogate-assisted techniques are employed to approximate the original numerical simulation processes for more efficient computation. Despite the widespread application of deep learning-based surrogate models, significant limitations remain, particularly the need for the frequent reconstruction of surrogate models when new observed data become available. This process requires additional high-fidelity simulation data, further increasing computational burden of history matching. To tackle this problem, we introduce a transfer learning-based framework within an image-to-sequence surrogate model to enable efficient updates as new data become available. This surrogate model consists of a geological feature extraction module and a sequence data regression module. To facilitate rapid learning of new tasks while retaining previously learned knowledge, we employ a decoupled learning strategy where the backbone structure remains fixed, and only the sequence data regression module is fine-tuned. The surrogate model is subsequently used for posterior sampling of the geological model using the Randomized Maximum Likelihood (RML) method. The proposed workflow is evaluated on both two-dimensional and three-dimensional oil-water flow models. Experiments indicate that our surrogate model achieves a Root Mean Square Error (RMSE) below 0.08 and a Coefficient of Determination (R 2 ) above 0.97. Moreover, by leveraging transfer learning, the model attains comparable predictive accuracy with only 40 % of the original training data. The results demonstrate the surrogate model's capability to reduce computational costs while preserving accuracy.}, archive = {J_EAAI}, author = {Wenhao Fu and Piyang Liu and Kai Zhang and Jinding Zhang and Xu Chen and Liming Zhang and Xia Yan}, doi = {10.1016/j.engappai.2025.110873}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110873}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Use deep transfer learning for efficient time-series updating of subsurface flow surrogate model}, volume = {153}, year = {2025}, }
• (2025). Unsupervised detail and color restorer for retinex-based low-light image enhancement. EAAI, 153, 110867. (www)

  @article{ , abstract = {Retinex-based methods have demonstrated promising results in restoring low-light images to their natural, normal-light appearance. However, existing approaches often inevitably amplify hidden artifacts because the Retinex theory does not consider the various uncertain degradation patterns in dark regions. Without modeling degradations, an algorithm may easily deviate from the original color and details of regions. To address this issue, we propose a novel detail and color modeling for Retinex-based low-light image enhancement. The modeling mechanism assists our Retinex-based solution in learning rich and diverse information hidden in the dark. In addition, we develop an unsupervised loss function to reduce the solution space of Retinex decomposition. It encourages all components to mutually constrain each other, further improving the adaptiveness in unknown complex scenarios. Extensive experiments demonstrate that our approach performs favorably against state-of-the-art methods. On the SICE dataset, our method achieves 19.71 Peak Signal-to-Noise Ratio (PSNR) and 0.773 Structural Similarity Index Measure (SSIM), surpassing all compared methods in PSNR and SSIM. Our framework also generalizes robustly to the LSRW-Huawei and LSRW-Nikon benchmarks, outperforming unsupervised approaches while maintaining competitive results against supervised counterparts. The code can be accessed via: https://github.com/starsky68/DCRetinex .}, archive = {J_EAAI}, author = {Yue Sun and Yutao Jin and Xiaoyan Chen and Yanbin Xu and Xiaoning Yan and Zefu Liu}, doi = {10.1016/j.engappai.2025.110867}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110867}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised detail and color restorer for retinex-based low-light image enhancement}, volume = {153}, year = {2025}, }
• (2025). Hierarchical reinforcement learning with curriculum demonstrations and goal-guided policies for sequential robotic manipulation. EAAI, 153, 110866. (www)

  @article{ , abstract = {Autonomous learning for robotic manipulation is a significant challenge when faced with complex sequential tasks, such as picking, stacking objects, and opening drawers in industrial scenes. Hierarchical reinforcement learning (HRL) has shown great potential. However, owing to insufficient exploration and non-stationary transitions, HRL still suffers from low learning efficiency and unstable training. Therefore, this article proposes an HRL method based on curriculum demonstrations and goal-guided policies (HCDGP) for sequential manipulation. At the high level, we adopt curriculum demonstrations (CD) to direct reinforcement learning (RL) policy over options. The demonstration options indicate the order of subtasks with increasing complexity. The proportion of options from demonstration and RL can be adjusted adaptively in a curriculum-based way. At the low level, hindsight experience replay (HER) is introduced, but the HER learning process is still slow and has hindsight bias. Thus, we propose goal-conditioned intrinsic curiosity (GCIC) and goal-oriented experience replay (GOER), which enable low-level policy to learn subtasks efficiently. In summary, HCDGP can improve the performance of both high-level and low-level policies, thereby ensuring the stability of parallel training. The experimental results show that the proposed method can efficiently complete sequential tasks and achieve excellent robotic manipulation performance.}, archive = {J_EAAI}, author = {Zihao Sun and Bao Pang and Xianfeng Yuan and Xiaolong Xu and Yong Song and Rui Song and Yibin Li}, doi = {10.1016/j.engappai.2025.110866}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110866}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical reinforcement learning with curriculum demonstrations and goal-guided policies for sequential robotic manipulation}, volume = {153}, year = {2025}, }
• (2025). Bilateral triple-interaction network: An accurate segmentation model of wet-blue hide surface defects for leather industry. EAAI, 153, 110864. (www)

  @article{ , abstract = {In the production environment of the leather industry, there is an urgent need to implement automated detection of surface defects in leather due to the increasing demand for high-quality leather. However, due to the similarity of leather defects to the background, complex textures, and the variable nature of defects, automated detection faces significant challenges. To address this, this study proposes a semantic segmentation model for the segmentation of surface defects in wet-blue hide, named Bilateral Triple-Interaction Segmentation Network. First, a Boundary Branch is introduced to more accurately identify and process the subtle boundaries between defects and normal leather textures. Secondly, a bidirectional transformer feature fusion module is proposed to guide feature fusion and help maintain the continuity of global features. Finally, a Triple Branch Integrator module is proposed to enhance the fusion of details, context, and boundary features. The proposed method was evaluated on a wet-blue hide surface defect dataset, achieving Intersection over Union (IoU) scores of 83.9 %, 95.3 %, and 88.1 % for brand, hole, and broken surface defects, respectively. This study establishes a theoretical framework for the precise segmentation of leather surface defects and is expected to drive advancements in automation technology within the leather industry.}, archive = {J_EAAI}, author = {Luwen Cao and Qixin Han and Rong Luo and Liangliang Yang and Yafei Sun and Weikuan Jia}, doi = {10.1016/j.engappai.2025.110864}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110864}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bilateral triple-interaction network: An accurate segmentation model of wet-blue hide surface defects for leather industry}, volume = {153}, year = {2025}, }
• (2025). Enhancing diabetes complications prediction through knowledge graphs and convolutional networks. EAAI, 153, 110863. (www)

  @article{ , abstract = {Early prediction of diabetes complications is crucial for timely intervention and effective disease management. However, current deep learning approaches often lack sufficient representation of diabetes knowledge and rely on large-scale annotated data. To address this, we propose a Knowledge Graphs-enhanced Diabetes Complications Prediction (KGDCP) model, which integrates a physical examination knowledge graph and an improved Diabetes Knowledge Graph (DiaKG) using self-attention-based convolutional neural networks. We construct the physical examination knowledge graph based on normal reference ranges of examination indicators and enhance DiaKG through knowledge deduplication and coreference resolution, improving its accuracy and comprehensiveness. Furthermore, we employ the embeddings of these two knowledge graphs to represent examination data and diagnostic information. A token-to-token (token2token) self-attention is utilized to explore dependencies among examination indicators, and convolutional neural networks extract local features to generate representation vectors. Additionally, source-to-token (source2token) self-attention assesses dependencies between diagnostic entities and the entire entity set, with entity-to-Physical Examination (entity2PE) attention gauging entity relevance to examination data representation vectors. Our model combines these features to enhance diabetes complications prediction, outperforming traditional models in predicting cerebral infarction and peripheral neuropathy.}, archive = {J_EAAI}, author = {Haitao Cheng and Qunli Zheng and Peng Li and He Xu}, doi = {10.1016/j.engappai.2025.110863}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110863}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing diabetes complications prediction through knowledge graphs and convolutional networks}, volume = {153}, year = {2025}, }
• (2025). The hybrid velocity prediction model for pipeline detection based on bidirectional long short-term memory and an improved attention mechanism. EAAI, 153, 110855. (www)

  @article{ , abstract = {Speed prediction of pipeline detectors is crucial for accurate pipeline positioning and defect detection. This paper proposes a novel hybrid prediction model for this purpose, combining dual-layer Bidirectional Long Short-Term Memory, a bidirectional input attention mechanism, Singular Spectrum Analysis, and Bayesian Optimization. The dual-layer Bidirectional Long Short-Term Memory captures both forward and backward information in time series for prediction. The attention mechanism assigns weights to multiple input features. Singular Spectrum Analysis reconstructs and extracts features from time series data, while Bayesian Optimization is used to obtain the optimal hyperparameters for the Singular Spectrum Analysis and Bidirectional Long Short-Term Memory algorithms. A pipeline experimental platform was constructed to conduct comparative tests of the proposed model under constant speed, variable speed, non-lubricated, and lubricated conditions, assessing both operation and prediction. The results indicate that, compared to the baseline model, the hybrid prediction model proposed in this paper achieves improvements of over 9 % in Root Mean Square Error and over 0.7 % in R-Square under the most severe variable-speed conditions, demonstrating superior performance in prediction accuracy and generalization capability.}, archive = {J_EAAI}, author = {Junjie Ma and Yiming Li and Zhongchao Zhang and Tongshan Liu and Guiqiu Song}, doi = {10.1016/j.engappai.2025.110855}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110855}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The hybrid velocity prediction model for pipeline detection based on bidirectional long short-term memory and an improved attention mechanism}, volume = {153}, year = {2025}, }
• (2025). Hybrid modelling of ruble exchange rates amidst the russo-ukrainian conflict using swarm and fuzzy neural networks. EAAI, 153, 110854. (www)

  @article{ , abstract = {The existing models for predicting the Ruble exchange rate against the Chinese Yuan (CNY), Euro (EUR), British Pound (GBP), and United States Dollar (USD) may prove inadequate considering the Russia-Ukraine war. This study employs the Autoregressive Integrated Moving Average (ARIMA), the Evidential Neural Network with Gaussian Random Fuzzy Numbers (EVNN), and the Artificial Neural Network optimised with Particle Swarm Optimisation (ANN-PSO), as well as hybrids of ARIMA and EVNN (ARIMA-EVNN) and ARIMA and ANN-PSO (ARIMA-ANN-PSO), to predict CNY, EUR, GBP, and USD. When compared with ARIMA, for the training sample, ARIMA-ANN-PSO enhances the accuracy of ARIMA by 66.89%, 66.20%, 72.97%, and 66.89% for CNY, EUR, GBP, and USD respectively, while ARIMA-EVNN improves the accuracy of ARIMA for CNY by 66.21%, EUR by 78.87%, GBP by 82.43%, and USD by 80.33%. For the testing sample, the ARIMA-ANN-PSO model enhances the predictive accuracy of ARIMA by 65.60%, 64.39%, 45.74%, and 55.28% respectively, while the ARIMA-EVNN model improves the accuracy by 58.73% for CNY, 80.30% for EUR, 83.72% for GBP, and 86.18% for USD. The Russia-Ukraine war dummy was included to capture structural changes in Ruble exchange rate dynamics. Including war-related information does not change the accuracy of ARIMA model for CNY likely because China has not imposed sanctions on Russia, but improves it for EUR, GBP, and USD. However, the accuracy of EVNN decreases after integrating this information for all exchange rates. The findings can provide assistance to bureaux de change, foreign exchange traders, and governments, enabling them to make well-informed decisions.}, archive = {J_EAAI}, author = {Raad Abdelhalim Ibrahim Alsakarneh and Sagiru Mati and Goran Yousif Ismael and Serag Masoud and Nazifi Aliyu and Ahmed Samour and Berna Uzun}, doi = {10.1016/j.engappai.2025.110854}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110854}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid modelling of ruble exchange rates amidst the russo-ukrainian conflict using swarm and fuzzy neural networks}, volume = {153}, year = {2025}, }
• (2025). Global-local coupled learning method for autonomous underwater vehicle side-scan sonar image recognition. EAAI, 153, 110853. (www)

  @article{ , abstract = {Independent real-time recognition of deep-sea targets is still challenging for autonomous underwater vehicles. The side-scan sonar plays an essential role in the fields of marine bottom topographic survey and resource exploration. For the deep and distant seas that are inaccessible to humankind, the traditional methods mainly collect information first and then manually recognize it offline, which has the shortcomings of weak recognition robustness and insufficient real-time. Unlike the previous methods, a global-local coupled learning method based on side-scan sonar image recognition is proposed, which can assist in independently exploring deep-sea targets. Firstly, global recognition is performed to extract the image texture information using the deep learning-based segmentation network module and initially recognize the marine target species. Secondly, local recognition is performed, and the deep learning-based local attention module is used to optimize global recognition and refine image types. Finally, the results of the previous two steps are fused using the confidence screening strategy to output the final recognition results. The coupled learning method is compared with other classical and lightweight methods based on the side-scan sonar datasets. Simulation experiments demonstrate that the proposed method is robust and real-time, which can be widely used in marine target recognition.}, archive = {J_EAAI}, author = {Fei Yu and Xiaodong Liu and Wei Liu and Jixin Liu}, doi = {10.1016/j.engappai.2025.110853}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110853}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Global-local coupled learning method for autonomous underwater vehicle side-scan sonar image recognition}, volume = {153}, year = {2025}, }
• (2025). Enhanced lightweight detection of small and tiny objects in high-resolution images using object tracking-based region of interest proposal. EAAI, 153, 110852. (www)

  @article{ , abstract = {Detecting small objects in high-resolution images is challenging in practical applications. Downsampling the image makes small objects difficult or impossible to detect, while processing multiple low-resolution detection windows using a sliding-window approach is both time-consuming and impractical. To deal with this engineering problem, we present a novel window-based object detection system designed for detecting tiny and multi-scale objects in high-resolution images. Our approach introduces two Region of Interest (ROI) Modules that select full-resolution areas for the detector to focus on. We use a low-resolution current frame to estimate ROIs with a segmentation-based branch combined with past detection metadata to predict object locations with a tracking-based branch. By fusing outputs from these modules, we effectively recover regions with tiny objects overlooked by the estimation branch, and we significantly reduce the number of object detection runs compared to the sliding-window approach, maintaining the method’s speed. Moreover, we employ a fusion of downscaling and sliding-window techniques within large ROIs, complemented by our novel Overlapping Box Suppression (OBS) algorithm to reduce partial false-positive detections. We analyze our system and all its components on two challenging datasets — SeaDronesSee and DroneCrowd to show superior performance compared to state-of-the-art object detectors. Our approach enhances both the Artificial Intelligence (AI) and engineering domains by improving the quality and efficiency of tiny object detection, facilitating its integration into demanding real-time robotics applications. The inference code is available at https://github.com/deepdrivepl/TinyROIFusion .}, archive = {J_EAAI}, author = {Aleksandra Kos and Karol Majek and Dominik Belter}, doi = {10.1016/j.engappai.2025.110852}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110852}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced lightweight detection of small and tiny objects in high-resolution images using object tracking-based region of interest proposal}, volume = {153}, year = {2025}, }
• (2025). Fully logits guided distillation with intermediate decision learning for deep model compression. EAAI, 153, 110850. (www)

  @article{ , abstract = {Knowledge distillation, as a model compression technique, has been widely applied in artificial intelligence to improve the efficiency of deep learning models, especially in resource-constrained environments. Considering that logit contains more decision information compared to the intermediate feature maps, fully logits guided distillation is proposed, which allows student networks to have better access to the guidance from both the intermediate and decision levels of the teacher’s network. Intermediate feature logicalization is designed, which perform logit transformations on the intermediate feature maps to obtain intermediate decision information. A logit matrisation strategy is proposed, which aim to capture inter-class information of the logits. Furthermore, cross layer distillation is presented in order to enable the final logit of the teacher to provide guidance to the intermediate layers of the student. The proposed mechanism can be embedded into State-of-the-art distillation frameworks to further improve the accuracy. Experiments conducted on the CIFAR-10, CIFAR-100, and ImageNet datasets demonstrate the effectiveness of the proposed method. Image classification accuracy was used as the evaluation metric, and the results show that the proposed method improves accuracy by an average of 1.84%, with a best improvement of 3.19% over the baseline models. Code is available at https://github.com/YiqinWang-JN/FLGD .}, archive = {J_EAAI}, author = {Yiqin Wang and Ying Chen}, doi = {10.1016/j.engappai.2025.110850}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110850}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fully logits guided distillation with intermediate decision learning for deep model compression}, volume = {153}, year = {2025}, }
• (2025). Optimization of solar farm design for energy efficiency in university campuses using machine learning: A case study. EAAI, 153, 110847. (www)

  @article{ , abstract = {The increasing energy demand in multifunctional buildings, like university buildings, hinders sustainability and cost efficiency. This study addresses this issue by designing and optimizing a multi-generation solar farm using Machine Learning to enhance system performance and reduce costs. By framing Response Surface Methodology (RSM) as an interpretable supervised machine learning technique, this study combines it with the Building Energy Optimization Tool (BEopt) and Engineering Equation Solver (EES) for data modeling and optimization. The integration leverages RSM's computational efficiency to optimize exergy efficiency and cost. The novelty lies in enhancing renewable energy integration in large-scale educational facilities, an underexplored domain, using Machine Learning-driven Multi-Objective Optimization. A university building in a cooling-dominant extreme climate was selected, with annual energy demands of 18.24 Gigawatt hours (GWh) for electricity, 6.57 GWh for heating, and 7.52 GWh for cooling. A multi-generation solar farm is proposed and optimized to meet this demand and provide surplus energy, which can be stored, utilized for additional applications, or exported to the grid. The optimized solar farm generates 22.8 GWh of electricity, 17.9 GWh of heating, and 12.9 GWh of cooling, achieving an exergy efficiency of 25.69 % with an operational cost of $10.15 per hour, and a CO 2 emissions reduction of 7395 metric tons per year. This study provides a scalable and modular framework for optimizing energy management in high-demand environments, contributing to sustainability goals and energy-efficient buildings. Future studies may explore dynamic climate variations, real-time demand forecasting, and hybrid renewable energy sources to improve system resilience, adaptability, and sustainability.}, archive = {J_EAAI}, author = {Ehsanolah Assareh and Nima Izadyar and Elmira Jamei and Mohammad amin Monzavian and Saurabh Agarwal and Wooguil Pak}, doi = {10.1016/j.engappai.2025.110847}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110847}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimization of solar farm design for energy efficiency in university campuses using machine learning: A case study}, volume = {153}, year = {2025}, }
• (2025). Dual evidence enhancement and text–image similarity awareness for multimodal rumor detection. EAAI, 153, 110845. (www)

  @article{ , abstract = {The widespread use of social media has significantly facilitated the dissemination and exchange of information, while simultaneously providing fertile ground for the propagation of rumors. The rampant spread of rumors, particularly those that are meticulously crafted, has led to substantial societal harm, including public panic, misinformation, and erosion of trust. Among these, a particularly challenging category is the deep-mismatched rumor, characterized by seemingly normal text and images with aligned superficial information, yet with fabricated underlying semantics. Detecting such rumors poses significant challenges, as they bypass traditional methods that focus solely on surface-level analysis. Most existing multimodal rumor detection approaches emphasize learning the deep semantics of individual modalities and integrating them through conventional fusion techniques, such as concatenation or element-wise addition. While these methods have achieved notable success in capturing complementary information across modalities, they are hindered by two critical limitations: (1) The shallow nature of cross-modal feature fusion often fails to capture semantic inconsistencies between text and images; (2) Exclusive reliance on the content of text and images renders them ineffective against deep-mismatched rumors, which are deliberately designed to deceive. To address these challenges, we propose a novel multimodal rumor detection framework leveraging dual evidence enhancement and text–image similarity awareness. Specifically, our framework comprises three core components: (1) A text–image similarity awareness module, which quantifies the semantic alignment between textual and visual content to identify potential inconsistencies; (2) A dual evidence enhancement module, which retrieves and filters relevant textual and visual evidence from external knowledge bases and aligns this evidence with the original post through cross-attention, substantially enhancing the model’s capability to detect deep-mismatched rumors; (3) A hierarchical feature fusion mechanism based on a gated neural network, which adaptively integrates features from different modalities by accounting for their varying roles and noise levels at different stages. Extensive experiments conducted on real-world datasets from Twitter and Weibo demonstrate that our proposed method consistently outperforms state-of-the-art baselines, particularly in the detection of deep-mismatched rumors. The source code and data are available at https://github.com/xuejianhuang/DEETSA .}, archive = {J_EAAI}, author = {Xuejian Huang and Tinghuai Ma and Huan Rong and Li Jia and Yuming Su}, doi = {10.1016/j.engappai.2025.110845}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110845}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual evidence enhancement and text–image similarity awareness for multimodal rumor detection}, volume = {153}, year = {2025}, }
• (2025). A temporal graph-based contrastive approach for financial time series forecasting. EAAI, 153, 110834. (www)

  @article{ , abstract = {Graph neural networks have been used for semi-supervised learning with inter-related stocks, but their application to capturing temporal patterns in financial time series is limited. Additionally, deep similarity-based learning, which is effective for learning distinctive embeddings and overcoming nonlinearity, has not been extensively studied in financial time series forecasting. In this study, we propose a temporal graph-based contrastive approach for forecasting multi-step movement direction in stock markets. We introduce a method for generating contrastive samples based on market behavior, followed by representing time series fluctuations using graphs. Then, we propose a graph-based contrastive learning approach to embed graph samples into a new space, bringing similar samples closer and distancing dissimilar ones. Ultimately, predicting movement direction is simplified to neighborhood analysis. The proposed framework outperforms state-of-the-art benchmarks across nine major market indices, delivering improvements in Balanced Accuracy (1%–5%) and F-score (1%–6%) over various horizons. It also demonstrates excellence in financial metrics such as the Sharpe Ratio and Maximum Drawdown, emphasizing both predictive performance and practical applicability in financial analysis.}, archive = {J_EAAI}, author = {Iman Barazandeh and Saman Haratizadeh and Georgios Sermpinis}, doi = {10.1016/j.engappai.2025.110834}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110834}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A temporal graph-based contrastive approach for financial time series forecasting}, volume = {153}, year = {2025}, }
• (2025). Iterative under-sampling for delisting risk prediction of chinese listed companies based on financial information. EAAI, 153, 110830. (www)

  @article{ , abstract = {The delisting of companies not only undermines corporate interests but also causes significant financial losses on investors, making the development of effective early warning models in delisting risk prediction (DRP) essential for improving corporate governance and guiding investment decisions. A major challenge in DRP is the class-imbalance problem between potential delisting and normal companies. The cost of false negatives is significantly higher than that of false positives, as missing a delisting event may lead to substantial financial losses. To address this, we propose a novel under-sampling framework inspired by the cautious decision-making mindset of investors. Our approach iteratively applies a soft-margin selection strategy during inter-group under-sampling to construct a balanced dataset, enabling base learners to focus on potential delisting cases while maintaining a robust precision–recall balance. We validate our method through extensive experiments on datasets collected from China’s security market by comparing it with eighteen existing imbalanced DRP techniques across four performance metrics. The results are rigorously analyzed using statistical methods. These findings demonstrate the superiority of our method for risk assessment across multiple timeframes in financial markets.}, archive = {J_EAAI}, author = {Yun-Hao Zhu and Zhong-Liang Zhang and Wan-An Liu}, doi = {10.1016/j.engappai.2025.110830}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110830}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Iterative under-sampling for delisting risk prediction of chinese listed companies based on financial information}, volume = {153}, year = {2025}, }
• (2025). Rapid prediction of thermal stress on satellites via domain decomposition-based hybrid fourier neural operator. EAAI, 153, 110826. (www)

  @article{ , abstract = {Rapid thermal stress analysis is crucial for the thermal design of satellites. To overcome the disadvantages of traditional algorithms in terms of efficiency, deep learning methods have been used to tackle these problems. However, using uniform grid-based techniques is challenging when faced with complex geometric shapes. To address this, we introduce the domain decomposition-based Hybrid Fourier Neural Operator (HFNO), a comprehensive framework for learning a multi-scale and end-to-end operator on two-dimensional point clouds. We then propose two decomposition metrics: a stress gradient-based metric for scenarios with prior knowledge of training data, and a mesh density-based metric for scenarios without prior knowledge. Leveraging K-Dimension tree-based domain decomposition optimized via Monte Carlo tree search, we decompose the computational domain into several disjoint rectangular subdomains. In the proposed hybrid framework, a Geometry-aware Fourier Neural Operator (Geo-FNO) is used to deal with subdomains with high-frequency information, while a Non-Uniform Fourier Neural Operator (NU-FNO) is used to deal with subdomains with low-frequency information. This framework effectively combines the advantages of two Fourier Neural Operator variants, overcoming the issue of large prediction errors on the subdomains with high-frequency information and ensuring stable prediction performance across different positions. Furthermore, we introduce a boundary loss term during the training process to enhance continuity across subdomain boundaries. The numerical results demonstrate that our method achieves a superior balance between efficiency and precision, surpassing that of a single algorithm.}, archive = {J_EAAI}, author = {Kangrui Zhou and Wei Peng and Xiaoya Zhang and Xu Liu and Wen Yao}, doi = {10.1016/j.engappai.2025.110826}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110826}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rapid prediction of thermal stress on satellites via domain decomposition-based hybrid fourier neural operator}, volume = {153}, year = {2025}, }
• (2025). TSSA-net: Transposed sparse self-attention-based network for image super-resolution. EAAI, 153, 110823. (www)

  @article{ , abstract = {Image super-resolution (SR) aims to reconstruct high-resolution images from low-resolution ones. Transformer-based methods have recently demonstrated remarkable results, but the conventional dense self-attention mechanism fails to capture local relationships between patches, leading to suboptimal performance. Moreover, the recovery of high-frequency information, crucial for edge reconstruction, is often insufficient. To address these challenges, we propose the Transposed Sparse Self-Attention (TSSA) mechanism, which improves local feature attention by restructuring the self-attention computation, without using convolutions. Additionally, we introduce a Segmented Convolutional Feed-Forward Network (SCFFN) to enhance high-frequency detail recovery and local feature acquisition, while maintaining a low parameter count. We combine TSSA, SCFFN, and a channel attention mechanism to develop TSSA-Net, an innovative network for image super-resolution. Comprehensive evaluations on classical, lightweight, and real-world SR tasks show that TSSA-Net outperforms recent methods on the Set14, Urban100, and Manga109 datasets, with improvements of 0.01–0.04 decibel (dB), 0.11–0.12 dB, and 0.01–0.08 dB, respectively. TSSA-Net achieves notable results in both metric-based and visual-based evaluations. The code is available at https://github.com/VMC-Lab-Chen/TSSA-Net .}, archive = {J_EAAI}, author = {Guanhao Chen and Dan Xu and Kangjian He and Hongzhen Shi and Hao Zhang}, doi = {10.1016/j.engappai.2025.110823}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110823}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {TSSA-net: Transposed sparse self-attention-based network for image super-resolution}, volume = {153}, year = {2025}, }
• (2025). A new hybrid data and model-centric predictive approach dedicated to industrial pipe maintenance. EAAI, 153, 110821. (www)

  @article{ , abstract = {Predictive Maintenance (PdM) for pipe clogging is a critical challenge in the industrial sector, particularly with the increasing adoption of Artificial Intelligence (AI) and the Internet of Things (IoT). Frequent clogging incidents, such as those faced by Orano/La Hague, lead to energy waste, operational inefficiencies, financial losses, and potential safety hazards, highlighting the critical need for effective maintenance solutions to protect both assets and personnel. This study proposes a novel hybrid approach that combines the strengths of data-centric and model-centric methodologies for Prognostic and Health Monitoring (PHM) of pipeline systems in constrained industrial environments. The approach utilizes passive acceleration measurements to predict clogging occurrences and quantify clogging severity under varying airflow rates. Experimental results indicate that the proposed method achieves up to 100% accuracy in clogging detection and robust performance across diverse operational conditions. This integrated methodology represents a significant step forward in predictive maintenance, offering scalable and adaptable solutions to enhance safety, operational efficiency, and cost-effectiveness in industrial settings.}, archive = {J_EAAI}, author = {Ahmad Braydi and Pascal Fossat and Alessandro Casaburo and Victor Pernet and Cyril Zwick and Mohsen Ardabilian and Olivier Bareille}, doi = {10.1016/j.engappai.2025.110821}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110821}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new hybrid data and model-centric predictive approach dedicated to industrial pipe maintenance}, volume = {153}, year = {2025}, }
• (2025). Using meta-reinforcement learning for solving the virtual network embedding problem. EAAI, 153, 110820. (www)

  @article{ , abstract = {Introduction of Network Function Virtualization (NFV), a technique for virtualization of infrastructure on physical servers, requires careful, efficient management of substrate network resources. This has resulted in formulation of Virtual Network Embedding Problem (VNEP) for efficient deployment of virtual network functions in the substrate network. While the holder of a substrate network earns the highest revenue as the acceptance rate of customers’ embedding requests increases, they should optimize the costs for embedding the required resources. Techniques employed so far for solving VNEP often suffer from poor design, long runtime, and the need for large training sets and maintaining precision in large-scale environments. In this work, we introduce Fast Context Adaptation via Meta-Learning Virtual Network Embedding Problem (CAVIA-VNEP), a technique based on meta-reinforcement learning (MRL) as a fusion of meta-learning – capable of learning to learn, and reinforcement learning (RL). We train RL at a high computational speed with a small amount of data while using CAVIA for efficient training of neural networks. Simulation results demonstrate that CAVIA-VNEP outperforms three baseline, heuristic, and RL-based techniques in terms of long-term revenue ratio, long-term revenue-to-cost ratio, long-term acceptance ratio, and utilization of substrate nodes and links while facing challenges for reducing the average path length.}, archive = {J_EAAI}, author = {Sahand Torkamani-Azar and Mohsen Jahanshahi and Alireza Hedayati}, doi = {10.1016/j.engappai.2025.110820}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110820}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Using meta-reinforcement learning for solving the virtual network embedding problem}, volume = {153}, year = {2025}, }
• (2025). A dynamic fuzzy video compression control algorithm for wireless advanced driver assistance systems. EAAI, 153, 110815. (www)

  @article{ , abstract = {As video-based Advanced Driver Assistance Systems (ADAS) become integral to modern vehicle safety, the demand for reliable, high-performance wireless solutions for retrofitting vehicles have grown. This study introduces a wireless ADAS that employs a novel dynamic video compression control algorithm, integrating a hardware-based Motion Joint Photographic Experts Group (MJPEG) compression engine with adaptive fuzzy logic control strategies. The system dynamically adjusts video compression levels based on real-time conditions such as wireless data rates and available bandwidth, addressing key challenges in maintaining video quality and minimizing latency in wireless environments. The adaptive control is governed by two distinct fuzzy control strategies: Fuzzy Rule-Based (FRB) and Evolutionary Fuzzy Rules (EFR). Both strategies were optimized using nature-inspired algorithms, including Differential Evolution (DE), Particle Swarm Optimization (PSO), and Genetic Programming (GP). Among these, the EFR-based control was found to offer the best overall performance. Key performance indicators such as compression efficiency, latency, and throughput rates were thoroughly evaluated. Experimental results demonstrated that the EFR-based system provided up to a 35% improvement in compression efficiency compared to traditional methods, reduced video latency by approximately 20%, and optimized data throughput. Furthermore, the EFR-based control showcased enhanced generalization capabilities, outperforming FRB-based control under previously unobserved conditions, which is critical for real-world vehicular applications where network conditions may vary significantly. The implementation of artificial intelligence in the form of EFR significantly enhanced the system’s ability to adapt to varying data rates and environmental conditions, making it a promising solution for real-time video compression in computationally constrained embedded systems.}, archive = {J_EAAI}, author = {M. Prauzek and P. Krömer and J. Konecny and M. Stankus and P. Musilek}, doi = {10.1016/j.engappai.2025.110815}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110815}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dynamic fuzzy video compression control algorithm for wireless advanced driver assistance systems}, volume = {153}, year = {2025}, }
• (2025). Altitude-informed fusion pyramid network for multi-scale waste detection in unmanned aerial vehicle images. EAAI, 153, 110814. (www)

  @article{ , abstract = {Accurate waste detection using unmanned aerial vehicles (UAVs) remains a significant challenge due to the non-canonical perspective, varying altitudes, and diverse nature of waste materials such as plastic, with its translucency and irregular appearance. Traditional object detection architectures struggle to adapt to these factors, leading to reduced detection accuracy. To address these challenges, we propose the use of additional contextual information by incorporating altitude information of UAV waste images to enhance the multi-scale detection capabilities of feature pyramid networks, dynamically assigning differential importance to feature fusion modules. We propose a novel altitude-informed tiny-object detection architecture (named AltiDet) in three configurations of backbone sizes, consisting of an Asymmetric Deep Aggregation (ADA) and High-Resolution Feature Extraction (HRFE) module as a backbone, an Altitude-Informed Fusion Pyramid Network (A-IFPN) and an altitude-scaled loss function. Our ADA+HRFE backbone aggregates extracted features iteratively using different nodes while capturing additional spatial and contextual features from the high-resolution layers, to account for the fine-grained and irregular features of waste. The A-IFPN fuses and aggregates Altitude-Weighted Frequency Attention (AWFA) modules to further extract meaningful feature maps from the backbone, improving multi-scale detection. Our altitude-scaled loss function ensures the more challenging higher altitude images are greater emphasized during training. Extensive experiments conducted on our collected multi-class aerial waste dataset (named AltiWaste), the Solid Waste Aerial Detection (SWAD), Trash Annotations in Context (TACO) and other datasets demonstrate the effectiveness and advantages of our proposed architecture, yielding a 1.41% improvement over state-of-the art detectors on the AltiWaste dataset.}, archive = {J_EAAI}, author = {Chan Yue Liew and Joanne Mun-Yee Lim and Chee Pin Tan and Raja Mazhar Mohar Bin Tun Mohar}, doi = {10.1016/j.engappai.2025.110814}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110814}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Altitude-informed fusion pyramid network for multi-scale waste detection in unmanned aerial vehicle images}, volume = {153}, year = {2025}, }
• (2025). Automatic inverse design of second-order differential metasurfaces based on reinforcement learning. EAAI, 153, 110812. (www)

  @article{ , abstract = {Second-order differential metasurfaces, as two-dimensional artificial nanostructures, hold great potential for various applications in optical computing, image processing, microscopy imaging, etc. They have the capability to perform second-order mathematical operations on incident optical fields. In general, the design of second-order differential optical processing metasurfaces needs to consider various constraints, including incident optical fields, material properties, and application requirements. This task is challenging and time-consuming, typically requiring manual intervention. Therefore, a reinforcement learning-based artificial intelligence algorithm is proposed for the automatic design of second-order differential metasurfaces. It can rapidly find the desired result by leveraging the powerful self-learning ability of reinforcement learning. This paper first introduces the basic principles of designing second-order differential optical processing metasurfaces. Then, it presents a new automatic design method for second-order differential metasurfaces based on reinforcement learning. Finally, a design of a second-order differential metasurface is presented. Simulation and experimental results validate the effectiveness of the proposed method. The proposed method establishes a basic framework for applying reinforcement learning to the automatic design of photonic devices and outlines the various constraint factors that need to be coupled and considered during the design process. A surrogate model based on neural networks is presented to fit the transmission calculated by finite difference time domain software. A new reward formula is given to sufficiently consider the transmission efficiency of whole devices. The proposed method effectively leverages the self-learning capability of reinforcement learning, circumventing challenges related to manual intervention, computational complexity, and time-consuming tasks in conventional methods.}, archive = {J_EAAI}, author = {Zhaohui Li and Shuo Yang and Delian Liu}, doi = {10.1016/j.engappai.2025.110812}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110812}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic inverse design of second-order differential metasurfaces based on reinforcement learning}, volume = {153}, year = {2025}, }
• (2025). Multi-agent reinforcement learning for vibration control of regenerative active suspension. EAAI, 153, 110809. (www)

  @article{ , abstract = {The comfort and responsiveness of active suspension systems have surpassed those of traditional suspension systems, but their development and implementation have been significantly constrained by high power requirements. In response to the current drive for energy conservation and emission reduction, regenerative active suspension systems effectively address the issues of high power consumption and energy loss while demonstrating considerable market potential. The main contribution of the investigation lies in designing a novel regenerative active suspension system specifically for electric vehicles powered by electric motors. The system integrates an electric motor and a generator to ensure precise power delivery and energy recovery from suspension dynamics. To fulfill the dual requirements of energy regeneration and ride smoothness, the Double Proximal Policy Optimization-Multi Dimensional Output (DPPO-MDO) algorithm has been devised based on multi-agent reinforcement learning. The algorithm features two agents equipped with the Proximal Policy Optimization (PPO) algorithm, each managing the operations of the motor and generator respectively. Furthermore, to accommodate diverse scenarios and balance comfort with energy efficiency, economic and comfort modes have been developed for the regenerative active suspension through the DPPO-MDO algorithm. Comprehensive experimental analysis demonstrates that both redesigned reward function and innovative reinforcement learning framework significantly enhance the application of multi-agent reinforcement learning in regenerative active suspension systems.}, archive = {J_EAAI}, author = {Xiaotian Gao and Yu Du and Shiyuan Han and Wenxiu Zhao and Jin Zhou and Tong Zhang and C.L. Philip Chen}, doi = {10.1016/j.engappai.2025.110809}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110809}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-agent reinforcement learning for vibration control of regenerative active suspension}, volume = {153}, year = {2025}, }
• (2025). Design of an efficient fault-tolerant quantum-computing circuit with quantum neural network learning. EAAI, 153, 110808. (www)

  @article{ , abstract = {Fault-tolerance is key to the practical realization of quantum computation, but the design of an efficient, low-overhead and fault-tolerant error-correction circuit remains a major challenge so far. To help address this issue, we first propose a noise-adaptive dissipative quantum neural network (DQNN) model to mitigate the effects of error propagation for constructing a fault-tolerant quantum circuit. Then, we develop a method for preparing a fault-tolerant auxiliary entangled state based on the DQNN model, reducing computational delays and qubit resource consumption. This method utilizes the adaptability of quantum machine learning to the distribution of noisy inputs and its practicality in noisy intermediate scale quantum devices, thus reducing interaction with classical computers and further optimizing the real-time requirements for active error-correction. By integrating quantum error-correction and quantum neural network learning, this DQNN scheme provides a novel solution for constructing scalable fault-tolerant quantum computation. Compared with existing fault-tolerant methods, the DQNN process requires fewer error-propagation efforts and offers higher fidelity in a noisy environment for error thresholds higher than 1 0 − 4 . The effectiveness of this method is verified through experimental simulations using the Qiskit. The code for experiments and model in this paper can be found on GitHub: https://github.com/Ricardo-Vv/Qiskit_exam/tree/master .}, archive = {J_EAAI}, author = {Chen Lin and Rucong Xu and Yun Li}, doi = {10.1016/j.engappai.2025.110808}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110808}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Design of an efficient fault-tolerant quantum-computing circuit with quantum neural network learning}, volume = {153}, year = {2025}, }
• (2025). Adaptive metagraph neural network assisted by metagraph search for financial fraud detection. EAAI, 153, 110807. (www)

  @article{ , abstract = {Financial transaction fraud detection is a critical technology for ensuring the security and stability of financial markets. Artificial intelligence, particularly graph neural networks, has demonstrated superior performance in fraud detection. However, challenges remain, such as limited interpretability, difficulty in adapting to new types of fraud in a timely manner, and incomplete data mining. To address these challenges, we propose a novel graph neural network model called Metagraph Fraud Detection Graph Neural Networks (MetaFraud-GNN), which leverages metagraph search and neural architecture search (NAS) techniques to automatically optimize the network structure for financial transaction fraud detection. MetaFraud-GNN extracts complex patterns from financial transaction networks through metagraph search algorithms, a technique that automatically mines key subgraph patterns. These metagraph capture fraudulent patterns and enable the model to more comprehensively uncover hidden information within transaction data, thus enhancing its processing efficiency. Additionally, the metagraph decoding algorithm optimizes the graph neural network structure by training on the most effective metagraph to adapt to evolving fraudulent methods. This approach improves both the accuracy and adaptability of fraud detection. We conduct experiments on three real-world public benchmark datasets YelpChi, Amazon, and Elliptic and demonstrate that our model significantly outperforms existing benchmark methods on various performance metrics. Such as F1-macro, Area Under the Receiver Operating Characteristic Curve(AUC) and Geometric Mean(GMean), by 4.46%, 2.67%, and 8.59% on YelpChi, 0.14% and 2.19% on Amazon,The F1 indicator has not been upgraded, respectively.}, archive = {J_EAAI}, author = {Guoxiang Tong and Junjie Qian and Jieyu Shen}, doi = {10.1016/j.engappai.2025.110807}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110807}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive metagraph neural network assisted by metagraph search for financial fraud detection}, volume = {153}, year = {2025}, }
• (2025). Enhanced medical image segmentation using U-net with residual connections and dual attention mechanism. EAAI, 153, 110794. (www)

  @article{ , abstract = {Convolutional Neural Network has an excellent ability to excavate local features but cannot perceive global information. Vision Transformer shows superior global modeling ability through the self-attention mechanism. Still, the lack of inductive bias unique to Convolutional Neural Networks and the significant computational complex parameters make it slightly less effective on small and medium-sized datasets. To address this issue, this paper presents an enhanced medical image segmentation model, the U-Net model, based on residual connections and a dual attention mechanism (RDTransUnet). RDTransUnet supplements the feature information lost during convolution by incorporating residual connections in the encoding stage, enabling more comprehensive image feature extraction. Furthermore, a dual attention mechanism is introduced in the skip connections to replace the traditional skip connection in U-Net. This mechanism reintegrates and superimposes the encoder outputs through spatial and channel attention, thereby better capturing global and local dependencies. In the decoding stage, an efficient channel attention feature fusion method is utilized to increase the network’s response to valuable features and suppress redundant ones, improving segmentation accuracy. Experiments on six public datasets have shown that RDTransUnet outperforms baseline models and the vast majority of state-of-the-art models and demonstrates more accurate segmentation performance on small and medium-sized datasets.}, archive = {J_EAAI}, author = {Leyi Xiao and Jiaojiao Song and Xia Xie and Chaodong Fan}, doi = {10.1016/j.engappai.2025.110794}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110794}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced medical image segmentation using U-net with residual connections and dual attention mechanism}, volume = {153}, year = {2025}, }
• (2025). Strategic tour operator selection in the tourism sector using a quantum picture fuzzy rough set-based multi-criteria decision-making approach. EAAI, 153, 110793. (www)

  @article{ , abstract = {Tour operator selection is critical for ensuring high-quality services, customer satisfaction, and sustainable tourism development. However, traditional decision-making methods often fail to address the complexities and uncertainties involved in this process. This study introduces a robust decision-making framework that integrates quantum picture fuzzy rough sets (QPFR) with advanced Multi-Criteria Decision-Making (MCDM) techniques to enhance the evaluation and selection of tour operators. The methodology incorporates QPFR, the Decision-Making Trial and Evaluation Laboratory (DEMATEL), and the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) to assess and rank seven prominent tour operators in the Turkish tourism sector. The evaluation is based on 16 comprehensive criteria: quality, safety, environmental impact, authenticity, and economic contribution. Expert inputs and artificial intelligence techniques were utilized to ensure the model's reliability and accuracy. The findings reveal that the proposed model effectively minimizes uncertainties, provides consistent rankings, and highlights the critical importance of specific criteria in decision-making. Sensitivity analysis confirms the robustness of the results, demonstrating the model's applicability to dynamic and complex decision-making contexts. This study offers theoretical contributions and practical insights for decision-makers, emphasizing the value of integrating advanced computational methods to support sustainable tourism development.}, archive = {J_EAAI}, author = {Ömer Faruk Görçün and Dragan Pamucar and Hasan Dinçer and Serhat Yüksel and Ismail İyigün and Vladimir Simic}, doi = {10.1016/j.engappai.2025.110793}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110793}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Strategic tour operator selection in the tourism sector using a quantum picture fuzzy rough set-based multi-criteria decision-making approach}, volume = {153}, year = {2025}, }
• (2025). Self-learning stationary subspace analysis for fault detection of industrial processes with varying operation conditions. EAAI, 153, 110792. (www)

  @article{ , abstract = {The Fault detection is critical for ensuring the safety and reliability of modern industrial processes. Due to load fluctuations, set point changes, etc., industrial processes often undergo time-varying operation conditions. Therefore, fault detection of nonstationary process has gained more and more attention in recent years. It is usually assumed that nonstationary industrial data remains in a long-term constant dynamic equilibrium relationship. However, this assumption is easily violated by significant changes in operation conditions, which breaks the original dynamic equilibrium relationship and brings great challenges to existing fault detection methods. To address the problem of monitoring industrial processes with varying operation conditions, a novel self-learning stationary subspace analysis algorithm with continual learning capability is proposed in this article. The proposed method involves building an initial model based on stationary subspace analysis and then updating the model only with the data from new operation conditions by the proposed asymmetric synaptic intelligence. This ensures that the updated model can effectively monitor both the new and historical operation conditions by adequately measuring the importance of the model parameters, and restricting the change of the critical parameters of the previous operation conditions. This method does not store old data, which significantly reduces the demands on storage space and computational efficiency, thereby facilitating the online deployment of the model. In addition, a parallel augmented Lagrangian algorithm is designed to estimate the model parameters, further enhancing the efficiency of the model. Finally, three challenging case studies with varying operation conditions are implemented to verify the effectiveness of the proposed method, including a numerical case, a closed-loop continuous stirred tank reactor, and a real roasting process.}, archive = {J_EAAI}, author = {Dehao Wu and Jianan Deng and Jingxin Zhang and Keke Huang and Chunhua Yang and Weihua Gui}, doi = {10.1016/j.engappai.2025.110792}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110792}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-learning stationary subspace analysis for fault detection of industrial processes with varying operation conditions}, volume = {153}, year = {2025}, }
• (2025). Autonomous control of soft robots using safe reinforcement learning and covariance matrix adaptation. EAAI, 153, 110791. (www)

  @article{ , abstract = {The control of soft robots (such as continuum robots) poses significant challenges due to their coupled dynamics with significant inherent nonlinearities. Recently, model-free reinforcement learning algorithms have been proposed as an attractive alternative to model-based methods to address such a challenging control problem through unsupervised learning. However, the safety of robots is usually ignored while training such algorithms. This is particularly important for medical applications of soft robots. Also, the curse of dimensionality in soft robots makes it difficult for a reinforcement learning algorithm to develop an optimal controller. In this work, we propose a safe phasic soft actor–critic algorithm with a covariance matrix adaptation network which is then tested on different soft robots. We demonstrate that the proposed algorithm could learn an optimal policy quickly while satisfying the safety constraints. We formulated and tested our algorithm for (i) multigait soft robot; (ii) soft gripper robot; and (iii) soft robotic trunk. The proposed algorithm achieved an average of 150% higher rewards compared to other state-of-the-art algorithms. Also, adding the safety layer helped reduce the tracking error by 8 times when compared to the algorithm without a safety layer. The policy is validated in Simulation Open Framework Architecture (SOFA) simulations against other state-of-the-art algorithms in terms of tracking errors.}, archive = {J_EAAI}, author = {Shaswat Garg and Masoud Goharimanesh and Sina Sajjadi and Farrokh Janabi-Sharifi}, doi = {10.1016/j.engappai.2025.110791}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110791}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Autonomous control of soft robots using safe reinforcement learning and covariance matrix adaptation}, volume = {153}, year = {2025}, }
• (2025). Handling heterogeneity in human activity recognition data by a compact long short term memory based deep learning approach. EAAI, 153, 110788. (www)

  @article{ , abstract = {The rapid growth and development of sensor technology in wearable devices has encouraged researchers to develop new Human Activity Recognition (HAR) models. The data collected from various devices is affected by different factors, such as users, environments, and sensor frequencies. These factors cause heterogeneity in the collected data. Traditional machine learning models and single deep learning approaches struggle to effectively manage and handle data heterogeneity. To address this challenge, a novel deep neural network based on combining Long Term Short Memory (LSTM) layers with convolutional layers as a hybrid approach is proposed in this paper. This model streamlines data preparation through efficient segmentation and standardization, avoiding time-consuming techniques and methods used in our previous study. Our compact architecture minimizes computational overhead while enhancing interpretability, making it suitable for resource-constrained environments. The LSTM layer of the model is used for temporal feature extraction, whereas the convolution layer is used for spatial feature extraction from the data. The parameters of the model were reduced by replacing a fully connected layer with a batch normalization (BN) layer, which offers a significant advantage over other methods in the field. The performance of the model was evaluated on four heterogeneous HAR datasets. The overall F1-score accuracy model of the datasets: phone accelerometer, watch accelerometer, phone gyroscope, and watch gyroscope reached 97%, 89%, 89%, and 87%, respectively. The results prove that the proposed hybrid Long Short Term Memory and Convolutional Neural Network (LSTM-CNN)the model has superior performance compared to other reported results. By effectively addressing the challenges of heterogeneous data and obtaining superior scores, this model not only demonstrates robust performance but also contributes valuable insights for future research in human activity recognition.}, archive = {J_EAAI}, author = {Ahmed Cemiloglu and Bahriye Akay}, doi = {10.1016/j.engappai.2025.110788}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110788}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Handling heterogeneity in human activity recognition data by a compact long short term memory based deep learning approach}, volume = {153}, year = {2025}, }
• (2025). Dual-phase airway segmentation: Enhancing distal bronchial identification with anatomical prior guidance. EAAI, 153, 110786. (www)

  @article{ , abstract = {Airway segmentation and reconstruction are critical for preoperative lesion localization and surgical planning in pulmonary interventions. However, this task remains challenging due to the intrinsically complex tree structure of the airway and the imbalance in branch sizes. While current deep learning methods focus on model architecture optimization, they underutilize anatomical priors such as the spatial correlation between pulmonary arteries and bronchi beyond geometric grading level III. To address this limitation, we propose a dual-decoding segmentation network (DDS-Net) integrated with a pulmonary-bronchial extension generative adversarial network (PBE-GAN), which explicitly embeds artery-bronchus adjacency priors to enhance distal bronchial identification. Experimental results demonstrate state-of-the-art performance, achieving a Dice Similarity Coefficient (DSC) of 88.46%, Branch Detection Rate (BD) of 88.31%, and Tree Length Detection Rate (TD) of 84.93%, with significant improvements in detecting peripheral bronchi near pulmonary arteries. This study confirms that incorporating anatomical relationships substantially improves segmentation accuracy, particularly for fine structures. Future work should prioritize clinical validation through multi-center trials and explore integration with real-time surgical navigation systems, while extending similar anatomical synergy principles to other organ-specific segmentation tasks.}, archive = {J_EAAI}, author = {Zhen Zhang and Wen Zhang and Liqin Huang and Lin Pan and Shaohua Zheng and Zheng Liu and Weisheng Chen and Penggang Bai}, doi = {10.1016/j.engappai.2025.110786}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110786}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-phase airway segmentation: Enhancing distal bronchial identification with anatomical prior guidance}, volume = {153}, year = {2025}, }
• (2025). Safe event-triggered control of unmanned surface vehicles with gaussian processes: Resilience in denial of service attacks and uncertain dynamics. EAAI, 153, 110776. (www)

  @article{ , abstract = {This paper addresses critical security challenges in cyber–physical systems arising from uncertain system dynamics and cyberattacks by proposing a learning-based event-triggered control protocol for networked unmanned surface vehicles (USVs). Leveraging Gaussian process regression, the proposed data-driven approach ensures the stabilization of USVs within a guaranteed error bound. Furthermore, a resilient event-triggered strategy is developed to maintain control performance under denial-of-service (DoS) attacks. Additionally, a rigorous stability analysis is conducted for USVs with unknown dynamics, specifying stabilization conditions of durations and frequencies of non-structured DoS attacks. Simulation results, including Monte Carlo tests, demonstrate the effectiveness of the proposed approach, highlighting its robustness and efficiency compared to time-triggered and non-learning-based methods.}, archive = {J_EAAI}, author = {Zewen Yang and Xiaobing Dai and Liang Fang and Jiajia Zhou and Zheping Yan}, doi = {10.1016/j.engappai.2025.110776}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110776}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Safe event-triggered control of unmanned surface vehicles with gaussian processes: Resilience in denial of service attacks and uncertain dynamics}, volume = {153}, year = {2025}, }
• (2025). A neural rendering system for fashion design process. EAAI, 153, 110773. (www)

  @article{ , abstract = {The translation of images into detailed three-dimensional (3D) models represents a critical challenge in digital content creation, particularly for the Creative Industries (CI). Traditional 3D modeling methods are resource-intensive, while recent advances in Neural Rendering (NR) have introduced efficient and automated solutions. In the fashion industry, where visual fidelity and rapid prototyping are crucial, NR techniques such as Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting (3DGS) are becoming relevant resources to digitize complex geometries and textures with a low-cost approach, enabling different applications like virtual fitting rooms, immersive e-commerce, and digital prototyping. However, the evaluation of their effectiveness in this field is still in its early phases. To address the industry’s needs, we propose Fashion immersive Neural Rendering Interface (FENRI) , a novel framework that integrates NR techniques for reconstructing 3D models from 2D images of fashion items. FENRI includes a WebXR-based visualization platform that allows immersive comparison and evaluation of NR-generated 3D models, supporting both experts and non-experts in selecting optimal designs. We applied FENRI to footwear design, collecting a novel dataset to compare NeRF and 3DGS methods through quantitative and qualitative analyses. In our study, the 3DGS method demonstrated superior performance over NeRF, as highlighted by the higher Peak Signal-to-Noise Ratio (PSNR) values (37.65 vs. 29.03, respectively) and Structural Similarity Index Measure (SSIM) scores (0.99 vs 0.96), while exhibiting a lower Learned Perceptual Image Patch Similarity (LPIPS) values (0.01 vs 0.04). Moreover, we showed that, by applying classical mesh post-processing techniques, we can increase the topological and visual quality of the 3D models synthesized by NR methods. These findings highlight the potential of NR techniques in the fashion industry’s digital pipeline. By enabling rapid, immersive, and visually compelling design iterations, FENRI offers a scalable solution to the fashion industry’s demand for high-quality 3D reconstructions, promoting innovation and sustainability.}, archive = {J_EAAI}, author = {Emanuele Balloni and Lorenzo Stacchio and Adriano Mancini and Emanuele Frontoni and Primo Zingaretti and Marina Paolanti}, doi = {10.1016/j.engappai.2025.110773}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110773}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A neural rendering system for fashion design process}, volume = {153}, year = {2025}, }
• (2025). Knowledge-guided and collaborative learning network for camouflaged object detection. EAAI, 153, 110771. (www)

  @article{ , abstract = {Camouflaged Object Detection (COD) task is more arduous than other target detection tasks because of the challenges of complex and variable contours, scale diversity, and high similarity to the background. Moreover, in pursuit of better detection performance, most existing COD methods usually require many parameters and computational complexity, which undoubtedly increases the difficulty and cost of implementation. In this paper, by revisiting this difficult task, we find that the collaborative effect of extracting explicit edge knowledge and global semantics highlights the camouflaged regions more and significantly affects detection efficiency. We propose a more comprehensible and efficient network for COD, namely the Knowledge-guided and Collaborative Learning Network (KCNet). It comes with more complex camouflaged targets by introducing rough features containing a large amount of edge knowledge and global semantics to interact with other components fully and collaborative learning. Specifically, we gradually introduce the rough features acquired by the preliminary perception unit to the layers by designing a knowledge-guided positioning module. Secondly, we design a detail enhancement module to enhance the extraction of the detailed parts of semantic information at deeper levels. Finally, we give a convolutional decoding unit to output the complete camouflaged target prediction information. Extensive experimental results show that our model reaches the state-of-the-art on four challenging datasets, outperforming 22 advanced models currently available. Meanwhile, KCNet exhibits a low number of parameters, a low computational complexity, and a very competitive inference speed, all of which are advantages over existing methods. The codes are released at https://github.com/wd61419/KCNet .}, archive = {J_EAAI}, author = {Dan Wu and Mengyin Wang and Jing Sun and Xu Jia}, doi = {10.1016/j.engappai.2025.110771}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110771}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Knowledge-guided and collaborative learning network for camouflaged object detection}, volume = {153}, year = {2025}, }
• (2025). Auto feature weighted c-means type clustering methods for color image segmentation. EAAI, 153, 110768. (www)

  @article{ , abstract = {To address the limitations of existing hard c -means (HCM) and fuzzy c -means (FCM) methods, we develop four novel clustering methods: vector-weighted alternative hard c -means (VWAHCM), matrix-weighted alternative hard c -means (MWAHCM), vector-weighted alternative fuzzy c -means (VWAFCM), and matrix-weighted alternative fuzzy c -means (MWAFCM). These methods enhance clustering performance by incorporating non-Euclidean norm metrics and vector-weighted and matrix-weighted schemes without adding extra parameters. Our methods modify the traditional weight constraint from a sum to a product of weights, thereby improving robustness and accuracy. Comprehensive experiments conduct on various real-world datasets and color image segmentation tasks demonstrate the superiority of the proposed methods over traditional HCM and FCM variants. The results show significant improvements in clustering Accuracy ( A C C ), Normalized mutual information ( N M I ), Rand index ( R I ), and Fowlkes–Mallows index ( F M ). Furthermore, the proposed methods exhibit fast convergence and robust performance, proving their effectiveness in practical applications.}, archive = {J_EAAI}, author = {Sijia Zhu and Zhe Liu and Sukumar Letchmunan and Haoye Qiu}, doi = {10.1016/j.engappai.2025.110768}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110768}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Auto feature weighted c-means type clustering methods for color image segmentation}, volume = {153}, year = {2025}, }
• (2025). DeepTrackStat: An end-to-end deep learning framework for extraction of motion statistics from videos of particles. EAAI, 153, 110761. (www)

  @article{ , abstract = {Particle tracking (PT) is a mature area of research that traditionally has used Gaussian filtering and nearest neighbors-based algorithms to detect and link features in a sequence of images. PT shares many similarities with the general task of object tracking, although it is specifically designed for tracking objects that are usually small and have a distinct shape shared amongst all particles in images with little to no background. To try and improve the use of modern computer vision techniques in particle tracking, we present DeepTrackStat (DTS), a novel class-based deep learning framework for the engineering application of extracting motion statistics from videos of particles. DTS is able to bypass the tracking process entirely and generate accurate statistics on speed, velocity components, and turn angle for a wide range of PT scenarios including trajectories derived from Brownian motion, Poiseuille flow, and porous media flow. On our simulated test set, DTS results in low errors of 2.8 pixels/frame for particle speed and 2.51 degrees for turn angle The model is robust to large variations in particle size, shape, brightness, speed, density, and signal to noise ratio, and can reduce the time required to obtain the target statistics from videos of moving particles by 6x (when compared with classical methods). In addition, we show that DTS outperforms a state-of-the-art (SOTA) method for a variety of simulated trajectories and experimental datasets of motile bacteria dispersing in porous media under a range of flow conditions.}, archive = {J_EAAI}, author = {Marc Berghouse and Rishi Parashar}, doi = {10.1016/j.engappai.2025.110761}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110761}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DeepTrackStat: An end-to-end deep learning framework for extraction of motion statistics from videos of particles}, volume = {153}, year = {2025}, }
• (2025). Visual geo-localization and attitude estimation using satellite imagery and topographical elevation for unmanned aerial vehicles. EAAI, 153, 110759. (www)

  @article{ , abstract = {In recent years, the autonomous flight capability of unmanned aerial vehicles (UAVs) has significantly improved. However, basic issues such as geo-localization and orientation still remain unsolved when Global Positioning System (GPS) is unavailable. Current visual geo-localization methods either assume a flat flight area using only satellite imagery for large-scale scenes or rely on three-dimensional (3D) reconstruction for small-scale areas, both of which limit practical UAVs applications. To achieve visual geo-localization in large-scale scenes with complex terrain variations, we propose a method that reliably achieves 6-degree-of-freedom pose estimation, including offline database and online inference. In the first stage, neural networks extract global and local features from satellite images to build feature databases. In the second stage, a fast search is conducted in the global feature database based on the global image descriptor of the aerial image, followed by fine matching based on local keypoint descriptors. Geographic coordinates for matched feature points are provided by satellite imagery and topographical elevation to achieve the UAV pose estimation. We conducted experimental validation by capturing aerial images at two non-overlapping locations. The results show that at Location 1, when using a top-down view, the overall localization error of the UAV decreased from 31.76 m to 17.94 m, improving accuracy by 43.61%; the recall rate increased from 67.59% to 71.94%. At Location 2, when using a front-down view, the overall localization error of the UAV decreased from 48.07 m to 10.84 m, with accuracy significantly improving by 77.45%; the recall rate increased from 64.96% to 69.71%.}, archive = {J_EAAI}, author = {Xiong Qiu and Shouyi Liao and Dongfang Yang and Yongfei Li and Shicheng Wang}, doi = {10.1016/j.engappai.2025.110759}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110759}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Visual geo-localization and attitude estimation using satellite imagery and topographical elevation for unmanned aerial vehicles}, volume = {153}, year = {2025}, }
• (2025). Statistical information assisted interaction reasoning for skeleton-only group activity recognition. EAAI, 153, 110756. (www)

  @article{ , abstract = {Group activity recognition is an essential task for video understanding and has shown significant value in practical application scenarios. Most existing methods mainly operate on visual information directly. Some skeleton-based methods replace the visual information with human skeletons to reduce computational pressure. However, due to the concise representation of skeletons, these methods typically leverage less information and show performance gaps with visual-based ones. In this paper, we introduce the statistical information of human joints and semantic information of action labels to provide more useful classification clues, and further propose a Statistical Information Assisted Interaction Reasoning Framework (SIAIR) for skeleton-only group activity recognition. It summarizes the joints statistical information to constructs the distribution maps, and explores semantic information of action labels to integrate both the skeleton-based individual representation and the statistical information. Extensive experiments on two public datasets validate the effectiveness of introducing statistical information, and the superiority of our framework over the state-of-the-art methods.}, archive = {J_EAAI}, author = {Zhuming Wang and Zun Li and Yihao Zheng and Lifang Wu}, doi = {10.1016/j.engappai.2025.110756}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110756}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Statistical information assisted interaction reasoning for skeleton-only group activity recognition}, volume = {153}, year = {2025}, }
• (2025). Neurodegenerative disorders: A holistic study of the explainable artificial intelligence applications. EAAI, 153, 110752. (www)

  @article{ , abstract = {Neuro Degenerative Disorders (NDDs) involve progressive nerve cell loss, impacting functions like sensation, movement, memory, and cognition, posing life-threatening risks. Despite extensive research, viable therapies remain elusive due to complex pathophysiology. Artificial Intelligence (AI), particularly Machine Learning (ML) and Deep Learning (DL), shows promise in NDD diagnosis and treatment by leveraging vast datasets for accurate predictions. However, because AI models are “black boxes,” explainable AI (XAI) had to be created to make sure that physicians and patients would trust and accept it. Early detection is critical to stop degeneration and make things better for patients. Many in-depth studies on XAI are designed explicitly for NDDs. Existing research does not constantly look at how to interpret NDDs, how to evaluate them, or how to keep them safe. This paper fills in these gaps by looking at and grouping XAI methods for different NDDs, to make them easier to understand and use in medical settings. In this paper, we look at the interpretability methods used in various NDD studies. The methods are split into five groups based on the conditions they are used to treat: Frontotemporal Dementia (FTD), Multiple Sclerosis (MS), Amyotrophic Lateral Sclerosis (ALS), and Alzheimer's Disease (AD). It organizes XAI methods into groups and talks about their pros, cons, and clinical importance. The study also finds some important research gaps. For example, it says that there are no good security frameworks and that XAI is hard to use in real-life healthcare settings. By giving helpful information and a plan for future research, this paper shows how XAI could change how NDDs are found, treated, and predicted. AI technologies will be used more in healthcare, and this will help us learn more about these challenging conditions.}, archive = {J_EAAI}, author = {Hongyuan Wang and Shiva Toumaj and Arash Heidari and Alireza Souri and Nima Jafari and Yiping Jiang}, doi = {10.1016/j.engappai.2025.110752}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110752}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neurodegenerative disorders: A holistic study of the explainable artificial intelligence applications}, volume = {153}, year = {2025}, }
• (2025). MirrorDiff: Prompt redescription for zero-shot grounded text-to-image generation with attention modulation. EAAI, 153, 110741. (www)

  @article{ , abstract = {Large-scale layout-conditioned text-to-image diffusion models have made significant progress and achieved remarkable results in generating diverse and high-quality images, realizing objects appearing in specific regions simultaneously. However, existing methods still fail with attribute coupling, unreasonable spatial relationships expressions and missing objects when the prompt is complex with multiple objects containing multiple attributes. In addition, it is difficult for users to give precise layout conditions for complex prompts. To address the above issues, we propose MirrorDiff, a novel training-free grounded text-to-image-to-text framework by redescription to correct inaccurate content expressions of synthetic images iteratively. Specifically, we first utilize large language models as layout generator which have the ability to understand visual concepts and support plausible arrangements to generate scene layout for complex prompts to help users obtain precision layout more conveniently. Subsequently, to solve small object missing, we design a layout-guided attention modulation strategy to properly adjust attention maps during diffusion generation process, which effectively increases attention of small objects. Additionally, semantic text regeneration supervision is proposed to constrain the redescription to keep consistent with the given text semantically, which aims to mitigate attribute coupling and failures of spatial relationships expressions. We conduct extensive experiments on four benchmarks and our method achieves the best results in all categories on the Holistic, Reliable and Scalable benchmark, which shows that our proposed MirrorDiff achieves state-of-the-art results both quantitatively and qualitatively compared with current superior models.}, archive = {J_EAAI}, author = {Chang Liu and Mingwen Shao and Zhengyi Gong and Xiang Lv and Lingzhuang Meng}, doi = {10.1016/j.engappai.2025.110741}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110741}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MirrorDiff: Prompt redescription for zero-shot grounded text-to-image generation with attention modulation}, volume = {153}, year = {2025}, }
• (2025). Comparison of unsupervised image anomaly detection models for sheet metal glue lines. EAAI, 153, 110740. (www)

  @article{ , abstract = {Accurate anomaly detection and localization in sheet metal glue line applications are crucial for quality assurance in automotive manufacturing. Most current vision-based inspection systems that rely on geometric deviations from a predefined shape often suffer from high false-positive rates, leading to unnecessary interventions and operational inefficiencies. This research investigates the potential of unsupervised deep learning models to significantly reduce false positives in the analysis of sheet metal glue line images, even with limited datasets. We conducted a comparative evaluation of 17 unsupervised deep learning models covering different categories with 28 backbones on datasets of approximately 300 industrial glue line images per part from a Swedish vehicle manufacturer. A data synthesis method was applied to balance the glue line dataset, further enhancing the reliability of the models. To address the challenge of limited training data and improve model generalization, we incorporated data augmentation techniques and performed robustness experiments to ensure applicability to real-world industrial conditions. Our findings demonstrate that deep learning approaches can effectively detect and localize anomalies, significantly reducing false positives and gluing machine downtimes compared to the existing system. Moreover, we proposed a multi-criteria decision-making based approach for model selection, enabling decision-makers to achieve optimal trade-offs between accuracy and inference time, thus improving operational efficiency. These advancements highlight that even with limited training data, unsupervised deep learning models can enhance anomaly detection reliability, streamline the automotive production process, and reduce unnecessary resource expenditures.}, archive = {J_EAAI}, author = {Siyuan Chen and Sunith Bandaru and Silvan Marti and Ebru Turanoglu Bekar and Anders Skoogh}, doi = {10.1016/j.engappai.2025.110740}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110740}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comparison of unsupervised image anomaly detection models for sheet metal glue lines}, volume = {153}, year = {2025}, }
• (2025). Continuous–Discrete alignment optimization for efficient differentiable neural architecture search. EAAI, 153, 110721. (www)

  @article{ , abstract = {Differential Architecture Search (DARTS) has become a prominent technique for neural architecture search in recent years. Despite its merits, the issue of discretization discrepancy within DARTS still necessitates further exploration, as it can degrade in performance. In this paper, we introduce a novel algorithm termed Continuous–Discrete Alignment Optimization (DARTS-CDAO), designed to address the discretization discrepancy and thereby enhance the robustness and generalization capabilities of the discovered neural architectures. Our proposed DARTS-CDAO algorithm seamlessly integrates the discretization process into the training phase of the architecture parameters, thereby bolstering the search algorithm’s adaptability to the inherent discretization processes. Specifically, our methodology commences by formalizing the process of architecture parameter discretization. Subsequently, we introduce a coarse gradient weighting algorithm that is employed to update the architecture parameters, effectively minimizing the divergence between the representation of continuous and discrete parameters. Rigorous theoretical analysis, coupled with extensive experimental outcomes, substantiates that our proposed approach can elevate the performance of the searched models. Notably, this enhancement is achieved without incurring additional search time, rendering DARTS more robust and endowed with a heightened capacity for generalization.}, archive = {J_EAAI}, author = {Wenbo Liu and Jia Wu and Tao Deng and Fei Yan}, doi = {10.1016/j.engappai.2025.110721}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110721}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Continuous–Discrete alignment optimization for efficient differentiable neural architecture search}, volume = {153}, year = {2025}, }
• (2025). Minimum-risk decision for the uncertain multiobjective cooperative task assignment problem of heterogeneous unmanned aerial vehicles. EAAI, 153, 110698. (www)

  @article{ , abstract = {The uncertain cooperative task assignment (CTA) problem based on uncertainty theory is a complex combinatorial optimization problem that cannot be effectively addressed using probability theory due to insufficient samples. In light of this limitation, this paper initially proposes an uncertain multiobjective CTA (UMCTA) problem, taking into account the uncertainties present in the battlefield environment. Consequently, the expected-value standard-deviation UMCTA model ( E σ -UMCTA) model, which emphasize the minimization of expected returns and stability of the objective functions, is developed. Nonetheless, this approach may yield suboptimal assignment schemes under predetermined risk levels. To address this issue, a minimum-risk model for the UMCTA problem is introduced by maximizing the belief degree that the objective functions do not exceed the predefined risk levels, and the concept of the minimum-risk efficient assignment scheme is delineated. Given the challenges in solving the minimum-risk model due to uncertain variables, two scenarios for addressing these uncertainties in the UMCTA problem are contemplated, and the minimum-risk model is transformed into a deterministic multiobjective programming problem. Subsequently, recognizing the combinational nature and complexity of the resulting deterministic multiobjective programming problem, an enhanced discrete particle swarm optimization (PSO) algorithm with non-dominated sorting is devised, where the crossover and mutation operators are employed to sustain diversity and avert premature convergence. Finally, a simulation study is conducted to substantiate the practicability of the proposed model and assess the efficacy of the designed algorithm.}, archive = {J_EAAI}, author = {Aoyu Zheng and Mingfa Zheng and Haitao Zhong and Zhichao Gao and Yu Mei}, doi = {10.1016/j.engappai.2025.110698}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110698}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Minimum-risk decision for the uncertain multiobjective cooperative task assignment problem of heterogeneous unmanned aerial vehicles}, volume = {153}, year = {2025}, }
• (2025). Max-one selection of equity prediction models for portfolio construction. EAAI, 153, 110694. (www)

  @article{ , abstract = {In many applications that involve prediction of a large number of variables, such as energy consumption or portfolio construction, it is common to employ a single model for predicting multiple target variables at once. Under this “monolithic” approach, the model is necessarily optimized on average across targets, and not for any one target in particular. Thus, the resulting predictions are likely to be underwhelming for a significant subset of targets. At the other end of the spectrum, optimizing and maintaining a separate prediction model for each of hundreds or thousands of targets, is unrealistic. This work proposes a novel model selection approach, termed max-one , that sits between these two extremes. Our methodology takes advantage of the standard hyperparameter tuning process used with most machine learning models and assigns to each target its own optimal model-hyperparameter combination, based on the criterion of one’s choice. In the context of portfolio optimization, which we use as a concrete example, the resulting family of models generates stock predictions which can then be used to construct a portfolio. Applying our suggested framework to an extensive 35-year data set with thousands of United States’ stocks, leads to an impressive 7x capital increase after transaction costs. This surpasses the performance of the “single model” approach and that of major global stock indices and compares favorably with recent relevant works, without sacrificing computational efficiency. Although our domain-specific application involves equities, our proposed approach could be used in settings where a model is tuned to predict multiple target variables.}, archive = {J_EAAI}, author = {Chariton Chalvatzis and Dimitrios Hristu-Varsakelis}, doi = {10.1016/j.engappai.2025.110694}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110694}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Max-one selection of equity prediction models for portfolio construction}, volume = {153}, year = {2025}, }
• (2025). The customer-based supplier selection and order allocation problem based on the waste management and resilience dimensions: A data-driven approach. EAAI, 153, 110692. (www)

  @article{ , abstract = {This work focuses on the Supplier Selection and Order Allocation (SSOA) problem with prominent features namely resiliency, circular economy, and customer-based dimensions. In this regard, this work proposes a novel data-driven approach based on the data mining and decision-making methods. In the current work, in the first step, the weights of the indicators are computed using the Fuzzy Best-Worst Method (FBWM). Then, the performance of the potential Raw Material Providers (RMPs) is assessed using the Decision Tree Regressor (DTR), and Data Envelopment Analysis (DEA) methods. In the next step, the best RMPs is identified and also the number of orders is specified by proposing a multi-objective mathematical model. In the next step, to deal with the mixed uncertainty, a hybrid data-driven method by combining the Robust Possibilistic-Stochastic Optimization (RPSO) and Prophet methods is proposed. Finally, to achieve the optimal solution, a new method called the Lexicographic Chebyshev Multi-Choice Goal Programming with Utility Function (LCMCGP-UF) is proposed. The achieved outputs demonstrate that cost, quality, waste management, service level, and robustness are determined as the most desirable indicators. The proposed model determines the best suppliers and also specifies the optimal location to establish the facilities. Additionally, the results confirm the efficiency and validity of the developed data-driven approach. Moreover, the results of the sensitivity analysis show that the total cost and non-resiliency of the supply chain have increased by increasing the demand parameter while the service level has decreased.}, archive = {J_EAAI}, author = {Borna Rezaie and Nikbakhsh Javadian and Mohammad Kazemi}, doi = {10.1016/j.engappai.2025.110692}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110692}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The customer-based supplier selection and order allocation problem based on the waste management and resilience dimensions: A data-driven approach}, volume = {153}, year = {2025}, }
• (2025). Polygon decomposition for obstacle representation in motion planning with model predictive control. EAAI, 153, 110690. (www)

  @article{ , abstract = {Model Predictive Control (MPC) is a powerful tool for planning the local trajectory of autonomous mobile robots. The paper considers a new algorithm for trajectory planning and obstacle avoidance based on the MPC technique known in Artificial Intelligence (AI) planning and robotics. We have proposed an original method for decomposing obstacles to form a potential field, which in turn is used as an additional component in MPC. Thus, we propose a new intelligent trajectory planning method that takes into account the special shape of obstacles, which in turn significantly improves the metrics of intelligent agent movement on the well-known Moving AI benchmark. The challenging aspect of MPC planning is collision avoidance on large and complicated grid maps. We propose the Polygon Segmentation for obtaining Artificial Potential Field (PolySAP). This local planner approximates the obstacles on the map with a set of polygons. We address the question of how to partition a map with polygons to make it fast and effective for a practical MPC planner. We propose a decomposition algorithm based on Straight Skeleton. Our algorithm returns a set of polygons, which are then convexified. Numerical experiments show that our method outperforms basic algorithms in performance and provides sufficient partition quality for effective planning. We propose an artificial potential function calculated for polygonal obstacles and added to the MPC objective for collision avoidance. We evaluate our approach on city map dataset and on a real robotic platform. Numerical experiments show that PolySAP allows for polygon decomposition that is five times faster than Interior Extensions. Our MPC solver provides a fast solution for the MPC task compared to the state-of-the-art MPC planners. Our planner ensured the safe motion of the real mobile robot through a narrow indoor environment. Our code is available at https://github.com/alhaddad-m/PolySAP .}, archive = {J_EAAI}, author = {Aleksey Logunov and Muhammad Alhaddad and Konstantin Mironov and Konstantin Yakovlev and Aleksandr Panov}, doi = {10.1016/j.engappai.2025.110690}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110690}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Polygon decomposition for obstacle representation in motion planning with model predictive control}, volume = {153}, year = {2025}, }
• (2025). Neuro-fuzzy tracking control of discrete-time nonlinear systems under linear matrix inequality region constraints. EAAI, 153, 110672. (www)

  @article{ , abstract = {This study investigates the problem of system identification and control for non-affine nonlinear deterministic systems using a generalised state-space neuro-fuzzy model. The proposed new model consists of a seven-layer neural network, where the consequent part comprises a finite set of discrete-time invariant linear state-space models. For tracking control design, quadratic stabilisers are integrated within a Parallel Distributed Compensation framework. Instead of merely ensuring the closed-loop stability of the state-augmented system, the feedback matrices are computed by solving a region-constrained Linear Matrix Inequality problem, which guarantees that the closed-loop eigenvalues remain within a D -stable region. The proposed generalised neuro-fuzzy model is proven to be a universal approximator on compact sets, with sufficient conditions for closed-loop stability established under the Neuro-Fuzzy-based Parallel Distributed Compensation framework. Experimental results on a nonlinear benchmark system validate the effectiveness and practical feasibility of the proposed neuro-fuzzy tracking control strategy.}, archive = {J_EAAI}, author = {Paulo Gil and Miguel João and Carolina Carvalho and Luís Brito Palma and Jorge Henriques}, doi = {10.1016/j.engappai.2025.110672}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110672}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neuro-fuzzy tracking control of discrete-time nonlinear systems under linear matrix inequality region constraints}, volume = {153}, year = {2025}, }
• (2025). A lightweight and robust detection network for diverse glass surface defects via scale- and shape-aware feature extraction. EAAI, 153, 110640. (www)

  @article{ , abstract = {As glass usage expands across industries, intelligent glass defect detection is essential for ensuring quality. However, the varying shapes and sizes of defects, coupled with numerous subtle defects and the demand for efficient detection, present challenges for existing methods in achieving both accurate and real-time detection. To address these, we propose the lightweight and robust G lass S urface D efect Net work (GSDNet) via scale- and shape-aware feature extraction. Specifically, the novel S hape-aware F eature E xtraction (SFE) block, which employs deformable convolution with special linear shape-adaptive offset constraints, forms the feature extraction network, enabling the adaptive extraction of local features for defects with irregular shapes. Meanwhile, the S cale- a ware (SA) attention is proposed, incorporating spatial attention mechanism to guide the model in focusing on key features across different receptive fields, enhancing defect detection at various scales. Finally, to enhance detection efficiency, the E fficient Bi directional P ath A ggregation N etwork (EBiPAN) is proposed as the feature aggregation module, integrating high-resolution information through bi-directional concatenation to improve small defect detection while avoiding significant additional computational burden. To validate the effectiveness of GSDNet, we compile the first multi-class glass defect dataset, covering 4 types of glass and 12 defect categories. Extensive experiments demonstrate GSDNet exhibits exceptional accuracy and robustness, consistently outperforming 9 advanced networks, with a 6.8% improvement in mean Average Precision and a notable 10.9% improvement in mean Average Precision small over the You Only Look Once version 8. Moreover, the optimal balance of accuracy and efficiency is achieved, with a detection speed of 68 frames per second. The dataset and code are publicly available at: https://github.com/FisherYuuri/GSDNet .}, archive = {J_EAAI}, author = {Huan Yu and Jin Wang and Jingru Yang and Yiming Liang and Zhihui Li and Zhan Wang and Haiyan He and Xusheng Zhang and Guodong Lu}, doi = {10.1016/j.engappai.2025.110640}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110640}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight and robust detection network for diverse glass surface defects via scale- and shape-aware feature extraction}, volume = {153}, year = {2025}, }
• (2025). Short-term wind speed prediction method based on prior wind direction knowledge and multi-period decoupling. EAAI, 153, 110596. (www)

  @article{ , abstract = {Accurate wind speed prediction improves power system stability and efficiency. Wind direction contains airflow dynamic information, but existing methods fail to fully capture it. The multi-periodicity of meteorological data causes irregular wind speed fluctuations, complicating the capture of correlations and variances across periods. To address these issues, this paper proposes the Multi-Period Wind Direction Graph Network(MWGN). Gaussian Spatial Module(GSM) calculates the relationship between wind turbines using Gaussian distribution along the crosswind direction, thus enhancing airflow motion capture. Multi-Period Time Series Decoupling Module(MPTSDM) selects the primary periods by analyzing the frequency domain and extracts local correlation and periodic correlation, to better model the changing pattern of time series. Compared with the existing models, MWGN achieves the consistent state-of-the-art in three datasets.}, archive = {J_EAAI}, author = {Zewen Shang and Xuewei Li and Zhiqiang Liu and Yingzhou Sun and Jian Yu and Mei Yu and Tianyi Xu and Wei Xiong}, doi = {10.1016/j.engappai.2025.110596}, journal = {Engineering Applications of Artificial Intelligence}, month = {8}, pages = {110596}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Short-term wind speed prediction method based on prior wind direction knowledge and multi-period decoupling}, volume = {153}, year = {2025}, }
• (2025). Assembling multipath service function chains in substrate graphs using sharing instances and deep learning. EAAI, 152, 110900. (www)

  @article{ , abstract = {The assembly of Service Function Chains (SFCs) within substrate graphs has emerged as a crucial challenge in the context of network function virtualization. SFCs are ordered sequences of Virtual Network Functions (VNFs) that must be executed to meet specific service requirements. Efficient management of network resources in this domain can be achieved through the reuse of sharable VNF instances, which reduces resource consumption and operational overhead. As network environments evolve with increasing demands and dynamic complexities, multipath SFCs have gained attention as a promising approach to enhance fault tolerance and optimize resource utilization. Multipath SFCs allow traffic to traverse multiple distinct paths in the network, ensuring load balancing, redundancy, and improved performance. With this motivation, this study presents a multipath SFC assembly framework via VNF instance sharing in substrate graphs, implemented using deep learning algorithms. The proposed algorithm leverages the Soft Actor-Critic algorithm to embed multipath SFCs effectively. By dynamically mapping SFCs to paths with minimal bandwidth consumption and latency, our algorithm optimizes network resource usage. The resource allocation mechanism in proposed algorithm is designed to adapt flexibly to diverse request demands. Moreover, by utilizing reusable VNF instances, our algorithm further enhances resource efficiency. Also, the framework ensures that all nodes and links in the physical network satisfy resource constraints during the mapping process. Simulation results demonstrate that proposed algorithm significantly improves performance metrics under multipath SFC assembly with instance sharing in substrate graphs. Specifically, proposed algorithm outperforms state-of-the-art methods by improving bandwidth consumption by 1.04 %, reducing latency by 0.82ms, and increasing the SFC assembly rate by 2.75 %, demonstrating efficient network resource management.}, archive = {J_EAAI}, author = {Zhanwei Chen and Amin Rezaeipanah}, doi = {10.1016/j.engappai.2025.110900}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110900}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Assembling multipath service function chains in substrate graphs using sharing instances and deep learning}, volume = {152}, year = {2025}, }
• (2025). Corrigendum to “Advanced multi-layer deep learning model for accurate estimation of heat transfer and flow designing parameters across diverse dataset configurations” [Eng. appl. artif. intell. 151 (2025) 110723]. EAAI, 152, 110869. (www)

  @article{ , archive = {J_EAAI}, author = {Ali A.H. Karah Bash and Ahmad Aboul Khail}, doi = {10.1016/j.engappai.2025.110869}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110869}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Corrigendum to “Advanced multi-layer deep learning model for accurate estimation of heat transfer and flow designing parameters across diverse dataset configurations” [Eng. appl. artif. intell. 151 (2025) 110723]}, volume = {152}, year = {2025}, }
• (2025). A coupling deterministic and probabilistic wind energy prediction based on information leakage prevention and distinctive deep learning network. EAAI, 152, 110862. (www)

  @article{ , abstract = {Accurate wind speed prediction is crucial for efficient wind energy utilization. Decomposition-integration methods, though widely used in wind speed prediction, are often limited by neglected information leakage during data processing and variability in prediction results among different decomposition models. To overcome these limitations, the present research proposes an enhanced framework for forecasting wind speed. The framework first extracts effective features from raw wind speed data using an improved singular spectrum analysis (ISSA) method. Then, a time convolutional network (TCN) is utilized to map signals and establish a relational paradigm, effectively addressing the information leakage issue. Furthermore, a deep learning network with multiple network interfaces is constructed, incorporating principles of quantitative analysis and parallel processing. Finally, quantile regression interval prediction is applied to assess the uncertainty of predicted values, enhancing model reliability. Validation using wind speed data from two Liaoning wind farms and one Zhoushan wind farm shows mean absolute percentage error values of 1.29 %, 1.53 %, and 13.70 % at the three sites, respectively. These results demonstrate a significant enhancement in the accuracy and efficacy of the proposed model for predicting wind speeds.}, archive = {J_EAAI}, author = {Jujie Wang and Yafen Liu and Shuqin Shu and Xuecheng He}, doi = {10.1016/j.engappai.2025.110862}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110862}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A coupling deterministic and probabilistic wind energy prediction based on information leakage prevention and distinctive deep learning network}, volume = {152}, year = {2025}, }
• (2025). An adaptive graph neural network-based intrusion detection system for airborne network. EAAI, 152, 110851. (www)

  @article{ , abstract = {With the rapid development of airborne information technology, the security of airborne networks has received more attention. To provide effective protection for airborne networks, we propose an adaptive graph neural network intrusion detection system; containing a graph generation process and a graph reconstruction process. Firstly, there is inherent uncertainty in the classification boundary for imbalanced datasets in airborne networks. We propose a Weight-based k-nearest neighbor to assign differential weights, thereby constructing the adjacent weight matrix. Then, we introduce a Gaussian graph aggregation classification method to improve the accuracy of the initial graph construction. In this process, the data with connected relationships are mapped into a low-dimensional Gaussian distribution through aggregation. We sample from the Gaussian distribution for each node data to establish new connections through the latent variables for anomaly intrusion detection. Finally, the proposed method was tested on public network intrusion detection and aviation data bus intrusion detection to verify its effectiveness. The experimental results demonstrate that the proposed method improves model classification performance by using an adaptive graph neural network that repeatedly aggregates features from similar nodes, leading to more accurate detection results.}, archive = {J_EAAI}, author = {Wenqi Liu and Shijia Li and Cong Gao and Yiqin Sang and Hongjuan Ge}, doi = {10.1016/j.engappai.2025.110851}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110851}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive graph neural network-based intrusion detection system for airborne network}, volume = {152}, year = {2025}, }
• (2025). Corrigendum to “Interpretable spatio-temporal prediction using deep neural network - Local interpretable model-agnostic explanations: A case study on leptospirosis outbreaks in malaysia” [Engineering applications of artificial intelligence 151 (2025) 110665]. EAAI, 152, 110849. (www)

  @article{ , archive = {J_EAAI}, author = {Fariq Rahmat and Zed Zulkafli and Asnor Juraiza Ishak and Ribhan Zafira Abdul Rahman and Wardah Tahir and Jamalludin Ab Rahman and Veianthan Jayaramu and Simon De Stercke and Salwa Ibrahim and Muhamad Ismail}, doi = {10.1016/j.engappai.2025.110849}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110849}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Corrigendum to “Interpretable spatio-temporal prediction using deep neural network - Local interpretable model-agnostic explanations: A case study on leptospirosis outbreaks in malaysia” [Engineering applications of artificial intelligence 151 (2025) 110665]}, volume = {152}, year = {2025}, }
• (2025). From smoke to fire: A forest fire early warning and risk assessment model fusing multimodal data. EAAI, 152, 110848. (www)

  @article{ , abstract = {Experienced forest firefighters can integrate knowledge of smoke patterns and risk factors to assess fire risk and anticipate potential fire spread and outbreaks in complex, variable environments. This study simulates the previously mentioned monitoring process and presents the Multimodal Smoke Risk Estimation Network (MM-SRENet), an innovative multimodal fusion model. This model uniquely integrates the Multi-Scale Dilation Block and Repulsion Loss into a lightweight and efficient target detection system to accurately identify smoke's presence. Furthermore, an advanced backbone based on star operation extracts the scene characteristics associated with smoke and merges them with various fire risk factors. The objective is to simulate fire risk assessments in smoke scenarios and to reduce the misallocation of fire resources resulting from false alarms and missed alerts. The proposed model was trained and validated on a multimodal dataset comprising multiple backgrounds. It successfully identified smoke features and fire potential risks in different scenarios, achieving a prediction accuracy of 93.06 %. Fusing smoke images with fire risk data resulted in an 18.75 % improvement in recognition accuracy compared to the single modal model. This work bridges the gap between multimodal data fusion and forest fire risk monitoring, providing a new direction for future intelligent forest fire prevention and control practices.}, archive = {J_EAAI}, author = {Peixian Jin and Pengle Cheng and Xiaodong Liu and Ying Huang}, doi = {10.1016/j.engappai.2025.110848}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110848}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {From smoke to fire: A forest fire early warning and risk assessment model fusing multimodal data}, volume = {152}, year = {2025}, }
• (2025). An extended robust optimisation approach for sustainable and resilient supply chain network design: A case of perishable products. EAAI, 152, 110846. (www)

  @article{ , abstract = {Disruptions and uncertainties pose considerable challenges in supply chain (SC) operations, requiring robust, strategic approaches. This study addresses financial resource allocation between supply chain network (SCN) design and external market investments, explicitly considering budget allocation and sales representative selection for perishable products. We propose a novel axis-shift robust (AxShR) method combined with an enhanced p-robust optimisation framework to manage multiple complex uncertainties. The model introduces a new resilience strategy allowing retailers to transfer products within the SCN, and apply artificial intelligence (AI) via a neural network to predict disruptions by evaluating risk indicators such as supplier location risk, raw material availability, transportation infrastructure reliability, energy supply stability, and labour availability. Multi-cut Benders decomposition enhances computational efficiency, while goal programming accommodates multiple conflicting objectives (profitability, environmental sustainability, and job creation), with decision-makers assigning weights reflecting managerial and policy priorities. Comparisons with classical approaches (including two-stage stochastic programming and the Bertsimas–Sim robust method) demonstrate significantly improved performance, as sensitivity analyses indicate that neglecting resilience increases expected and worst-case costs by 10.21 % and 16.54 %, respectively. Higher disruption levels demand more extensive, combined resilience strategies rather than reliance on a single measure. Additionally, optimal pricing decisions balance profitability with demand fulfilment, guiding policymakers to reallocate resources between SCN design and external investments when gross profit margins shift. The proposed robust approach outperforms classical methods in managing uncertainty, and employing multi-cut Benders decomposition yields faster computational performance than commercial solvers and traditional methods, strengthening both the model's applicability and overall operational resilience.}, archive = {J_EAAI}, author = {Seyed Ashkan Hosseini Shekarabi and Reza Kiani Mavi and Flavio Romero Macau}, doi = {10.1016/j.engappai.2025.110846}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110846}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An extended robust optimisation approach for sustainable and resilient supply chain network design: A case of perishable products}, volume = {152}, year = {2025}, }
• (2025). A domain generalization method for deploying driver distraction detection models to practical application scenarios. EAAI, 152, 110844. (www)

  @article{ , abstract = {Driver distraction detection is crucial for reducing traffic accidents and enhancing driving safety. However, there is often a domain shift between data from real-world driving scenarios and the model training data, which limits the deployment of existing driver distraction detection models. To facilitate deployment of these models in practical application scenarios, a domain generalization method based on the contrastive language-image pretraining (CLIP) model and the constraint of center loss (DGCCL) is proposed in this paper. Firstly, the image encoder of the pre-trained CLIP model is adopted as the feature extraction module to improve the domain generalization ability of the proposed model. Furthermore, the constraint of center loss is introduced to promote the samples of different datasets to follow an approximately identical distribution in the feature space, thereby alleviating the domain shift problem. Additionally, the classification loss with additive angular margin penalty (AAMP) is introduced to further improve the cross-domain performance of the proposed model. In order to demonstrate the effectiveness of the proposed method, extensive experiments have been conducted on three publicly available driver distraction detection datasets: AUC-DDD, State-Farm, and SAM-DD. The experimental results verify that our method can achieve much better performance than various well-known models in the cross-domain driver distraction detection tasks.}, archive = {J_EAAI}, author = {Lie Yang and Henglai Wei and Zhongxu Hu and Chen Lv}, doi = {10.1016/j.engappai.2025.110844}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110844}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A domain generalization method for deploying driver distraction detection models to practical application scenarios}, volume = {152}, year = {2025}, }
• (2025). A mathematical model and a memetic algorithm for berth allocation at inland waterway terminals in developing countries. EAAI, 152, 110843. (www)

  @article{ , abstract = {This paper addresses the Berth Allocation Problem (BAP) in multipurpose Inland Water Transport (IWT) terminals operated in developing countries like India. Conceptually, these are different from the specialized IWT ports operated in developed countries. The IWT terminals in developing countries face issues due to complex combination of cargo types, their mix on barges, variability in handling equipment, and direct transloading in trucks. This paper adds to the research on BAP which otherwise has primarily focused on specialized seaports or IWT ports in developed countries. We developed a mathematical model to deal with BAP in multipurpose IWT terminals which minimizes the waiting and total completion times of the barges. We further propose a novel memetic algorithm (MA), an artificial intelligence (AI) based optimization technique that integrates a genetic algorithm with a local search to obtain near-optimal solutions efficiently. Using real-world data from an Indian IWT terminal, extensive computational experiments are conducted to validate the models' effectiveness and the algorithm's efficiency. A sensitivity analysis is carried out to deliver managerial insights. The developed AI-driven method can replace the first-come-first-served (FCFS) approach that is presently in use at IWT terminals in developing nations like India. It would improve their operational efficiency and competitiveness.}, archive = {J_EAAI}, author = {Adnan Pasha and Jiyin Liu and Rajat Rastogi}, doi = {10.1016/j.engappai.2025.110843}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110843}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A mathematical model and a memetic algorithm for berth allocation at inland waterway terminals in developing countries}, volume = {152}, year = {2025}, }
• (2025). A module partition method for complex product based on the knowledge hypergraph. EAAI, 152, 110842. (www)

  @article{ , abstract = {The modular design can effectively control the development cycle and cost of complex product, with product module partition (PMP) serving as the foundation of modularization. However, models constructed based on expert knowledge are inadequate in effectively capturing the relationships within complex product, which undermines the efficiency and accuracy of PMP. To solve this problem, this paper introduces hypergraph theory into the field of PMP, specifically, proposes a PMP method based on the knowledge hypergraph (KHG). First, the multiple coupling relationships between complex product are defined from the perspective of "Function-Behavior-Structure-Constraint", to form the pattern layer of the KHG. Then, the joint learning algorithm, which contains the pretraining model, Bi-directional Long Short-Term Memory network, Conditional Random Field and Attention layer, is proposed to automatically extract design knowledge from large-scale text data to form the data layer of the KHG. Furthermore, considering that the PMP model needs to learn nonlinear relationship features, achieve end-to-end optimization, and have strong anti-noise ability, hypergraph neural networks are used to partition the complex product modules, which contains the importance calculation, hypergraph convolution, modularity maximum and self-supervised module. Finally, a case study is conducted using a snow removal equipment as an example, the knowledge extraction accuracy reaches 91.67 %, and the PMP modularity is 0.68, thus validating the feasibility of the proposed method. Additionally, the comparison is made with other knowledge extraction and hypergraph clustering algorithms using public datasets, which further confirms the feasibility and superiority of the proposed method.}, archive = {J_EAAI}, author = {Pengchao Wang and Jianjie Chu and Suihuai Yu and Fangmin Cheng and Ning Ding and Yangfan Cong}, doi = {10.1016/j.engappai.2025.110842}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110842}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A module partition method for complex product based on the knowledge hypergraph}, volume = {152}, year = {2025}, }
• (2025). Unsupervised rapid lowlight enhancement via deep curve and statistic loss. EAAI, 152, 110841. (www)

  @article{ , abstract = {Lowlight images suffer from poor illumination and noise due to the limited information captured by smaller sensor devices such as smartphone cameras. While supervised approaches to lowlight image enhancement have shown promise, they require paired image datasets, which are often expensive and difficult to obtain, limiting their practical applicability. Previous unsupervised network approaches have attempted to address these challenges but often fall short in terms of quality or speed. To overcome these limitations, we present an unsupervised network specifically designed for lowlight image enhancement. Our method employs diverse strategies within the loss functions to guide the model in generating images with normal lighting that appear natural to the human eye. This approach, combining rapid processing, a lightweight model, and decent-quality outputs trained with unpaired data, makes it an ideal choice for real-world applications such as consumer electronics which is helpful for various kinds of engineering. Furthermore, to address the common issue of noise amplification in enhanced images, we incorporate a denoising model trained also with unpaired data, which can effectively remove the noises. Our quantitative comparisons demonstrate that our approach achieves superior and comprehensive scores while maintaining a low number of trainable parameters, around 10k. Additionally, our model processes a 512 × 512 color image in just 43 ms, which highlights its efficiency. Using the LOLv2-real (LOw-Light real-world version 2) dataset, it achieved a PSNR (Peak Signal-to-Noise Ratio) of 20.23 dB, which is 1.27 dB higher than the second-best method, a LPIPS (Learned Perceptual Image Patch Similarity) of 0.168, and an SSIM (Structure SIMilarity) of 0.77, demonstrating its effectiveness.}, archive = {J_EAAI}, author = {Min Si Young and Chang Hong Lin}, doi = {10.1016/j.engappai.2025.110841}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110841}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised rapid lowlight enhancement via deep curve and statistic loss}, volume = {152}, year = {2025}, }
• (2025). Identification and selection of systems security engineering approach via bipolar fuzzy multi-criteria decision-making technique based on generalized power operators. EAAI, 152, 110825. (www)

  @article{ , abstract = {Organizations need to choose an appropriate systems security engineering approach to enable them to design and create systems that are secure and resistant to threats in areas where the impacts of security threats are severe. However, this decision-making process may inherently contain bipolarity and vagueness because the criteria used to evaluate systems security engineering approaches may have positive and negative aspects (characteristics). Overlooking these bipolar aspects may result in inadequate or skewed assessments of the risks, which in turn will hamper the ability to address them appropriately. In response to this challenge, this research develops a new methodological framework that combines bipolar fuzzy set theory and generalized power aggregation operators into multi-criteria decision-making for choosing systems security engineering approaches. This new mathematical framework would be known as the bipolar fuzzy multi-criteria decision-making technique and is applicable to solve all sorts of multi-criteria decision-making dilemmas, where bipolarity is involved such as selection of systems security engineering, selection of artificial intelligence models, etc. Thus, further, this manuscript analyzes and solves a case study of the identification and selection of the finest systems security engineering approach through a bipolar fuzzy multi-criteria decision-making technique. To reveal the benefits and supremacy of the initiated technique over the existing ones, this manuscript contains a comparative analysis.}, archive = {J_EAAI}, author = {Xiaopeng Yang and Ubaid ur Rehman and Kholood Alnefaie and Tahir Mahmood}, doi = {10.1016/j.engappai.2025.110825}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110825}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identification and selection of systems security engineering approach via bipolar fuzzy multi-criteria decision-making technique based on generalized power operators}, volume = {152}, year = {2025}, }
• (2025). A novel wasserstein autoencoder-enhanced thermo-mechanical coupled reduced-order model for high pressure turbine blades life monitoring. EAAI, 152, 110819. (www)

  @article{ , abstract = {To facilitate effective assessment of loads (e.g., temperature and stress) and life management of high-pressure turbine (HPT) blades, a wasserstein autoencoder (WAE)-enhanced thermodynamically coupled reduced-order model (ROM) is proposed in this paper. The advanced ROM for the nonlinear thermomechanical coupling fields is developed by introducing the deep learning model of WAE in the proper orthogonal decomposition (POD) method. The proposed method improves the prediction accuracy of loads in locally focused regions and generalization performance. The accuracy and efficiency of this method are validated through 30 sets of validation conditions. Results indicate that the proposed approach achieves higher accuracy and better generalization performance than traditional POD-based methods, with errors maintained within 10. Additionally, computational speed is improved by nearly 1400 times compared to conventional numerical methods. The WAE-enhanced ROM is applied for load and life assessment of the HPT blades throughout their service life. The evaluation time for a single aeroengine performance parameter is 1.7 s, and for a single flight evaluation, it is 67 s, which highlights the effectiveness of the proposed method in enabling the assessment of the loads and remaining life of HPT blades.}, archive = {J_EAAI}, author = {Rongqiao Wang and Ruoqi Chen and Yan Zhao and Tianbao Shen and Gaoxiang Chen and Dianyin Hu and Zhimin Jiang and Xuemin Wang}, doi = {10.1016/j.engappai.2025.110819}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110819}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel wasserstein autoencoder-enhanced thermo-mechanical coupled reduced-order model for high pressure turbine blades life monitoring}, volume = {152}, year = {2025}, }
• (2025). Reduced mobility of elderly travelers in airports: Artificial neural networks approach. EAAI, 152, 110818. (www)

  @article{ , abstract = {Internationally, the number of elderly people, especially those over 65, continues to increase. The improved living standards extend the human life expectancy and improve quality of life, which results in an increase in the number of elderly travelers utilizing airports. There are many factors that make commuting in the airport difficult for the elderly such as deteriorating sight, hearing as well as psychological challenges of anxiety such as running late for a flight. This paper tackled the biggest issues faced by elderly people and people with reduced mobility in Kuwait International airport. The aim of this study was achieved through constructing a survey based on factors found such as wayfinding, fatigue, technology and equipment, and amenities. Artificial Neural Networks (ANN) were utilized for exploring the correlation and ranking of different factors from our survey conducted at terminal 1 and 4 of Kuwait International Airport. One of the study's findings is the identification of hidden correlations between terminal facilities and travelers' personal information at Kuwait International Airport. These findings are significant for practical applications, as they can inform improvements in airport operations and service efficiency, with potential for broader real-world use in the future.}, archive = {J_EAAI}, author = {Sharaf AlKheder and Fatma Al-Hajri and Farah Buarki}, doi = {10.1016/j.engappai.2025.110818}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110818}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reduced mobility of elderly travelers in airports: Artificial neural networks approach}, volume = {152}, year = {2025}, }
• (2025). Cross-scale prediction of aluminum dust concentration based on image fusion physics-informed neural networks. EAAI, 152, 110817. (www)

  @article{ , abstract = {Research on predicting dust concentration can effectively help reduce the occurrence of dust explosion accidents. However, existing methods struggle to accurately and quickly reconstruct and predict concentration fields across scales in turbulent dust diffusion processes. This paper proposes a neural network framework that combines particle physics information with image inverse mapping to achieve rapid, multi-source, and cross-scale prediction of turbulent dust concentration fields. We conducted a 280 kg aluminum dust dispersion experiment, collecting ultrasound attenuation signals and image data for our dataset. Based on this, by incorporating the Maxwell-Stefan equation, our approach addresses the underfitting issues of existing neural networks in predicting microscale turbulence within concentration fields. Additionally, the inverse mapping of images provides macroscopic diffusion trends for the concentration field. Results demonstrate that our method reconstructs aluminum dust concentration and predicts future 0.06 s states in 0.011 s, with a mean squared error of only 0.0003. Compared to existing Convolutional Neural Networks, Physics-Informed Neural Networks, and Computational Fluid Dynamics methods, our approach shows significant improvement in cross-scale prediction, making accurate concentration prediction possible. This advancement offers quantitative prediction data crucial for preventing dust explosions.}, archive = {J_EAAI}, author = {Nanxi Ding and Wenzhong Lou and Zihao Zhang and Yizhe Wu and Chenglong Li and Wenlong Ma and Zhengqian Zhang}, doi = {10.1016/j.engappai.2025.110817}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110817}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-scale prediction of aluminum dust concentration based on image fusion physics-informed neural networks}, volume = {152}, year = {2025}, }
• (2025). Multi-stage self-training social bot detection based on graph neural network. EAAI, 152, 110816. (www)

  @article{ , abstract = {With the popularity of social networks, social bots are increasingly interfering with human social activities, which have negatively impacted social network ecology and user experience. How to detect social bots effectively has become a challenging task. Most of existing social bot detection methods are based on the supervised learning mechanism, but the labelled data collection is resource-intensive. In this paper, a Multi-stage Self-training Social Bot detection method based on a Graph Neural Network (MSSBot) is proposed to improve the performance of social bot detection with a small amount of labelled data. MSSBot exploits pseudo-labels to expand the dataset and trains a Relational Graph Convolutional Network in multiple stages, while uncertainty-aware and deep clustering are used to reduce the noise present in the pseudo-labels acquisition process. The experiments were conducted on a real-world social network dataset Twibot-20, a small high-quality dataset focused on social bot behavior patterns Cresci-2015, and a large-scale graph-based dataset Twibot-22 with high annotation quality. The experimental results show that our method achieves accuracies of 0.9259, 0.9979, and 0.8358 on the Twibot-20, Cresci-2015, and Twibot-22 datasets, respectively, outperforming the state-of-the-art methods, with improvements of 5.68 %, 2.64 %, and 3.92 % over the existing state-of-the-art methods.}, archive = {J_EAAI}, author = {Keke Wang and Xiujuan Wang and Kangmiao Chen and Zhengxiang Wang and Kangfeng Zheng}, doi = {10.1016/j.engappai.2025.110816}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110816}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-stage self-training social bot detection based on graph neural network}, volume = {152}, year = {2025}, }
• (2025). Predictive modeling of combustion cycle variations in spark ignition engine based on backpropagation neural network and artificial bee colony algorithm. EAAI, 152, 110813. (www)

  @article{ , abstract = {This study proposes a back-propagation neural network (BPNN) method optimized by the artificial bee colony (ABC) algorithm for predicting cyclic variations in an ignition engine. The proposed method integrates a meta-heuristic (MH) algorithm with deep learning techniques to develop a high-precision prediction model. This model utilizes engine speed, exhaust gas recirculation (EGR) ratio, and excess air ratio (λ) as input parameters, with indicated mean effective pressure (IMEP) as the output parameter. A multi-stage optimization strategy is employed in this study. Initially, the optimal structural parameters of the BPNN are determined through experimental design, including 90 and 84 neurons in the two hidden layers, the Sigmoid activation function, and a learning rate of 0.01. Subsequently, the ABC algorithm is utilized to globally optimize the initial weights and biases of the BPNN, thereby mitigating the tendency of the BPNN to converge to local optima. Experimental results demonstrate that the optimized ABC-BPNN model exhibits exceptional prediction performance, achieving a coefficient of determination (R 2 ) of 0.9407, a root mean square error (RMSE) of 0.3673, and a mean absolute error (MAE) of 0.2986. Furthermore, the ABC-BPNN model exhibits superior prediction capability for IMEP when compared to other advanced algorithms. Lastly, analysis of the experimental results shows that an increase in the excess air ratio and EGR ratio aggravates combustion instability. Consequently, the novelty of this study is to improve the accuracy and reliability of engine combustion process prediction and to provide a new path for performance optimization of complex industrial systems.}, archive = {J_EAAI}, author = {Mingzhang Pan and Yuede Pan and Changcheng Fu and Xinxin Cao and Song Zhang and Jun Yang and Hailang Sang and Wei Guan}, doi = {10.1016/j.engappai.2025.110813}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110813}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predictive modeling of combustion cycle variations in spark ignition engine based on backpropagation neural network and artificial bee colony algorithm}, volume = {152}, year = {2025}, }
• (2025). A bayesian-physical informed conditional tabular generative adversarial network framework for low-carbon concrete data augmentation and hyperparameter optimization. EAAI, 152, 110811. (www)

  @article{ , abstract = {Data shortage, unbalanced data distribution and multi-factor coupling mechanism of materials all increase the difficulty of design. This paper proposed a physical constraint-conditional generative adversarial network (PI-CTGAN) to solve the above problems. Firstly, residual layers are added to the generator to enhance model stability. The continuous differentiable function and wasserstein_distance were constructed to embed physical loss functions into the generator, including water-cement ratio, supplementary cementitious materials (SCMs) ratio, and aggregate water absorption ratio. Based on this, Bayesian optimization (BO) was used to optimize the hyperparameters of PI-CTGAN. The results showed that BO effectively optimized the model's hyperparameters, reducing the total of Kolmogorov-Smirnov distribution (K-S tot ) of the generated dataset by 27.9 %. Additionally, applying physical loss to the optimized model can improve the model's data recognition capability, with generation accuracy increasing by 16.2 %. The influence of physical weight (W PI ) and activation functions on data generation quality was compared. Revealing that K-S tot initially decreased and then increased with W PI . The model using the rectified linear unit exhibited the best generation accuracy, with a K-S tot of 0.37 and anomaly data ratios (water-binder ratio and supplementary cementitious materials/binder ratio) of 11.67 % and 3.2 %, respectively. The generated data and the experimental data show statistical similarity and conform to the physical law. By constructing the target dataset and related physical constraints, the proposed PI-CTGAN can effectively solve the issues of multi-source data sets with data shortage and imbalance, thereby providing numerous datasets to guide engineering design.}, archive = {J_EAAI}, author = {Shiqi Wang and Peng Xia and Fuyuan Gong and Yuxi Zhao and Peng Lin}, doi = {10.1016/j.engappai.2025.110811}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110811}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A bayesian-physical informed conditional tabular generative adversarial network framework for low-carbon concrete data augmentation and hyperparameter optimization}, volume = {152}, year = {2025}, }
• (2025). Electroencephalogram-based floor vibration serviceability evaluation using convolutional neural network and ensemble learning. EAAI, 152, 110806. (www)

  @article{ , abstract = {Discomfort or even panic would occasionally be felt from human-induced vibrations in public buildings such as shopping malls and offices, raising concerns on serviceability. However, the evaluation of human perception is rather subjective. In this study, the electroencephalogram (EEG) was utilized to directly reflect the human perception to floor vibrations. EEG signals and the corresponding perception factors (PFs) were recorded from 12 volunteers experiencing the floor vibrations induced from random heel drop, jump, and hammering. Each EEG time-domain curve was processed with a continuous wavelet transform to combine both time and frequency domain features into one wavelet time-frequency (WTF) image for further analysis. To identify different PFs through WTF images, three convolutional neural networks (CNNs) (Visual geometry group 16-layer network, Google Lenet, and Residual Network 18) and two ensemble learning methods (bagging and stacking) were tested. Based on the comparison of algorithm performances, Google Lenet is recommended to perform the PF recognition as it demonstrates an 89.4 % accuracy with a running time of 84 s. This study proves the feasibility of using the neuroscientific technique EEG with its time-frequency features to quantitatively grade human perceptions to floor vibration. Furthermore, the correlation establishment between EEG data and perception factors provides a novel approach for evaluating floor vibration serviceability.}, archive = {J_EAAI}, author = {Jiang Li and Xuhong Zhou and Jiepeng Liu and Weizhao Tang and Mingyue Xiao and Y. Frank Chen}, doi = {10.1016/j.engappai.2025.110806}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110806}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Electroencephalogram-based floor vibration serviceability evaluation using convolutional neural network and ensemble learning}, volume = {152}, year = {2025}, }
• (2025). Enhancing crop disease recognition via prompt learning-based progressive mixup and contrastive language-image pre-training dynamic calibration. EAAI, 152, 110805. (www)

  @article{ , abstract = {Crop diseases cause approximately 40 % of global crop losses annually and severely affecting crop yields. Traditional deep learning-based crop disease recognition methods rely on large-scale labeled data and demonstrate relatively weak generalization capabilities. Inspired by vision-language large-scale pre-trained models (VLMs) like Contrastive Language-Image Pre-training (CLIP), which combined with prompt learning techniques have demonstrated tremendous potential across multiple downstream tasks, we propose Progressive Mixup Prompt Learning (PMPL) for agricultural disease recognition. In the feature extraction phase, PMPL performs feature Mixup according to the hierarchical depth of the visual backbone network and subsequently uses the fused feature information for prompt tuning. To further enhance the model's generalization capability, we designed a test time adaptation (TTA) module called CLIP Dynamic Calibration (CDC). This module can dynamically calibrate prompts during testing, thereby achieving efficient unsupervised domain adaptation. Experimental results on eight publicly available datasets demonstrate that PMPL achieves average accuracy of 88.77 % (base) and 45.32 % (new), with a Harmonic mean (H) of 59.89 % which significantly outperforms baseline methods. In Cross-dataset transfer experiments, CDC improved the model's domain generalization capability by approximately 6 % on average. This indicates that our method has significant advantages over existing prompt learning methods in terms of accuracy and generalization for crop disease recognition, which has important practical significance for improving agricultural production efficiency and reducing disease-related losses.}, archive = {J_EAAI}, author = {Hao Chen and Haidong Li and Jinling Zhao and Chao Ruan and Linsheng Huang}, doi = {10.1016/j.engappai.2025.110805}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110805}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing crop disease recognition via prompt learning-based progressive mixup and contrastive language-image pre-training dynamic calibration}, volume = {152}, year = {2025}, }
• (2025). Utility of machine learning algorithms to classify age groups based on gait pattern: A biomechanical calendar. EAAI, 152, 110804. (www)

  @article{ , abstract = {The Biomechanical Calendar of Biomechanical Age serves as a crucial framework for understanding the impact of aging on human gait, providing valuable insights into health and performance assessment. To achieve this, this research implements a machine learning framework to classify age cohorts using comprehensive biomechanical kinetic and kinematic datasets. In this light, a dataset comprising three-dimensional motion capture and force plate data from 84 subjects across three age groups has been recorded and analyzed, focusing on features like joint angles and forces in the hip, knee, and ankle. Using Support Vector Machine (SVM) methodology with weighted loss functions to address class imbalance and Grid Search for hyperparameter optimization, the study achieved accuracies of 94 %, 100 %, and 98 % for the overall dataset and moment/angular classifications, respectively. Additionally, a comparative analysis of various algorithms, including SVM with different kernels, K-Nearest Neighbors (KNN), Decision Trees (DT), and Naïve Bayes (NB), was conducted. The findings indicate that machine learning models can effectively classify age groups based on biomechanical data, offering a crucial tool for monitoring age-related changes and identifying health and performance related factors. This study paves the way for future research on gait-based age classification and its applications in healthcare and sports medicine.}, archive = {J_EAAI}, author = {Rozhin Molavian and Behnam Faraji and Ali Fatahi and Hamed Abbasi and Davood Khezri}, doi = {10.1016/j.engappai.2025.110804}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110804}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Utility of machine learning algorithms to classify age groups based on gait pattern: A biomechanical calendar}, volume = {152}, year = {2025}, }
• (2025). Multi-dimensional real-time data-driven identification for progressive fracturing characteristics of sandstone under freeze-thaw cycles and seepage-stress coupling. EAAI, 152, 110803. (www)

  @article{ , abstract = {The melting of ice and overlying snow creates seepage fields in freeze-thaw (F-T) rock masses. Coupled with mining-induced stress changes, this elevates geological hazard risks. This study analyzes the progressive fracturing characteristics of sandstone under F-T cycles and seepage-stress coupling. The Mel-frequency cepstral coefficients (MFCCs) method was employed to extract acoustic emission (AE) signal characteristics during progressive fracturing. By integrating AE signal features with deep learning, enhancing the predictive capabilities of long short-term memory (LSTM) and bidirectional gated recurrent unit models via convolutional neural network (CNN) feature extraction and an attention mechanism (AM). This established a novel system combining artificial intelligence and rock mechanics for precise identification of progressive fracturing characteristics in rock masses. The results indicated that the fatigue-resistant properties of the sandstone under F-T cycles and seepage-stress coupling fundamentally accounts for the V-shaped evolution of its permeability during the unstable crack propagation stage, and manifests prematurely due to the cumulative effect of F-T damage. High-amplitude responses with dense MFCCs distributions prior to peak stress serve as precursors to impending sandstone failure under F-T cycles and seepage-stress coupling. The deep learning model, optimized with a CNN and AM, accurately identified the progressive fracturing characteristics of sandstone under F-T cycles and seepage-stress coupling, achieving over 90 % identification accuracy. This approach effectively integrates the dynamic optimization of deep learning models with long-term hazard prevention and control in cold-region rock masses.}, archive = {J_EAAI}, author = {Sen Zhang and Yongjun Song and Xiao Wang and Jianxi Ren}, doi = {10.1016/j.engappai.2025.110803}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110803}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-dimensional real-time data-driven identification for progressive fracturing characteristics of sandstone under freeze-thaw cycles and seepage-stress coupling}, volume = {152}, year = {2025}, }
• (2025). Internet of things based parking slot detection and occupancy classification for smart city traffic management. EAAI, 152, 110802. (www)

  @article{ , abstract = {With the rising issue of traffic congestion, there is a growing need for smart parking management systems. The goal is to support the idea of smart city traffic management by providing real-time data on the occupancy of outdoor and interior parking slots. It is a huge problem, therefore, to find a cost-effective image-based detection technology that can replace costly sensor-based technologies. This research presents a vision-based parking slot identification and occupancy categorization system that is both very accurate and built to work in real-time, and it is powered by the Internet of Things (IoT). Using the Improved Weight-Based White Shark Optimizer (IW-WSO) to further tune the hybrid Single Shot MultiBox Detector (SSD)-ShuffleNet model, the system is constructed to boost computing efficiency while preserving detection robustness across different environmental circumstances. The SSD-ShuffleNet model generates bounding boxes and confidence scores for parking spaces, while ShuffleNet ensures lightweight and computationally efficient processing. With an optimum support vector machine (OSVM) fine-tuned with IW-WSO, in this current research can detect the occupancy status (occupied or unoccupied) with a high accuracy of 96.5 %, which is better than typical deep learning-based approaches. Reliable performance in varying lighting and weather circumstances is shown by further performance parameters, which include a recall of 95.1 %, an AUC of 0.97, and an accuracy of 94.8 %. Operational in real-time on IoT devices, the system runs at an average inference time of 50 ms per frame. Congestion reduction, improved urban mobility, and sustainable smart city growth are all goals of the suggested framework, which provides a scalable and cost-effective solution for smart parking systems. Integrating this technology into current intelligent transportation systems is made possible by the findings, which show that it is resilient across a variety of environmental circumstances.}, archive = {J_EAAI}, author = {Hanadi Alkhudhayr}, doi = {10.1016/j.engappai.2025.110802}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110802}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Internet of things based parking slot detection and occupancy classification for smart city traffic management}, volume = {152}, year = {2025}, }
• (2025). A hypergraph-based dual-path multi-agent trajectory prediction model with topology inferring. EAAI, 152, 110799. (www)

  @article{ , abstract = {—Anticipating the motion of multiple agents in dynamic scenarios is a challenging yet crucial task in autonomous driving. Current research typically takes advantage of graphs to demonstrate pairwise interactions while underestimating the essentiality of groupwise interactions. To address this challenge, we present a dual-path social modeling framework that simultaneously captures both pairwise and groupwise interactions among agents. To facilitate social modeling, we develop an end-to-end adaptively topology inferring (ATI) mechanism that generates more robust structures as part of the proposed dual-path social modeling. The proposed mechanism leverages learnable thresholds and benefits from both commonly used thresholding and k-nearest neighbor methods, with the ability to determine the node connectivity in the scene from latent feature space and not be limited to a fixed number of neighbors. Moreover, for lower predictive deviation, we integrate a gated network into the ordinary Laplacian decoder where interaction-aware hidden states for agents are refined by corresponding mode-related time dependencies. We leverage the Average Displacement Error (ADE) and Final Displacement Error (FDE) as metrics. Lower ADE and FDE values mean that the predictions are closer to the ground truth. Extensive experiments are conducted on multiple public datasets. Benchmarks on the Standford Drone Dataset show that the ADE and FDE are 7.6 and 12.7, which outperforms a bunch of recent state-of-the-art methods in real-world scenarios. Furthermore, comprehensive ablation studies verify the effectiveness of the proposed methods. Results imply that the proposed mechanism has the ability to learn better topologies for agent-to-agent interactions as well as more accurate predictions.}, archive = {J_EAAI}, author = {Yu Hu and Xiaobo Chen and Yongjie Zhou and Hang Yang and Zhou Zhang and Jun Liang}, doi = {10.1016/j.engappai.2025.110799}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110799}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hypergraph-based dual-path multi-agent trajectory prediction model with topology inferring}, volume = {152}, year = {2025}, }
• (2025). A real-time lightweight object detection algorithm based on improved you only look once version 8 for unmanned surface vehicle. EAAI, 152, 110798. (www)

  @article{ , abstract = {Visual detection plays an indispensable role in the research and application of unmanned surface vehicle (USV), such as navigational safety and environmental perception. However, achieving a balance between detection accuracy and computational complexity is challenging due to limitations in memory resources and computational power. In this paper, a real-time lightweight object detection algorithm called CG-YOLOv8 is proposed based on YOLOv8(You Only Look Once Version 8) with CARAFE (Content-Aware Reassembly of Features) and Ghost modules. CG-YOLOv8 incorporates several key components to improve detection performance. Firstly, a specific detection layer is added to the backbone, and depth-wise separable convolution is applied to enhance detection efficiency for xlarge (Extra-large) objects. Secondly, the lightweight operator CARAFE is used to guide the up-sampling in the neck network, and the Ghost module is integrated to form the CGNeck (The Neck Network of Proposed Model), which can improve the expression capability of multi-scale features and reduce the burden of model. Finally, the robustness in complex environments is improved by introducing the loss function GIoU (Generalized Intersection over Union). The proposed model is validated on ship dataset collected by our USV under different marine environments. Compared to the YOLOv8, the CG-YOLOv8 increases the mAP50-95 (Mean Average Precision) to 85.9 % by improvement of 3.2 %, reduces the computation by 19.8 % and achieves a model size of 6.8M, with an FPS (Frames Per Second) of 125. Moreover, our model demonstrates superior detection performance, low computational complexity and high detection accuracy compared to other detection models.}, archive = {J_EAAI}, author = {Yinfeng Gong and Jiucai Jin and Deqing Liu and Peng Ren}, doi = {10.1016/j.engappai.2025.110798}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110798}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A real-time lightweight object detection algorithm based on improved you only look once version 8 for unmanned surface vehicle}, volume = {152}, year = {2025}, }
• (2025). Designing novel deep learning based frameworks for performance improvement of automatic modulation classification. EAAI, 152, 110796. (www)

  @article{ , abstract = {Efficiently recognizing modulated signals has grown more crucial as communication technologies advances. Unfortunately, obstacles including a large number of data and inaccurate identification provide a barrier to advancement in this field. Automatic modulation classification (AMC) based on deep learning (DL) has been suggested as a productive way to get good classification performance. However, under different noise situations, particularly at low signal-to-noise ratios (SNR), the majority of existing DL-AMC algorithms have limited generalization capabilities. In this paper, we developed three AMC models: a Bidirectional Long Short-Term Memory-Gated Recurrent Unit (BiLSTM-GRU), a hybrid Convolutional Neural Network (BiLSTM-GRU-CNN) for spatiotemporal features, and Bidirectional LSTM-GRU-CNN-Attention (AMC-BLGCA) with multihead attention. Our models can be implemented on the receiver end of a transmission system as modulation classifiers, which are capable of differentiating modulated schemes in various scenarios. According to the experiment, the AMC-BLGCA has the best overall accuracy on both datasets at 62.54% and 65.00% on RadioML2016.10a and RadioML2016.10b dataset respectively. The BiLSTM-GRU model takes second place with much fewer parameters and a very close accuracy score to the AMC-BLGCA model. The parallel hybrid-BiLSTM-GRU-CNN model has the lowest performance among the proposed methods, but still outperforms the recent AMC models. Our proposed methods are very lightweight and easily implementable to edge devices as they use fewer parameters and the need for computation resources is very low. The performance is 3% to 9% better than the current state-of-the-art (SOTA) AMC methods.}, archive = {J_EAAI}, author = {Soyabul Islam and Mirza Mohd Shahriar Maswood}, doi = {10.1016/j.engappai.2025.110796}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110796}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Designing novel deep learning based frameworks for performance improvement of automatic modulation classification}, volume = {152}, year = {2025}, }
• (2025). A multi-temporal granularity feature driven convolutional ensemble model for electricity theft detection. EAAI, 152, 110795. (www)

  @article{ , abstract = {Electricity theft causes substantial economic losses and safety hazards. While the widespread adoption of advanced metering infrastructure has significantly reduced electricity theft, perpetrators continue to find ways to exploit the system, employing increasingly covert and intricate methods. To address the ongoing challenge, this paper proposes an attention mechanism optimized multi-temporal granularity feature driven convolutional ensemble model for enhanced accuracy and robustness in electricity theft detection (ETD). For comprehensive feature extraction across diverse temporal scales, the proposed framework integrates two specialized feature extraction modules. The first module, a squeeze-and-excitation network-optimized temporal convolutional network, selectively focuses on informative temporal features within the electricity consumption data. The second module, a dual-dimensional attention enhanced deep residual network composed of residual blocks embedded with the convolutional block attention module, facilitates the model's concurrent learning of informative spatial and temporal features. Then, the features from each module are fused and classified through a fully connected layer. To validate the effectiveness of the proposed ETD method, this paper conducted simulation experiments using the publicly available dataset from the State Grid Corporation of China. The experimental results show that the model optimized with the attention mechanism significantly improves the performance of ETD. Compared to other ETD models, the proposed model performs excellently in various indicators under different training set ratios and sample imbalance scenarios, demonstrating good generalization and robustness. Additionally, the model was deployed on a Raspberry Pi edge computing device to further verify its feasibility in practical engineering applications.}, archive = {J_EAAI}, author = {Mingfa Yang and Qinyu Huang and Yulong Liu and Xidong Zheng and Tao Jin and Mohamed A. Mohamed}, doi = {10.1016/j.engappai.2025.110795}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110795}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-temporal granularity feature driven convolutional ensemble model for electricity theft detection}, volume = {152}, year = {2025}, }
• (2025). MVIFSA: Enhancing relation detection in knowledge base question answering through multi-view information fusion and self-attention. EAAI, 152, 110790. (www)

  @article{ , abstract = {In the domain of knowledge base question answering systems, relation detection plays a pivotal role in interpreting entities and their intricate networks. Traditional relation detection approaches model primarily from a question and relation-level perspective without taking full advantage of word-by-word relation representations. Existing hierarchical relation detection models typically use a unidirectional, layer-specific learning manner that ignores training losses across layers. They do not take into account the critical fraction of feature vectors from the input questions and relations. Towards this, we propose MVIFSA, a multi-view information fusion model based on self-attention for relation matching. Our approach employs a multi-view relational embedding strategy, which increases the word-by-word relational embedding dimension to capture the correlations between questions and relations more efficiently. The residual learning layer is designed to reduce training loss and avoid local optimum. This is achieved by constructing residual connections between the output of the information fusion layer and the complex relation representation layer. In addition, the introduction of the self-attention mechanism allows the model to selectively pay attention to the vectors of input questions and relations, thus improving its ability to align questions with the corresponding relations. Experimental results show that MVIFSA outperforms state-of-the-art methods on benchmark datasets such as SimpleQuestions, WebQuestionsSP, and PathQuestionLarge.}, archive = {J_EAAI}, author = {Guoxiang Tong and Maolin Wang and Qi Hu}, doi = {10.1016/j.engappai.2025.110790}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110790}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MVIFSA: Enhancing relation detection in knowledge base question answering through multi-view information fusion and self-attention}, volume = {152}, year = {2025}, }
• (2025). Adaptive self-evolving extreme learning machine-based terminal sliding mode control with application in retinal vein injection. EAAI, 152, 110789. (www)

  @article{ , abstract = {Retinal vein occlusion (RVO) is a serious condition that can lead to blindness. Injecting drugs into the retinal vein is a promising procedure for treating RVO. Due to the fragility of the retinal tissue, maintaining a precise drug flow rate (DFR) with a fast response is critical. Considering the unknown disturbance from piston dynamic and the drug-vein interaction, an adaptive self-evolving neural terminal sliding mode (ASNTSM) controller is proposed for DFR tracking. The integral terminal sliding surface is adopted to track the desired DFR in finite-time. The extreme learning machine (ELM) is utilized to estimate overall disturbances, and the adaptive switching gain is employed to compensate for the estimation error without requiring prior bounds. To achieve a compact ELM structure, a self-evolving mechanism is designed to implement the growth or pruning strategy of the hidden neurons. Theoretical analysis has proven that the ASNTSM controller can guarantee finite-time stability. Comparative experiments are conducted using a silicon phantom with simulated blood flow disturbances. The experimental results illustrate that the ASNTSM controller not only achieves lower transient time and average steady-state error, but also exhibits lower fluctuation and chattering effect. The self-evolving mechanism enhances the practicability of neural network in artificial intelligence-based medical engineering. Therefore, the ASNTSM controller is suitable for retinal vein injection tasks to improve surgical efficiency.}, archive = {J_EAAI}, author = {Bo Hu and Shiyu Xu and Lu Liu and Rongxin Liu and Mingzhu Sun and Xin Zhao}, doi = {10.1016/j.engappai.2025.110789}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110789}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive self-evolving extreme learning machine-based terminal sliding mode control with application in retinal vein injection}, volume = {152}, year = {2025}, }
• (2025). Joint feature extraction and alignment in object tracking with vision-language model. EAAI, 152, 110787. (www)

  @article{ , abstract = {Vision-language tracking is a new rising research topic that focuses on locating the target object in a video sequence using its language description. The main challenge is to model the correspondence between the vision-language input references and the test image, that is, vision-language fusion learning. Nevertheless, current vision-language trackers still follow the conventional framework in vision-only tracking, overlooking the heterogeneity between vision and language. To tackle the problem, we present a novel framework for vision-language tracking by combining joint feature extraction, alignment, and interaction. Before fusing, we first perform joint feature extraction and modality alignment in a vision-language embedding space learned by our proposed Adapter-equipped model, so as to obtain semantically unified feature representation. The model, named A-CLIP, which integrates some lightweight Adapters into CLIP (Contrastive Language-Image Pre-training), is an efficient method of seamlessly transferring the large-scale foundation model to downstream tracking task. It inherits the remarkable generalization ability from the large-scale model, providing a solution to the challenge of limited training data in vision-language tracking. Further, a transformer-based deep fusion is designed to model the multi-source correlations, highlighting the localization-relevant cues for accurate reasoning. Our proposed method has been extensively evaluated on three benchmark datasets, and the experimental results provide both quantitative and qualitative analyses that demonstrate its superior performance compared to the state-of-the-arts.}, archive = {J_EAAI}, author = {Hong Zhu and Qingyang Lu and Lei Xue and Guanglin Yuan and Kaihua Zhang}, doi = {10.1016/j.engappai.2025.110787}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110787}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Joint feature extraction and alignment in object tracking with vision-language model}, volume = {152}, year = {2025}, }
• (2025). Non-monotonic lyapunov function based membership function dependent robust control of Takagi–Sugeno fuzzy systems. EAAI, 152, 110785. (www)

  @article{ , abstract = {In the tasks of output robust stabilization control for discrete-time Takagi–Sugeno (T–S) fuzzy systems, a novel method is proposed to give better control performance and relax the conservatism. Firstly, a robust control method based on membership functions dependent (MFD) is obtained by utilizing a classic piecewise linear function approximation approach. A traditional membership functions independent (MFI) robust synthesis method is also presented. Subsequently, a new non-monotonic H ∞ robust control concept is proposed to better describe the system performance more comprehensively. Then, based on the non-monotonic Lyapunov function (NLF) approaches, system non-monotonic robust conditions are established in the form of linear matrix inequalities (LMl) with reduced conservatism. Finally, two simulation examples are utilized to verify the validity of the designed methods.}, archive = {J_EAAI}, author = {Wen-Bo Xie and Jin-Sheng Liang and Zi-Hao Wang and Jie Yang}, doi = {10.1016/j.engappai.2025.110785}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110785}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Non-monotonic lyapunov function based membership function dependent robust control of Takagi–Sugeno fuzzy systems}, volume = {152}, year = {2025}, }
• (2025). Adaptive learning network for detecting pavement distresses in complex environments. EAAI, 152, 110784. (www)

  @article{ , abstract = {In pavement detection, simultaneously identifying multiple defects and surface design features can save time and reduce operational costs. ShuttleNet version 3 (slice-based), referred to as ShuttleNetS3, builds on ShuttleNet to enhance pixel-level pavement detection by addressing the challenge of identifying diverse defects and surface design features, particularly cracks, potholes, and other irregular issues. This improved architecture introduces key innovations, including a learnable down-sampling module, slice-based positional sampling, which uses slicing and positional encoding to capture diverse features, and a learnable up-sampling module, slice-based positional upsampling, which ensures accurate detail restoration through bi-directional mapping. Additionally, Deformable Convolution version 4 allows the model to adapt to morphological variations, enabling precise extraction of complex defect features. With an average F1 Score of 97.48 % and an Intersection over Union of 95.19 % on a test set of 1500 images, ShuttleNetS3 excels in detecting subtle and irregular defects. Its adaptive learning capability ensures optimal performance in dynamic environments, providing an intelligent and precise solution for road management and maintenance.}, archive = {J_EAAI}, author = {Dingfeng Wang and Allen A. Zhang and Yi Peng and Yifan Wei and Huixuan Cheng and Jing Shang}, doi = {10.1016/j.engappai.2025.110784}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110784}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive learning network for detecting pavement distresses in complex environments}, volume = {152}, year = {2025}, }
• (2025). Adaptively multi-modal contrastive fusion network for molecular properties prediction. EAAI, 152, 110782. (www)

  @article{ , abstract = {Molecular property prediction has become the mainstream approach for revealing the underlying mechanisms of biomedical systems with molecular representations. Existing prediction methods based on deep learning typically learn features from molecules at a specific modality or simple fusion solution, failing to consider the inconsistency, complexity, and relationships inherent in multi-modal data. To solve this issue, an adaptively multi-modal contrastive fusion network (AMCFNet) is proposed to adaptively extract the complementary features from interaction and consensus between multi-modal representations for molecular property prediction of breast cancer. The proposed model begins with a two-stream feature extractor module, which learns both one-dimensional (1D) and two-dimensional (2D) molecular representations simultaneously. The basic part of the network is the adaptively contrastive fusion module, contrastively learning features between similar and different molecules with consensus scores, which can adaptively allocate weight to fuse semantic and structural information while avoiding cognitive gaps caused by inconsistencies within multi-modal. Additionally, the final complementary molecular representation is derived by integrating 1D, 2D, and fused 1D-2D features to enhance the prediction of molecular properties in breast cancer. The proposed AMCFNet model is evaluated on five estrogen receptor alpha (ER α ) and five compound public datasets, consistently outperforming state-of-the-art baselines in classification and regression tasks of molecular property prediction including single- and multi-modal methodologies.}, archive = {J_EAAI}, author = {Wenyan Tang and Meng Li and Yi Zhan and Bin Chen}, doi = {10.1016/j.engappai.2025.110782}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110782}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptively multi-modal contrastive fusion network for molecular properties prediction}, volume = {152}, year = {2025}, }
• (2025). Transformer-based deep probabilistic network for load forecasting. EAAI, 152, 110781. (www)

  @article{ , abstract = {Accurate electric power load forecasting is critical for power utility companies as it increases control over the relevant infrastructure, resulting in significant improvements in energy management and scheduling. However, point forecasting appears to fall short of providing these businesses with enough information to prepare for the worst. This paper proposes an encoder–decoder model that takes advantage of the expressiveness of Transformer-based encoders to produce probabilistic forecasts, i.e., a distribution over future predictions. Two real-world datasets are utilized to incorporate the performance of the proposed model on two different types of data: hourly load data from the power supply company of the city of Johor in Malaysia and hourly load consumption data from one of Grenoble Institute of Technology’s buildings. The former represents aggregated data, which makes identifying patterns and trends easier, but the latter was taken from a single building (non-aggregated), which increases the difficulty of forecasts. The model’s performance is discussed across multiple time horizons, including 24-hour, 1-week, and 1-month predictions. It achieved notable improvements compared to the used baseline, Amazon DeepAr. For 24 h ahead forecasting, accuracy was increased from 87.2 percent to 96.2 percent for Malaysian data and from 52.3 percent to 68.2 percent for Grenoble data. And for 1 month ahead forecasting, it was improved from 84.7 percent to 89.7 percent for Malaysian data, and from 45.5 percent to 57.2 percent for Grenoble data.}, archive = {J_EAAI}, author = {Omar Bouhamed and Maher Dissem and Manar Amayri and Nizar Bouguila}, doi = {10.1016/j.engappai.2025.110781}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110781}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer-based deep probabilistic network for load forecasting}, volume = {152}, year = {2025}, }
• (2025). Boosting meta-training with base class information for robust few-shot learning. EAAI, 152, 110780. (www)

  @article{ , abstract = {Few-shot learning aims to develop classifiers that can recognize new, unseen classes with only a few labeled examples. Meta-learning, with methods like Model-Agnostic Meta-Learning (MAML) and Prototypical Networks, has emerged as a key approach for this challenge. A recent advancement, Meta-Baseline, utilizes sequential pre-training and meta-training to achieve state-of-the-art performance. However, during meta-training, designed to further adapt the model to few-shot classification tasks, the class transferability gained during pre-training can be compromised, leading to suboptimal performance. We propose an end-to-end training paradigm with two alternating loops. The outer loop calculates cross-entropy loss across the entire training set while updating only the final linear layer. The inner loop uses the original meta-learning method to calculate the loss and integrate gradients from the outer loss to guide parameter updates. This approach enables effective adaptation to few-shot learning tasks while preserving robust generalization, outperforming existing baselines. Our findings suggest that leveraging information from the entire training set and the meta-learning training paradigm could mutually enhance one another. Our extensive experiments on the m i n i ImageNet, t i e r e d ImageNet, and CUB datasets demonstrate the effectiveness of our method. In the 5-way 1-shot setting, we achieve accuracies of 64.01%, 69.73%, and 68.07%, outperforming the best baselines by 0.94%, 1.11%, and 1.94%, respectively. In the 5-way 5-shot setting, our accuracies are 81.00%, 84.91%, and 83.44%, outperforming the best baselines by 1.74%, 1.17%, and 3.37%, respectively. Additionally, our framework is model-agnostic, yielding significant performance improvements when integrated with existing baseline training frameworks, providing an approximate 1% performance boost.}, archive = {J_EAAI}, author = {Weihao Jiang and Guodong Liu and Di He and Kun He}, doi = {10.1016/j.engappai.2025.110780}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110780}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Boosting meta-training with base class information for robust few-shot learning}, volume = {152}, year = {2025}, }
• (2025). Study on the deformation mode domain of shrink energy-absorbing structures based on curve feature classification method. EAAI, 152, 110779. (www)

  @article{ , abstract = {Shrink energy-absorbing structures play a key role in engineering applications by absorbing impact energy and ensuring passenger safety. However, inappropriate structural parameters and contact conditions can lead to buckling instability or folding collapse, which reduces the energy absorption efficiency. For this purpose, a deformation mode classification method based on the curve feature was proposed. A Long Short Term Memory (LSTM) network was used to predict the crushing force curve, followed by feature extraction and mode classification to establish the mapping relationships from design parameters to deformation modes. The deformation mode domain was then constructed using the classification model for data expansion, and its boundaries were precisely defined using surface fitting techniques. The critical cone angles of the shrink deformation mode at different friction coefficients were obtained by two-dimensional analysis. In addition, a structural design strategy was also proposed to maximize the specific energy absorption (SEA) of the structure under the shrink deformation mode. The results show that the classification method can effectively predict the deformation modes with 97 % accuracy. Further analysis of the deformation mode domain reveals that the critical cone angle of the shrink deformation mode decreases with the increase of the friction coefficient. Overall, this study predicts the deformation modes of shrink energy-absorbing structures and analyzes the variation of the critical cone angle, providing important guidance for structural optimization and improving energy absorption efficiency.}, archive = {J_EAAI}, author = {Jiaxing He and Ping Xu and Jie Xing and Shuguang Yao and Bo Wang and Xin Zheng}, doi = {10.1016/j.engappai.2025.110779}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110779}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Study on the deformation mode domain of shrink energy-absorbing structures based on curve feature classification method}, volume = {152}, year = {2025}, }
• (2025). Covariance attention guidance mamba hashing for cross-modal retrieval. EAAI, 152, 110777. (www)

  @article{ , abstract = {With the rapid development of artificial intelligence and multimedia technology, cross-modal hashing (CMH) has been widely applied in multimedia retrieval, recommendation systems, and large-scale data search due to its efficient query processing and low storage requirements, and has become a pivotal research area in both academia and industry. However, existing CMH algorithms fall short in exploiting the potential inter-modal correlations, leading to considerable semantic gaps. To overcome this issue, this paper proposes an innovative CMH framework called Covariance Attention Guidance Mamba Hashing (CAGMH) for Cross-Modal Retrieval. The framework enables deeper semantic alignment between modalities through a novel multi-feature fusion mechanism. This mechanism narrows the semantic gap and enhances the expressive power of each modality. Specifically, CAGMH exploits the distributional properties of covariance to optimize hash code generation and combined with the Mamba strategy to further improve cross-modal retrieval robustness. In addition, we design a novel loss function computation strategy that combines modal correlation with semantic consistency to optimize the model’s convergence and generalization ability. Experiments on four public benchmark datasets show that CAGMH surpasses state-of-the-art CMH methods, offering improved accuracy and efficiency in large-scale cross-modal similarity search. The corresponding code is available at https://github.com/Rooike111/CAGMH .}, archive = {J_EAAI}, author = {Gang Wang and Shuli Cheng and Anyu Du and Qiang Zou}, doi = {10.1016/j.engappai.2025.110777}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110777}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Covariance attention guidance mamba hashing for cross-modal retrieval}, volume = {152}, year = {2025}, }
• (2025). Pulling the carpet below the learner’s feet: Genetic algorithm to learn ensemble machine learning model during concept drift. EAAI, 152, 110772. (www)

  @article{ , abstract = {Data-driven models, in general, and machine learning (ML) models, in particular, have gained popularity over recent years with an increased usage of such models across the scientific and engineering domains. When using ML models in realistic and dynamic environments, users often need to handle the challenge of concept drift (CD). In this study, we explore the application of genetic algorithms (GAs) to address the challenges posed by CD in such settings. Formally, we propose a novel two-level ensemble ML model, which combines a global ML model with a CD detector, operating as an aggregator for a population of ML pipeline models, each one with an adjusted CD detector by itself responsible for re-training its ML model. In addition, we show that one can further improve the proposed model by utilizing off-the-shelf automatic ML (AutoML) methods. Through extensive synthetic dataset analysis, we show that the proposed model statistically significantly outperforms an ML pipeline with a CD algorithm, particularly in scenarios with unknown CD characteristics or a mixture of moving and shifting CDs. Moreover, we show a sub-linear decline in the proposed method’s performance with respect to a higher drifting rate and robustness to the underlying AutoML method utilized.}, archive = {J_EAAI}, author = {Teddy Lazebnik}, doi = {10.1016/j.engappai.2025.110772}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110772}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pulling the carpet below the learner’s feet: Genetic algorithm to learn ensemble machine learning model during concept drift}, volume = {152}, year = {2025}, }
• (2025). A composite insulator aging level classification method based on fourier transform infrared spectroscopy and deep learning model. EAAI, 152, 110770. (www)

  @article{ , abstract = {Composite insulators are critical components in power systems that ensure operational reliability, and accurate assessment of their aging status is crucial for grid safety. In this study, a large-scale dataset of Fourier Transform Infrared Spectroscopy (FTIR) data from methyl vinyl silicone rubber (MVQ) composite insulators is collected, and a classification model for aging levels is developed. An end-to-end composite insulator aging classification model based on deep learning, named Multiscale Depthwise Separable Convolution-Attention (MDSC-Attention), is proposed. The model effectively reduces computational complexity and improves efficiency by introducing one-dimensional depthwise separable convolutions. Additionally, it significantly enhances feature extraction and global modeling capabilities for FTIR data by combining multi-scale feature extraction modules with a self-attention mechanism. Experimental results demonstrate that the MDSC-Attention model can automatically extract key features from FTIR data and achieve high-precision classification of composite insulator aging levels, with a classification accuracy of 99.68 % on the test set. This study validates the potential application of deep learning in composite material aging assessment and provides significant support for state monitoring and intelligent evaluation in power systems.}, archive = {J_EAAI}, author = {Zhenan Zhou and Chuyan Zhang and Kun Zhang and Yu Deng and Chen Gu and Yinan Lin and Xinzhe Yu}, doi = {10.1016/j.engappai.2025.110770}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110770}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A composite insulator aging level classification method based on fourier transform infrared spectroscopy and deep learning model}, volume = {152}, year = {2025}, }
• (2025). Real-time monitoring of thermoelastic deformation of a silicon wafer with sparse measurements in the photolithography process using a physics-informed neural network and fourier neural operator. EAAI, 152, 110767. (www)

  @article{ , abstract = {This study proposes a novel approach for real-time prediction of temperature distribution and warpage in silicon wafers using only a few temperature sensors during photolithography processes. To achieve this, a physics-informed neural network-based surrogate model (PBSM) is employed to predict the full temperature distribution using a limited number of temperature sensor inputs. Subsequently, assuming minimal warpage caused by thermal effects, a one-way coupling approach is used to implement a Fourier neural operator (FNO) for predicting warpage based on the temperature field obtained from the PBSM. The FNO model is trained using supervised learning with paired data consisting of full temperature fields and corresponding warpage data generated by a commercial structural analysis solver. To evaluate the effectiveness of the proposed approach, a numerical example of the PEB process for a silicon wafer is analyzed. Various factors such as noise levels in sensor data, the number of sensors, and the number of training cases are examined, demonstrating the robustness of the approach. The results confirm that the PBSM and FNO operate sequentially in an end-to-end manner, enabling accurate real-time inference of both the temperature and warpage of the wafer. Furthermore, the approach achieves significantly superior anomaly detection for temperature uniformity and warpage compared to a data-driven approach using a deep neural network (DNN).}, archive = {J_EAAI}, author = {Hong-Kyun Noh and Myeong-Seok Go and Jae Hyuk Lim}, doi = {10.1016/j.engappai.2025.110767}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110767}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time monitoring of thermoelastic deformation of a silicon wafer with sparse measurements in the photolithography process using a physics-informed neural network and fourier neural operator}, volume = {152}, year = {2025}, }
• (2025). Location choice in solar power plants by applying meteorological data to multi-criteria decision-making method. EAAI, 152, 110766. (www)

  @article{ , abstract = {This study aims to determine the optimum generation locations for new solar power plants by evaluating meteorological data according to analytical hierarchy process (AHP). For this purpose, meteorological data for 2021 from three different provinces in Turkey were evaluated using AHP in order to determine which province has the best solar power plant energy potential. To test the accuracy of the obtained results, various scenarios were analyzed based on annual and seasonal data, and then compared against actual plant production data. Comparisons were made by taking into account both AHP scores and average plant production data. The AHP scores obtained for Kilis, Çanakkale, and Bursa provinces were 0.63, 0.38, and 0.16, respectively, whilst the annual average per-unit production of the plants were 0.21, 0.15, and 0.06, respectively. In both the average of real production data and AHP score results, Kilis was determined as the province with the highest potential. All these applications were realized in the same way according to seasonal periods. The AHP score for Kilis province, which was determined as the best location across all four seasons, was determined as 0.55, 0.66, 0.75, and 0.71 for spring, summer, autumn, and winter, respectively.}, archive = {J_EAAI}, author = {İsrafil Karadöl and Rıfat Yıldırım}, doi = {10.1016/j.engappai.2025.110766}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110766}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Location choice in solar power plants by applying meteorological data to multi-criteria decision-making method}, volume = {152}, year = {2025}, }
• (2025). Subtle defect detection network: More accurately detect subtle defects on the printed circuit board surface. EAAI, 152, 110765. (www)

  @article{ , abstract = {Printed circuit boards (PCBs) are the hardware foundation of large-scale integrated circuits, where surface quality inspection plays a critical role in manufacturing reliability. We proposed a subtle defect detection network (SDD-Net) to solve the detection problems in PCB surface defects, such as complex background, difficulty in distinguishing between foreground and background, random defect shape, area, and position. We propose a lightweight receptive field augmentation network (LRFA-Net) as the backbone effectively augments the receptive field, reduces parameters, and enhances feature extraction capabilities. A more lightweight multi-scale feature and coordinate information interaction mechanism was designed to enhance the capacity of the network to discern small targets in complex backgrounds. The combination of Varifocal Loss and Complete Intersection over Union Loss (CIoU) addresses the issue of distinguishing between the foreground and background, as well as adapting to PCB surface defects with variable shapes and positions. A lightweight Omni-dimensional dynamic convolutional prediction head (OD-Head) is designed to introduce multi-dimensional attention to effectively perceive small defects on the PCB surface. Compared with other algorithms, SDD-Net has a mean average precision ( mAP 0.5 ) of 99.6 % on the PCB Defect Augmented dataset, and the detection speed reaches 53 frames per second, which achieves a balance between accuracy and speed, and the effect is better than other algorithms. At the same time, SDD-Net is also experimentally verified on the real PCB surface welding defect data set, and the results show that SDD-Net also has the best detection effect.}, archive = {J_EAAI}, author = {Liying Zhu and Sen Wang and Mingfang Chen and Yang Zhu and Kaizhe Xing and Aiping Shen}, doi = {10.1016/j.engappai.2025.110765}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110765}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Subtle defect detection network: More accurately detect subtle defects on the printed circuit board surface}, volume = {152}, year = {2025}, }
• (2025). An evolutionary multitasking algorithm for multi-objective feature selection using dual-perspective reduction. EAAI, 152, 110764. (www)

  @article{ , abstract = {Feature selection inherently involves two conflicting objectives: minimizing the number of selected features and maximizing the classification accuracy. The exponential growth of the search space and complex interactions between features make high-dimensional feature selection challenging. Existing multi-objective methods suffer from slow convergence and limited search capabilities. Moreover, there is a lack of efficient methods for identifying feature subsets with equivalent objective values, which could offer diverse options. To address these issues, this paper proposes an evolutionary multitasking algorithm for multi-objective feature selection using dual-perspective reduction, called DREA-FS. First, a dual-perspective dimensionality reduction strategy is designed to generate simplified and complementary tasks through improved filter-based and group-based methods, facilitating the rapid identification of promising regions. To enable effective information sharing, a dual-archive multitasking optimization mechanism is proposed, which incorporates a diversity archive to preserve feature subsets with equivalent performance and maintain diversity. Coupled with an elite archive that offers convergence guidance, this mechanism achieves a balance between convergence and diversity across tasks, thereby enhancing the ability to search for equivalent feature subsets. Experimental results on 21 datasets demonstrate that the proposed method outperforms state-of-the-art multi-objective algorithms in classification performance. Besides, DREA-FS can identify different feature subsets with equivalent objective values, supporting decision-makers with diverse options and better interpretability.}, archive = {J_EAAI}, author = {Mengyue Wang and Hongwei Ge and Xia Wang and Liang Sun and Yaqing Hou and Bin Li}, doi = {10.1016/j.engappai.2025.110764}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110764}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An evolutionary multitasking algorithm for multi-objective feature selection using dual-perspective reduction}, volume = {152}, year = {2025}, }
• (2025). Towards machinery incremental fault diagnosis based on inverted transformer lifelong learning with learnable pruning mechanism. EAAI, 152, 110763. (www)

  @article{ , abstract = {To overcome catastrophic forgetting in deep learning for bearing diagnosis in wind turbines, it is necessary to boost stability-plasticity in lifelong learning that ensures the generation of high-quality exemplars translated from time-series signals in numerous sensors while incrementally learning multiple fresh classes. Therefore, an Inverted transformer lifetime learning method is forwarded to address the abovementioned limitations without tedious retraining for machinery fault diagnosis. First, the backbone of this method is the Inverted Transformer, which independently embeds the time-series signals of every sensor into tokens that simultaneously aggregate the global representations of series and enlarge the local receptive field via booming attention mechanisms. Second, the Inverted transformer expansion is developed to enable learning new and old knowledge by adding new branches based on the Inverted transformer to incrementally learn multiple new classes. Next, the learnable pruning mechanism is introduced to alleviate the dilemma caused by predefined and fixed structures in the previous stage and enhance the learning ability of the added fresh branch. Finally, a multi-objective training strategy is designed to overcome the class imbalance issues induced by several faults added in the incremental stage. The experimental results demonstrate the effectiveness and feasibility of the novel lifelong learning method.}, archive = {J_EAAI}, author = {Hai-hong Tang and Jia-wei Li and Wu-wei Feng and Peng Chen and Hong-tao Xue}, doi = {10.1016/j.engappai.2025.110763}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110763}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards machinery incremental fault diagnosis based on inverted transformer lifelong learning with learnable pruning mechanism}, volume = {152}, year = {2025}, }
• (2025). Real-time asphalt pavement ice detection and annotation with a transformer-based model framework. EAAI, 152, 110758. (www)

  @article{ , abstract = {Road icing is a common weather-related traffic hazard in cold winters, which increases the risk of traffic accidents. Traditional road ice detectors often damage road structures and are prone to interference. This paper develops a non-destructive ice detection model framework implemented based on the Transformer. The framework uses transfer learning to train high-accuracy transformer-based model. The framework sequentially integrates image classification and segmentation models to analyze the severity of asphalt pavement icing. The classification model (Swin Transformer) first classifies road conditions into dry, wet, or icy categories. Then, Masked-attention Mask Transformer (Mask2Former) segmentation model only annotates icy regions in the images, improving operational efficiency. The models were trained and validated on 4200 field images. The impact of training data volume on model performance, as well as the interference of road signs and cracks on model detection, was analyzed. Results show that the detection model demonstrated an accuracy of 98.5% in identifying icy roads. The annotation model achieved an Intersection over Union (IoU) score of 94.8% for icy regions, with a processing speed of 3.48 frames per second (FPS). To achieve the optimal performance, at least 1800 field images should be used for training the road icing detection model and annotation model. Road signs lead to a 6.77% decrease in the overall accuracy of the ice detection framework, while the impact of road cracks on the model is relatively minor. The framework presented in this paper is useful for transportation agencies in winter de-icing operations.}, archive = {J_EAAI}, author = {Jiaqi Chen and Huaqiang Wang and Yueyan Li and Shujun Yu}, doi = {10.1016/j.engappai.2025.110758}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110758}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time asphalt pavement ice detection and annotation with a transformer-based model framework}, volume = {152}, year = {2025}, }
• (2025). Conditional generative adversarial networks for spectrum-compatible ground motion generation of scenario earthquakes. EAAI, 152, 110755. (www)

  @article{ , abstract = {Obtaining acceleration time histories compatible with the design response spectrum is essential for structural seismic analysis. However, traditional methods for generating spectrum-compatible acceleration time histories often fail to incorporate physical characteristics, and the response spectrum cannot adequately represent key features such as spectral characteristics and non-stationarity. Therefore, this paper proposes a spectrum-compatible ground motion generation model, referred to as the conditional generative adversarial network-response spectrum (CGAN-RS). The model integrates a response spectrum discriminator (RS-D) and conditional variables, including moment magnitude, rupture distance, fault type, and time-averaged shear-wave velocity at the top 30 m, to ensure spectrum compatibility and physical consistency in the generated acceleration time histories. By integrating the response spectrum discriminator (RS-D) and conditional variables, the CGAN-RS model ensures spectrum compatibility and physical consistency in the generated acceleration time histories. Comparative analysis in the time-frequency domain between generated and recorded data under identical conditions reveals high similarity in peak ground acceleration, waveform characteristics, Arias intensity, and Fourier amplitude spectra. Furthermore, under identical scenario earthquakes, the CGAN-RS model achieves significant improvements in the coefficient of determination (R 2 ) between the target spectrum and the mean and median spectra of 50 generated spectra, with increases of 39.9 % and 19.7 % over the traditional CGAN model. Additionally, the mean absolute error (MAE) and the mean squared error (MSE) demonstrate substantial improvements. For the mean spectrum, the MAE is reduced by 62.23 %, and the MSE by 77.37 %. For the median spectrum, the MAE is reduced by 66.70 %, and the MSE by 83.99 %.}, archive = {J_EAAI}, author = {Zhenning Ba and Yang Ma and Jingxuan Zhao and Fangbo Wang}, doi = {10.1016/j.engappai.2025.110755}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110755}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Conditional generative adversarial networks for spectrum-compatible ground motion generation of scenario earthquakes}, volume = {152}, year = {2025}, }
• (2025). A lightweight deep learning approach for patient-specific electrocardiogram beat classification using local and long-term dependencies. EAAI, 152, 110754. (www)

  @article{ , abstract = {An electrocardiogram (ECG) is a graphical tool used to assess patients’ cardiac activity. Long-term ECG recordings, typically spanning 24 to 48 h, are crucial for detecting cardiac disorders. This paper introduces a novel, lightweight deep-learning architecture for classifying ECG beats as per the AAMI (Association for the Advancement of Medical Instrumentation) standard. The model integrates the advantages of Convolutional Neural Networks (CNN) and Bi-directional Long Short-Term Memory (Bi-LSTM) mechanisms in a single network to effectively capture local, temporal, and sequential patterns in ECG signals. Unlike conventional training, which often relies on fixed learning rates or predefined epochs, the proposed method dynamically adjusts learning parameters based on validation performance. Two Bi-LSTM layers effectively capture rich temporal dependencies, without requiring additional depth. The proposed method concatenates extracted CNN and BiLSTM features before the compact, dense layer, which will reduce the number of parameters significantly. This lightweight model ensures fast inference and low computational costs. Experimental results show that the proposed method achieves an accuracy of 99.21%, sensitivity of 98.66%, precision of 99.19%, and an F-score of 0.987. Additionally, the model demonstrates strong generalization capabilities, achieving high accuracies of 96.17% over different databases. The model‘s robustness and reliability in classifying ECG beats make it a practical and efficient tool for real-time monitoring applications.}, archive = {J_EAAI}, author = {Allam Jaya Prakash and Mohamed Atef}, doi = {10.1016/j.engappai.2025.110754}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110754}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight deep learning approach for patient-specific electrocardiogram beat classification using local and long-term dependencies}, volume = {152}, year = {2025}, }
• (2025). An intelligent Q-learning-based tree routing method in underwater acoustic sensor networks. EAAI, 152, 110753. (www)

  @article{ , abstract = {Nowadays, underwater acoustic sensor networks (UASNs) have emerged as an advanced and promising technology for developing various underwater applications. However, several routing protocols have been suggested for these networks in recent years. This subject is still facing many challenges such as low propagation speed, low bandwidth, and energy restrictions. To solve the above-mentioned challenges, this paper proposes an intelligent Q -learning-based t ree r outing method called QTRU for u nderwater acoustic sensor networks. The proposed scheme includes a network bootstrapping process to be aware of local network topology and calculate the neighboring table related to each node. QTRU also contains a Q-learning-based tree construction process to transmit data from sensor nodes to the sink node. In the routing tree construction process, the reward function in the Q-learning algorithm consists of four parameters, including the depth of the node, remaining energy, successful transmission probability, and the size of the candidate set. In addition, to calculate the state set in the Q-learning algorithm, each node carries out two screening operations on its neighboring nodes on the network. The first screening operation ensure that each node in the routing tree has the least number of hops to the sink node. The second screening operation is to avoid the formation of routing loops between sensor nodes and ensure a tree-based network topology. QTRU also designs a recovery mechanism and allows the sensor nodes present in the void area to select their best parent node in the routing tree. Finally, QTRU is implemented in network simulator version 2 (NS2), and its results are compared with three routing methods, namely reinforcement learning-based opportunistic routing protocol (RLOR), Q-learning-based multi-level routing protocol (MURAO), and energy-efficient depth-based routing protocol (EE-DBR). These results show that QTRU improves the packet delivery rate (about 8.94%), data integrity (about 5.95%), delay (about 7.31%), energy consumption (about 9.74%), and the number of hops in the communication route (about 5.03%).}, archive = {J_EAAI}, author = {Parisa Khoshvaght and Amir Haider and Amir Masoud Rahmani and May Altulyan and Monji Mohamed Zaidi and Mohammad Sadegh Yousefpoor and Efat Yousefpoor and Mehdi Hosseinzadeh}, doi = {10.1016/j.engappai.2025.110753}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110753}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An intelligent Q-learning-based tree routing method in underwater acoustic sensor networks}, volume = {152}, year = {2025}, }
• (2025). Adaptive protocols for hypervisor security in cloud infrastructure using federated learning-based anomaly detection. EAAI, 152, 110750. (www)

  @article{ , abstract = {A static security layer protecting the hypervisor from ever-evolving cyber attacks raises concerns about cloud computing security in the dynamic cybersecurity landscape. As cybercriminals modify their approaches, the security protocols should adapt accordingly. This paper introduces an adaptive communication protocol enhanced by federated learning (FL) to improve hypervisor security in cloud infrastructures. Federated learning is a decentralized machine learning (ML) approach that prevents data sharing while still allowing models to learn collaboratively across multiple hypervisors. Artificial Intelligence (AI)-based anomaly detection is incorporated into this framework to enhance hypervisor security in cloud infrastructures. The proposed system utilizes local and global anomaly detection models to dynamically adjust security protocols and protect hypervisors against threats such as hyperjacking, side-channel attacks, and virtual machine (VM) escape. Experimental results demonstrate the protocol’s effectiveness, achieving a detection accuracy of 92.6%, significantly higher than the 85.2% from centralized learning and 78.4% from static protocols. Furthermore, the adaptive approach reduced communication overhead by 55% and training time by 32%, emphasizing its efficiency and operational performance. This research highlights the potential of integrating adaptive protocols with federated learning to enhance cloud security, offering a robust defense against evolving cyber threats.}, archive = {J_EAAI}, author = {Moutaz Alazab and Albara Awajan and Areej Obeidat and Nuruzzaman Faruqui and Aaron Bere and Saqib Ali and Wei Wei}, doi = {10.1016/j.engappai.2025.110750}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110750}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive protocols for hypervisor security in cloud infrastructure using federated learning-based anomaly detection}, volume = {152}, year = {2025}, }
• (2025). Intelligent predictive networks of swimming of microorganism for tangent hyperbolic fluid flow with heat and mass transfer and thompson and troian slip conditions using levenberg-marquardt algorithm. EAAI, 152, 110749. (www)

  @article{ , abstract = {The objective of this study is to devise a predictive model for mass and heat transfer in electrically conductive tangent hyperbolic nanofluids under the influence of gyrotactic microorganisms and an effect of a horizontally stretched sheet with Thompson and Troian slip conditions. Improving the Levenberg-Marquardt Algorithm (LMT) models so that it can effectively train a Backpropagation Neural Network (BP-NN) for predicting accurate fluid flow, temperature, concentration, and microorganism density profiles at different physical parameters. By means of similarity transformations, the mathematical model is first converted to a system of nonlinear Partial Differential Equations (PDEs) into Ordinary Differential Equations (ODEs) which is solved through the BVP4C method created in MATLAB. The study confirms that the LMT accompanied with BP-NN provides a reliable and fast tool to investigate the complexities of non-Newtonian fluid, flow and diffusive mass transfer induced by tangent hyperbolic fluids and gyrotactic microorganisms. That advances the application of machine learning methods to fluid mechanics, including thermal management, and manufacturing cooling systems or biomedical fluid dynamics, by improving predictive capabilities.}, archive = {J_EAAI}, author = {M.B. Arain and D. Hussain and Fuad A.M. Al-Yarimi and S. Shaheen and J. Hu}, doi = {10.1016/j.engappai.2025.110749}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110749}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent predictive networks of swimming of microorganism for tangent hyperbolic fluid flow with heat and mass transfer and thompson and troian slip conditions using levenberg-marquardt algorithm}, volume = {152}, year = {2025}, }
• (2025). Dual-sampling feature fusion for three-dimensional object detection using four-dimensional radar and camera. EAAI, 152, 110747. (www)

  @article{ , abstract = {Bird’s-eye view (BEV) fusion of four-dimensional (4D) radar and camera provides a cost-effective solution for three-dimensional (3D) object detection, ensuring all-weather perception in autonomous vehicles. However, existing BEV-based fusion methods experience progressive degradation of camera-generated features due to overfitting and inconsistent generalization during training, which limits their effectiveness in multi-modal fusion. Although complex networks estimate 3D information from images, their accuracy remains suboptimal in end-to-end training, while their high computational and storage costs hinder practical deployment. To address these challenges, we propose a novel approach for multi-modal 3D object detection using 4D radar and camera fusion, termed dual-sampling feature fusion (DSFusion). We design two modules: predefined voxel-based view transformation (PVVT) and radar pillar-based view transformation (RPVT). Both modules aim to sample image features through point cloud matrix transformations. The RPVT module effectively captures more reliable image features by sampling the raw radar point cloud, mitigating the feature degradation caused by direct view transformation. Meanwhile, the PVVT module employs a predefined point cloud to sample additional image features and utilizes the radar occupancy estimation network (ROENet) to enhance feature selection efficiency. Finally, multiple BEV features are fused through convolution, and a unified detection head predicts object classes and locations. Experimental results demonstrate that DSFusion achieves state-of-the-art radar-camera fusion detection performance on the View-of-Delft and TJ4DRadSet datasets, approaching the accuracy of high-resolution light detection and ranging while maintaining all-weather capability. Furthermore, DSFusion enables real-time inference at over 10 frames per second, meeting the real-time requirements of autonomous vehicle applications.}, archive = {J_EAAI}, author = {Xin Bi and Caien Weng and Panpan Tong and Wei Tian and Lu Xiong}, doi = {10.1016/j.engappai.2025.110747}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110747}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-sampling feature fusion for three-dimensional object detection using four-dimensional radar and camera}, volume = {152}, year = {2025}, }
• (2025). An insulator defect detection network combining bidirectional feature pyramid network and attention mechanism in unmanned aerial vehicle images. EAAI, 152, 110745. (www)

  @article{ , abstract = {Insulator defect detection is a crucial aspect of power transmission line inspection. To achieve effective detection of insulator defects, a deep learning-based object detection network is applied to unmanned aerial vehicle images, offering advantages such as high accuracy, high efficiency, and low cost. While the you only look once version8 network demonstrates superior performance in insulator defect detection compared to other methods, it still struggles to achieve ideal results in scenarios with variable scales and complex backgrounds. To address this issue, an improved network is proposed in this paper, tailored to the characteristics of insulator defect detection using unmanned aerial vehicle images. Firstly, an attention mechanism is introduced into the backbone network to attenuate the influence of background in the target region. Secondly, the bidirectional feature pyramid network structure is incorporated, and cross-layer connections and weighted fusion are implemented during feature extraction and fusion to enable the network to better focus on insulator defect features and suppress the impact of noise. Lastly, the detection head at the lowest layer is replaced with a small target detection head to enhance the network's attention to small target. Experimental results on the self-made dataset demonstrate that recall and precision of the network are 89.9 % and 96.5 %.}, archive = {J_EAAI}, author = {Fu Feng and Xiaoxia Yang and Ronghao Yang and Hao Yu and Fangzhou Liao and Qiqi Shi and Feng Zhu}, doi = {10.1016/j.engappai.2025.110745}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110745}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An insulator defect detection network combining bidirectional feature pyramid network and attention mechanism in unmanned aerial vehicle images}, volume = {152}, year = {2025}, }
• (2025). Multi-level feature decomposition and fusion model for video-based multimodal emotion recognition. EAAI, 152, 110744. (www)

  @article{ , abstract = {The application of artificial intelligence to Multimodal Emotion Recognition (MER) significantly enhances human–computer interaction. However, effectively fusing heterogeneous multimodal information, such as facial expressions and body gestures, remains a significant challenge. Most existing MER approaches force these features into a common subspace, leading to a loss of modality-private information. In addition, the temporal dynamics and mutual influences of multiple simultaneous physiological responses in videos remain an understudied area of research. To address these, a Multi-level Feature Decomposition and Fusion (MLDF) model is proposed to improve the Video-based MER. First, an event-based spatio-temporal vision Transformer is employed to extract crucial emotional changes and spatio-temporal features from facial and body information in video clips. Furthermore, a multi-level feature decomposition loss is utilized to disentangle modality-private and shared representations, enhancing the model’s ability to capture fine-grained details and high-level semantic information. Finally, a dynamic feature fusion strategy is proposed to adaptively integrate multimodal multi-level features, enhancing emotion recognition performance. The MLDF outperforms the existing state-of-the-art methods on both EmoReact and ElderReact datasets. Its ability to effectively capture temporal information in videos and improve heterogeneous fusion makes it suitable for smart video analysis systems for MER.}, archive = {J_EAAI}, author = {Xiaofang Liu and Guotian He and Shuge Li and Fan Yang and Songxiying He and Lin Chen}, doi = {10.1016/j.engappai.2025.110744}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110744}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-level feature decomposition and fusion model for video-based multimodal emotion recognition}, volume = {152}, year = {2025}, }
• (2025). DF-LSH: An efficient double filters locality sensitive hashing for approximate nearest neighbor search. EAAI, 152, 110742. (www)

  @article{ , abstract = {Locality-sensitive hashing (LSH) is an effective randomized technique for high-dimensional approximate nearest neighbor search. In practice, LSH has been used in large-scale recommendation and information retrieval tasks. However, it often suffers from the problem of generating many false positives due to the randomness of hash functions, which significantly increases the query overhead needed to retrieve these false positives. To this end, recent LSH variants have proposed more rigorous search schemes or used more compact hash codes to identify eligible candidates. However, as the number of candidates increases, query performance often degrades. To effectively address this problem, we propose a Double Filters Locality Sensitive Hashing scheme, called DF-LSH, designed for efficient approximate nearest neighbor search in high-dimensional datasets. DF-LSH first learns data-aware hash functions for a Bloom filter to decrease false positives, and then leverages the geometrical nature of random projection for further filter. By employing these double filters, DF-LSH provides probability guarantees similar to that of standard LSH schemes for query performance. Experiment results demonstrate that DF-LSH achieves superior query performance in both accuracy and efficiency over various high-dimensional datasets, with an average query time reduction of up to around 45x compared against previous LSH techniques while maintaining identical query accuracy.}, archive = {J_EAAI}, author = {Zhenpei Shan and Defu Zhang and Yunqi Lei}, doi = {10.1016/j.engappai.2025.110742}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110742}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DF-LSH: An efficient double filters locality sensitive hashing for approximate nearest neighbor search}, volume = {152}, year = {2025}, }
• (2025). Finite-time consensus of leader-following multi-agent systems with event-triggered control strategy utilizing absolute velocity information. EAAI, 152, 110738. (www)

  @article{ , abstract = {This paper investigates the problem of finite-time consensus (FTC) of second-order multi-agent systems (MASs) with intrinsic nonlinear dynamics. A novel distributed event-triggered control (ETC) protocol has been proposed. This protocol achieves the control objectives in scenarios where both the velocity information of the leader and its neighboring agents is simultaneously absent, and introduces new triggering strategy, effectively reducing the demand for communication bandwidth and ensures the tracking of the leader’s trajectory. Subsequently, rigorous proofs are conducted using finite-time stability theory and Lyapunov stability theory to ensure leader-following consensus and establish the settling time while excluding Zeno behavior. Finally, through simulations involving multi-Unmanned Aerial Vehicle (multi-UAV) collaboration systems, the effectiveness of the proposed control protocol is demonstrated.}, archive = {J_EAAI}, author = {Guoliang Tan and Hongwei Ren and Feiqi Deng}, doi = {10.1016/j.engappai.2025.110738}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110738}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Finite-time consensus of leader-following multi-agent systems with event-triggered control strategy utilizing absolute velocity information}, volume = {152}, year = {2025}, }
• (2025). Advanced computational modeling of darcy-forchheimer effects and nanoparticle-enhanced blood flow in stenosed arteries. EAAI, 152, 110737. (www)

  @article{ , abstract = {Enhancing blood flow in stenosed arteries through the optimization of nanoparticle-based solutions is one potential application of this research to improve medication delivery systems and therapeutic therapies for cardiovascular disorders. This research employed artificial neural networks (ANNs) to analyze the Darcy-Forchheimer flow of hybrid nanoparticles embedded in magnetized blood flow across the stenosed arteries. The ANNs were specifically used to consider the influence of heat generation and activation energy on this complex flow scenario. Heat transfer analysis accounts for multiple factors, such as thermal radiation, viscous dissipation, heat sources, and Joule heating. These elements collectively exert a substantial influence on the overall heat transfer process. Similarity transforms are applied to convert the original Partial Differential Equation (PDE) into a more manageable Ordinary Differential Equation (ODE). This ODE is solved numerically with MATLAB's built-in bvp4c scheme. The Levenberg–Marquardt Algorithm (LMA) is considered in multi-layer perceptron models with 10 neurons in the hidden layers. The ANN model is structured with a configuration that includes 9 input layers for data entry, 3 output layers for results, and 10 hidden layers that process information between the input and output stages. Increased porous parameter values indicate thermal energy is held longer within the flow. Increasing the heat source and thermal radiation leads to a rise in the Nusselt number while increasing the flow parameter causes it to drop.}, archive = {J_EAAI}, author = {Fatima Shafiq Hira and Qammar Rubbab and Irshad Ahmad and Afraz Hussain Majeed}, doi = {10.1016/j.engappai.2025.110737}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110737}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advanced computational modeling of darcy-forchheimer effects and nanoparticle-enhanced blood flow in stenosed arteries}, volume = {152}, year = {2025}, }
• (2025). A novel dynamic fractional-order discrete grey power model for forecasting china's total solar energy capacity. EAAI, 152, 110736. (www)

  @article{ , abstract = {-Precise prediction of the total solar energy capacity is pivotal for the progress of the nation's solar energy industry, the optimization of energy structure and the sustainable development of energy systems. This study proposes a novel dynamic fractional order discrete grey power model (DFDGPM(1,1)) for predicting China's total solar energy capacity. The model introduces a power exponent to capture the nonlinear characteristics among system behavior variables. Additionally, it incorporates a fractional-order accumulation operator and a dynamic time-delay function, which not only describe the time-delay effect between China's economic development and solar energy growth but also enhance the model's adaptability to different samples. The model demonstrates strong compatibility and can degenerate into 10 existing grey models. Empirical research shows that the model's fitting error is close to 0 %, with a prediction error of only 1.07 %, which is significantly better than 11 other methods. The forecast findings indicate that China's total solar energy capacity will experience an annual growth rate of 29.41 % from 2022 to 2030. This method promotes the development of dynamic forecasting technology and provides the necessary technical and data support for renewable energy field.}, archive = {J_EAAI}, author = {Lin Xia and Yuhong Wang and Youyang Ren and Ke Zhou and Yiyang Fu}, doi = {10.1016/j.engappai.2025.110736}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110736}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel dynamic fractional-order discrete grey power model for forecasting china's total solar energy capacity}, volume = {152}, year = {2025}, }
• (2025). Multi-modal feature integration network for visible-depth-thermal salient object detection. EAAI, 152, 110734. (www)

  @article{ , abstract = {In recent years, the task of salient object detection in multi-modal scenarios has attracted more and more attention, where the increase of modalities is beneficial for improving the detection performance of models. However, though the existing saliency models have achieved encouraging performance, they overlook the unbalanced information content between visible modality and other auxiliary modalities ( i.e., depth and thermal modalities), and lack the full utilization of multi-level features. This will lead to insufficient multi-modal fusion and multi-level integration. Therefore, in this paper, we propose a multi-modal feature integration network (MFINet) for Visible-Depth-Thermal (VDT) salient object detection (SOD), which contains three key modules. Firstly, we utilize the three-modal feature fusion (TMFF) module to enhance and fuse the multi-modal features by emphasizing effective feature channels and enlarging the receptive fields of features, where we further emphasize the visible cues. Secondly, we present a neighborhood layer feature enhancement (NLFE) module, which can utilize the complementary information from adjacent TMFF modules to enhance the decoder features by using different spatial attention strategies. Thirdly, a multi-level cascade feature integration (MCFI) module is proposed to aggregate the multi-level decoder features in a cascade way, acquiring the final high-quality saliency maps. Comprehensive experiments conducted on the VDT-2048 dataset demonstrate that our model outperforms the state-of-the-art models in terms of all evaluation metrics. The code is available at https://github.com/banjamn/MFINet .}, archive = {J_EAAI}, author = {Fengyv Cui and Xiaofei Zhou and Liuxin Bao and Bin Wan and Ran Shi and Qiang Chen and Jiyong Zhang}, doi = {10.1016/j.engappai.2025.110734}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110734}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modal feature integration network for visible-depth-thermal salient object detection}, volume = {152}, year = {2025}, }
• (2025). A binary linear predictive evolutionary algorithm with feature analysis for multiobjective feature selection in classification. EAAI, 152, 110733. (www)

  @article{ , abstract = {Multiobjective feature selection (MOFS), which aims to obtain a set of Pareto optimal feature subsets by simultaneously maximizing classification accuracy and minimizing the number of selected features, has attracted considerable attention recently. However, most existing studies still face a challenge that locating more well-distributed Pareto optimal feature subsets, especially for high-dimensional complex datasets. In response to this challenge, this paper proposes a binary linear predictive evolutionary algorithm with feature analysis (MBLPE) for MOFS. In this paper, a feature analysis-based selection method is proposed to select effective solutions into the next generation. Concretely, two subset evaluation indicators are designed to efficiently deal with duplicated feature subsets during evolution. Then, a fitness allocation is constructed to select effective solutions, improving population diversity. Moreover, a fisher score-based initialization scheme is designed when handling high-dimensional complex datasets. The proposed scheme effectively removes irrelevant and redundant features in search space by identifying features with strong discriminant performance in advance, thereby reducing computational cost. In comparison with seven state-of-the-art algorithms on 18 classification datasets with different characteristics, the proposed MBLPE finds more diverse feature subsets with better convergence in solving the MOFS problems.}, archive = {J_EAAI}, author = {Ting Zhou and Limin Wang and Xuming Han and Zhiquan Liu and Minghan Gao}, doi = {10.1016/j.engappai.2025.110733}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110733}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A binary linear predictive evolutionary algorithm with feature analysis for multiobjective feature selection in classification}, volume = {152}, year = {2025}, }
• (2025). Multisource data-driven intelligent method for detecting surface defects in cold-rolled copper strips. EAAI, 152, 110730. (www)

  @article{ , abstract = {Surface defects are critical in determining the quality of cold-rolled copper strips. Although deep learning-based detection techniques are widely used, they rely heavily on large amounts of annotated data, while in actual production, there are practical bottlenecks in sample acquisition and annotation. To address these challenges, a multi-source data-driven lightweight strip surface defect detection method (Ms-LSSD) is proposed. First, a dataset containing nine types of defects was curated, named the Cold-Rolled Copper Strip defect dataset (CRCS9-DET). Then, the architecture, based on Denoising Diffusion Probabilistic Model (DDPM), improves dataset diversity and mitigates issues related to sample size and imbalance. Next, a lightweight real-time object detector is designed based on Real-Time DEtection TRansformer (RT-DETR) to achieve precise detection of surface defects in cold-rolled copper strips. A novel convolution operator, Adaptive Partial Convolution (APConv), is proposed by introducing channel attention mechanism to enhance representational capacity across different tasks. And an Efficient Adaptive Residual network (EARNet) is established as the foundation for achieving a lightweight model. Finally, experiments were conducted on CRCS9-DET dataset and a public dataset (NEU-DET). Ms-LSSD achieves mean Average Precision (mAP 50 ) of 90.3 % on CRCS9-DET dataset and mAP 50 of 78.7 % on NEU-DET, with a detection speed of 52.2 Frames Per Second (FPS), striking a balance between detection performance and inference speed. Compared to state-of-the-art algorithms, Ms-LSSD demonstrates more comprehensive performance potential, making it suitable for industrial applications.}, archive = {J_EAAI}, author = {Bowei Duan and Dongcheng Wang and Yuehua Ma and Guodong Wang and Hongmin Liu}, doi = {10.1016/j.engappai.2025.110730}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110730}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multisource data-driven intelligent method for detecting surface defects in cold-rolled copper strips}, volume = {152}, year = {2025}, }
• (2025). A new approach to failure mode and effect analysis based on regret theory and group satisfaction index in basin-type insulators. EAAI, 152, 110729. (www)

  @article{ , abstract = {Gas-insulated metal-enclosed switchgear, as an important part of the power transmission and transformation project of the global energy Internet, has the advantages of high reliability and small footprint. As an important part of the switchgear, basin-type insulators will seriously affect the reliability and safety of the switchgear operation in case of failure such as breakage or partial discharge. There are still many limitations in the current risk assessment methods for basin-type insulators, resulting in inaccurate risk identification and classification. Therefore, this paper proposes an extended failure mode and effect analysis (FMEA) based on the defects of traditional FMEA model based on regret theory(RT), group satisfaction index and the combined compromise solution (CoCoSo) method to assess the failure mode of basin-type insulators. First, the expert applies the 5Why analysis to obtain the failure modes of basin-type insulators. Then the risk assessment information given by the experts in uncertain linguistic terms (ULT) is assembled by probabilistic uncertain linguistic terms sets (PULTSs). Secondly, in order to improve the accuracy of the expert group information during the calculation, this paper proposes a new group satisfaction index for describing the degree of satisfaction and disagreement of expert opinions, and constructs a PUL-MEREC-SWARA integrated weight acquisition method based on this index for determining the weights of risk factors. Finally, in order to improve the accuracy and stability of the ranking, a new PUL-RT-CoCoSo integrated decision framework is proposed to rank the importance of failure modes in a more rational way. The stability, validity and superiority of the proposed model are demonstrated through the sensitivity analysis of parameters and comparison with several typical ranking methods.}, archive = {J_EAAI}, author = {Jianping Fan and Yihua Duan and Meiqin Wu}, doi = {10.1016/j.engappai.2025.110729}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110729}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new approach to failure mode and effect analysis based on regret theory and group satisfaction index in basin-type insulators}, volume = {152}, year = {2025}, }
• (2025). Virtual workflows and adaptive optimization scheduling of production process with feedback constraints. EAAI, 152, 110728. (www)

  @article{ , abstract = {Feedback structure process often exists in production processes with precise quality requirements. The constraint of feedback process is conditional and reverse, and these peculiarities increase the complexity of scheduling problem. A dynamic scheduling model based on virtual workflows is designed and a scheduling optimization method with a new adaptive sequencing rule strategy is proposed for flexible production scheduling problem with feedback constraints. Based on feedback structure virtualization and virtual nodes responding to feedback disturbances, the dynamic scheduling model implements a mechanism for synchronizing the updating of task workflows and machine workflows with activities of feedback processes to trigger rescheduling. To adapt to the correlation between key dynamic features of scheduling scenario and multi-objective balance, the adaptive sequencing rule strategy with mean tardiness as dominant objective and makespan and energy consumption as regular objectives is proposed to improve multi-objective dynamic trade-off optimization performances of the scheduling method based on hybrid decision-making mechanism. Comparative tests are conducted to verify that the hybrid decision-making mechanism incorporating the proposed adaptive rule strategy can effectively improve the comprehensive dominance level of scheduling optimization results. Simulation tests show that the dynamic scheduling model can become a way to respond to feedback constraint disturbances and trigger rescheduling.}, archive = {J_EAAI}, author = {Zhen Quan and Yan Wang and Xiang Liu and Zhicheng Ji}, doi = {10.1016/j.engappai.2025.110728}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110728}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Virtual workflows and adaptive optimization scheduling of production process with feedback constraints}, volume = {152}, year = {2025}, }
• (2025). Machine learning methods for predicting shoreline change from submerged breakwater simulations. EAAI, 152, 110726. (www)

  @article{ , abstract = {The design of submerged breakwaters for coastal protection requires many iterations of process-based simulations to determine combinations of input variables’ influence on the consequent shoreline response. By using machine learning, the complex non-linear multidimensional relation between input and modelled shoreline output can be modelled to predict shoreline change on unseen combinations of inputs with much faster simulations. To prove this, a neural network and Gaussian process regressor were trained on a dataset of 243 combinations of structural and wave forcing input parameters paired with their process-based modelled shoreline change output on a straight shoreline after one year. The models were capable of fitting the dataset of input–output relations comprising outputs of accretion and erosion, allowing for predictions on incremental input changes. The proposed hybrid modelling approach takes advantage of the computational advantages of machine learning to leverage physics-driven output data generated from process-based models to optimise the design options of submerged breakwaters relative to shoreline outcomes. Given the multidimensional multiparameter processes, the solution space is large, however, by focusing on key parameters of interest, resource-intensive processes can be optimised to collect high-quality data for important site parameters and using this machine learning method to develop a shoreline change model.}, archive = {J_EAAI}, author = {Subaru Ken Muroi and Edoardo Bertone and Nick Cartwright and Fernando Alvarez}, doi = {10.1016/j.engappai.2025.110726}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110726}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning methods for predicting shoreline change from submerged breakwater simulations}, volume = {152}, year = {2025}, }
• (2025). A faster heterogeneous parallel computing method for tucker decomposition. EAAI, 152, 110725. (www)

  @article{ , abstract = {Artificial intelligence (AI) technology is developing explosively in application fields such as speech recognition, semantic understanding, and computer vision. In particular, the new generation of knowledge enhancement large language models have gradually become the infrastructure of social productivity. With the development of large models towards multi-mode, the input data and model parameters characterized by tensors are becoming increasingly large. This puts high demands on computation and storage. Tucker decomposition obtains the optimal low-rank representation of the natural tensor by factor matrices and a core tensor, reducing storage and computation requirements in big data and artificial intelligence applications. However, the existing Tucker decomposition methods show limited computational speed and convergence performance. In this paper, a general heterogeneous computing framework of the Tucker decomposition method is proposed, which analyzes the row independence of factor matrices and the column independence of Kruskal matrices, and updates the Kruskal matrices instead of the core tensor in a column-wise manner and factor matrices in a row-wise manner respectively to reduce the storage overhead in computation. Further, the proposed method employs a heterogeneous computing platform to speed up the computing bottlenecks and takes full advantage of fine-grained parallel optimization technology to improve memory access efficiency. The experimental results show that the computation speed achieves a 3.1 to 75.4 times improvement compared to the latest methods. Among all the methods, it exhibits the best convergence performance.}, archive = {J_EAAI}, author = {Xiaosong Peng and Laurence T. Yang and Jie Li and Wenjun Jiang}, doi = {10.1016/j.engappai.2025.110725}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110725}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A faster heterogeneous parallel computing method for tucker decomposition}, volume = {152}, year = {2025}, }
• (2025). An optimized updating adaptive federated learning for pumping units collaborative diagnosis with label heterogeneity and communication redundancy. EAAI, 152, 110724. (www)

  @article{ , abstract = {Rapid developments in the petroleum industry create a strong demand for the remote fault diagnosis of pumping units. As pumping units are often located in harsh and communication-constrained environments, collecting and transmitting data with heterogeneous labels is severely restricted, which poses a serious challenge to traditional centralized diagnostic models that rely on data aggregation. To address this issue, we propose an optimized updating adaptive federated learning method for decentralized fault diagnosis of pumping units with label heterogeneity and communication redundancy. The method uniquely integrates data distillation-based initialization, divergence metric strategy, and adaptive optimization to tackle label heterogeneity and optimize communication efficiency in a unified framework. Specifically, a data distillation-based initialization trick is designed to compensate for label heterogeneity across local clients. Then, a divergence metric strategy is used to reduce local communication overhead. Besides, it is integrated into the adaptive optimization mechanism to guide the global model aggregation. Extensive experiments are carried out to validate the superiority of the proposed method using both private and public datasets. Compared with the benchmarks, the diagnostic accuracy and communication efficiency of the proposed method are improved by 1.58% and 9.53%, respectively.}, archive = {J_EAAI}, author = {Zhi-Wei Gao and Yaohai Xiang and Shixiang Lu and Yuanhong Liu}, doi = {10.1016/j.engappai.2025.110724}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110724}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An optimized updating adaptive federated learning for pumping units collaborative diagnosis with label heterogeneity and communication redundancy}, volume = {152}, year = {2025}, }
• (2025). Orthogonal diversity nonnegative matrix factorization for multi-view clustering. EAAI, 152, 110715. (www)

  @article{ , abstract = {In the context of rapid development of artificial intelligence, how to extract valuable information from complex multidimensional data has become a core research problem. Multi-view clustering methods based on non-negative matrix factorization (NMF) are widely used in multi-view data analysis, but still face many challenges in practical applications. Current multi-view clustering methods usually solve the problem of diversity among viewpoints by orthogonalization of view representations. However, they fail to fully utilize the rich features of each viewpoint because data from different viewpoints may be interrelated. In addition, existing methods fail to fully consider the orthogonality between base matrices while emphasizing the diversity of view representations. For this reason, this paper proposes a new orthogonal diversity non-negative matrix factorization method (ODNMF). First, ODNMF explores the orthogonality of the representations of sample pairs between different viewpoints. This approach preserves the characteristics of each perspective and enhances the diversity of data representations. Second, ODNMF orthogonalizes the basis matrix of each viewpoint to reduce redundant features and enhance data interpretability and representation. Finally, ODNMF introduces graph regularization for each view to reveal the intrinsic geometric and structural information of features. Experimental results show that ODNMF significantly outperforms existing state-of-the-art algorithms on seven datasets.}, archive = {J_EAAI}, author = {Xinling Zhang and Chengcai Leng and Jinye Peng and Irene Cheng and Anup Basu}, doi = {10.1016/j.engappai.2025.110715}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110715}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Orthogonal diversity nonnegative matrix factorization for multi-view clustering}, volume = {152}, year = {2025}, }
• (2025). Prediction of compressive strength of multiple types of fiber-reinforced concrete based on optimized machine learning models. EAAI, 152, 110714. (www)

  @article{ , abstract = {The accurate prediction of compressive strength of fiber-reinforced concrete (FRC) is essential for its design optimization and performance assessment, as it can significantly reduce testing costs. However, the high variability of FRC's compressive strength poses considerable prediction challenges. Current research has predominantly focused on developing prediction models for single-type FRC, while the prediction of compressive strength across multiple types of FRC remains a critical and unresolved issue in the field. To address this gap, this study proposes a novel hybrid approach integrating Deep Neural Networks (DNN), Generalized Regression Neural Networks (GRNN), and Extreme Gradient Boosting (XGBoost) with optimization techniques—Particle Swarm Optimization (PSO), Bayesian Optimization (BO), and Bald Eagle Search (BES). A comprehensive dataset of 386 peer-reviewed compressive strength measurements was utilized, with K-means++ algorithm ensuring balanced training and testing set distributions. Hyperparameter optimization for DNN, GRNN, and XGBoost was conducted by combining PSO, BO, and BES with five-fold cross-validation. Results demonstrate strong model performance, with the BES-XGBoost model achieving the highest accuracy, exhibiting deviations of approximately 15 % between actual and predicted values. Additionally, Shapley Additive Explanations (SHAP) and partial dependence plots were employed to analyze the feature importance on compressive strength and the coupling effects of fiber characteristics. The proposed approach not only provides enhanced prediction accuracy for multiple types of FRC but also delivers valuable insights for FRC proportioning design, advancing the field of FRC performance evaluation.}, archive = {J_EAAI}, author = {Ning Zhao and Haonan Zhang and Peilun Xie and Xiaowei Chen and Xuewei Wang}, doi = {10.1016/j.engappai.2025.110714}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110714}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction of compressive strength of multiple types of fiber-reinforced concrete based on optimized machine learning models}, volume = {152}, year = {2025}, }
• (2025). Gossypium herbaceum: Folium disease identification and classification using efficient net-coordinate convolutional neural network (EcoNet). EAAI, 152, 110701. (www)

  @article{ , abstract = {Globally, agriculture plays a key role in economic growth, with cotton being one of the most essential fiber crops, often referred to as “silver fiber” and widely used in manufacturing industries. However, cotton plants face productivity challenges due to disease outbreaks, making early detection crucial. Manual monitoring is tedious, and conventional cnn-based deep transfer learning models suffer from spatial information loss and translational variance, reducing classification accuracy. To address these challenges, this study proposes EcoNet, a novel Efficient Net-Coordinate Convolutional Neural Network (EcoNet) designed to enhance spatial feature extraction and positional encoding for improved cotton leaf disease identification and classification. The dataset was compiled using an amalgamated method: in-field fungal and healthy cotton leaf images were collected from Thoppulampatti, Madurai, India, while bacterial and insect-borne disease images were obtained from Kaggle's public repository. Pre-processing involved image resizing (224 × 224 pixels) and normalization to optimize model performance. EcoNet integrates Coordinate Convolution Neural Network (Coordconv) layers to capture local and global spatial dependencies, while residual blocks improve hierarchical feature extraction and reduce computational overhead. Experimental results demonstrate that EcoNet achieves 98.32 % accuracy with a Cohen's Kappa coefficient of 0.9804, outperforming conventional models.}, archive = {J_EAAI}, author = {S. Naveena and K. Kavitha}, doi = {10.1016/j.engappai.2025.110701}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110701}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Gossypium herbaceum: Folium disease identification and classification using efficient net-coordinate convolutional neural network (EcoNet)}, volume = {152}, year = {2025}, }
• (2025). Large-scale analytic hierarchy process method based on fuzzy-rough-advantage relation. EAAI, 152, 110699. (www)

  @article{ , abstract = {In the big data era, the complexity of decision-making problems on a large scale has risen, particularly when confronted with large-scale alternatives (objects). Traditional multi-attribute decision-making methods, however, are becoming less effective in handling these problems. To address this, the present study proposes a Large-scale Analytic Hierarchy Process (LAHP) approach predicated on fuzzy-rough-advantage relations. Initially, an unsupervised learning technique, known as K-means++, is applied to cluster the alternatives (objects). Subsequently, the fuzzy-rough-advantage relationship is utilized to form the judgment matrix for these clusters. Following this, the Analytic Hierarchy Process (AHP) principles are employed to determine the weights of each cluster based on various evaluation criteria. These weights are then utilized to compute the overall scores of each cluster and identify the optimal group. Finally, the applicability of the LAHP method is confirmed through case analysis, with the ranking outcomes being subsequently analyzed and discussed.}, archive = {J_EAAI}, author = {Bin Yu and Zeyu Xiao and Yinglong Dai and Zeshui Xu}, doi = {10.1016/j.engappai.2025.110699}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110699}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Large-scale analytic hierarchy process method based on fuzzy-rough-advantage relation}, volume = {152}, year = {2025}, }
• (2025). Efficient crack segmentation with multi-decoder networks and enhanced feature fusion. EAAI, 152, 110697. (www)

  @article{ , abstract = {In infrastructure management, intelligent crack detection is vital, particularly for maintaining crucial elements such as road networks in urban areas. Detecting pavement defects promptly and accurately is essential for timely repairs and hazard prevention. However, this task is challenging due to factors, such as complex backgrounds, micro defects, diverse defect shapes and sizes, and class imbalance issues. Innovative approaches and advanced technologies are needed to address these challenges and effectively manage infrastructure complexities. In this study, we propose a novel framework for crack segmentation, called CrackMaster. CrackMaster utilizes advanced neural network architectures, leveraging the next generation of convolutional networks (ConvNeXt) as an encoder and dual decoders customized for distinct tasks. The first decoder adopts a self-supervised learning paradigm to reconstruct images, thereby enhancing feature extraction capabilities. Meanwhile, the second decoder combines deep labelling network for semantic image segmentation (Deeplabv3+) with a light deep neural network (LinkNet) to facilitate precise segmentation. Notably, we introduce an Enhanced Feature Fusion (EFF) block to improve features quality, enhancing information flow and context preservation, thus boosting segmentation performance. Experimental results conducted on three diverse datasets, including our in-house Road Crack Dataset (RCD), DeepCrack537, and Yang Crack Dataset (YCD) datasets, demonstrate the effectiveness of our framework achieving outstanding Intersection over Union scores (IoU) of 86.0%, 87.8%, and 76.9%, respectively, showing superior accuracy and robustness in crack segmentation tasks. These findings underscore the potential applicability of our framework in real-world infrastructure management scenarios. The code is publicly available at: https://github.com/AmmarOkran/CrackMaster .}, archive = {J_EAAI}, author = {Ammar M. Okran and Hatem A. Rashwan and Adel Saleh and Domenec Puig}, doi = {10.1016/j.engappai.2025.110697}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110697}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient crack segmentation with multi-decoder networks and enhanced feature fusion}, volume = {152}, year = {2025}, }
• (2025). Adaptive few-shot image augmentation for fine-grained industrial defects based on region-level modeling. EAAI, 152, 110695. (www)

  @article{ , abstract = {Image augmentation can improve detection performance in industrial inspection. However, traditional image augmentation fails to create novel defects. Current defect image generation often overlooks normal structures and struggles to capture fine-grained defect features with limited training samples. To address these challenges, we propose AHDEN (Adaptive Hierarchical Defect Editing Networks), which consists of AdaSDN (Adaptive Shape Design Networks) and H-DEN (Hierarchical Defect Editing Networks), to synthesize high-quality fine-grained defects with accurate binary masks by modeling the relationship between defects and normal regions. Firstly, AdaSDN generates adaptive and diversified defect masks by fusing the normal structures and random styles employing self-attention, incorporating the proposed proportion regularization and double discriminators to ensure that the mask aligns with the defect characteristics. Secondly, H-DEN designs defects based on normal textures and binary masks. The modeling of region-level relationships between defect and normal textures enables the generation of defects solely based on a single sample, and the defect strength can be controlled by fine-tuning fusion factors. Thirdly, we demonstrate the inadequacy of conventional metrics in evaluating industrial defects and propose a new metric, Texture Distance (TD), based on detailed texture features. Experimental results in various real-world industrial scenarios demonstrate that our TD metric provides more realistic evaluation results, and our AHDEN successfully generates diversified and high-quality defect samples even with only one training sample. Finally, AHDEN significantly improves the performance of various segmentation models and surpasses state-of-the-art methods with the IoU (Intersection over Union) improvement of 6.33 % and the F1 score improvement of 5.95 %.}, archive = {J_EAAI}, author = {Jing Wei and Qingfeng Shi and Zhengtao Zhang and Fei Shen and Huiyuan Luo}, doi = {10.1016/j.engappai.2025.110695}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110695}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive few-shot image augmentation for fine-grained industrial defects based on region-level modeling}, volume = {152}, year = {2025}, }
• (2025). FSGFuse: Feature synergy-guided multi-modality image fusion. EAAI, 152, 110691. (www)

  @article{ , abstract = {The purpose of infrared and visible image fusion (IVIF) is to integrate the complementary properties of the two modalities to obtain a fused image with a more comprehensive information representation. However, existing multi-modal image fusion (MMIF) methods mainly rely on auto-encoders to extract source features and achieve feature complementarity through fusion mechanisms. These methods fail to fully utilize the potential of the encoder in feature decoupling and information integration, resulting in difficulties in effectively capturing and fusing cross-modal features, which in turn introduces redundant information. To tackle the challenge, this study proposes a Feature Synergy-Guided Fusion (FSGFuse) network, which achieves IVIF by synergizing complementary features to preserve texture details and acquire thermal target information. In the pre-training stage, the feature synergism loss is utilized to guide the double-branch encoder to perceive the consistency and difference of cross-modal features, which in turn motivates the model to capture the more discriminatory feature representations in the source features. In the fusion stage, a Feature Complementary Fusion (FCF) module is designed to guide feature fusion. This module not only integrates complementary features, but also captures long-range contextual relationships between features, which facilitates cross-modal interaction while enriching the semantic information. Experiments on publicly available benchmark datasets show that the fusion performance of FSGFuse significantly outperforms existing state-of-the-art methods and positively affects the effectiveness of the downstream object detection task, with a mean average precision (mAP) improvement of 18.57%. This result fully demonstrates its practical application value.}, archive = {J_EAAI}, author = {Qiuhan Shao and Zheng Guan and Xue Wang and Hang Li}, doi = {10.1016/j.engappai.2025.110691}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110691}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {FSGFuse: Feature synergy-guided multi-modality image fusion}, volume = {152}, year = {2025}, }
• (2025). Dimensionality reduction of rolling bearing fault data based on graph-embedded semi-supervised deep auto-encoders. EAAI, 152, 110689. (www)

  @article{ , abstract = {Conventional shallow models struggle to extract effective features from high-dimensional and sparse bearing fault data, which negatively impacts fault classification accuracy. To address this issue, we propose a dimensionality reduction method for rolling bearing fault datasets based on a Graph Embedding Semi-supervised Deep Auto-Encoder (GESDAE). This method constructs traditional fault datasets into a data network graph embedding structure and inputs them into a sparse dropout regularized semi-supervised deep auto-encoder for dimensionality reduction. The supervised component utilizes first-order proximity to maintain the local network graph embedding structure, while the unsupervised component employs second-order proximity to capture the global network graph embedding structure. By reasonably introducing a sparse dropout regularization term in the original objective function, the method helps prevent model overfitting and enhances the robustness and generalization ability of GESDAE. The effectiveness of the proposed method is validated using vibration signals from rolling bearing systems on two different experimental platforms. Results show that the proposed method achieves a correct recognition rate of 97.3 % for rolling bearing faults, with a 4.8 % improvement compared to state-of-the-art dimensionality reduction methods. Furthermore, visualization and convergence analysis demonstrate that the proposed method not only retains highly nonlinear local and global network graph embedding structures but also significantly enhances the separability of fault features, showcasing its superior robustness and generalization performance in fault diagnosis systems.}, archive = {J_EAAI}, author = {Kongyuan Wei and Rongzhen Zhao and Haixia Kou and Pengfei Chen and Yongyong Cao and Yuqiao Zheng and Linfeng Deng}, doi = {10.1016/j.engappai.2025.110689}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110689}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dimensionality reduction of rolling bearing fault data based on graph-embedded semi-supervised deep auto-encoders}, volume = {152}, year = {2025}, }
• (2025). Deep content and contrastive perception learning for automatic fetal nuchal translucency image quality assessment. EAAI, 152, 110687. (www)

  @article{ , abstract = {Automatic quality assessment of fetal nuchal translucency ultrasound images can assist physicians in obtaining standard planes and improve the reproducibility of nuchal translucency screening. At present, there are no special studies and methods for the quality assessment of fetal nuchal translucency ultrasound images. For this task, main challenges are low image quality, content identification of structural integrity and relative position relationship, time consumption for data collection and fine-grained annotation. To address these challenges, we propose a framework based on DenseNet model, which includes preprocessing module, content perception module, attention learning module and contrastive regularization module. Experiments show that the modules are effective for improving the quality assessment framework performance. And this framework is better than the other fourteen deep learning models. This framework can provide the sonographer with a model interpretable reference map. Bland–Altman experimental analysis also verifies the consistency between the results obtained by the automatic quality assessment framework and the manually annotated clinical dataset. Therefore, the proposed quality assessment framework for fetal nuchal translucency ultrasound images has the prospect and value of clinical application.}, archive = {J_EAAI}, author = {Lili Zhao and Yuanyuan Xu and Jian Xu and Weiping Ding and Jinzhao Yang and Huiyu Zhou and Yiming Du and Bin Hu and Lichi Zhang and Qian Wang}, doi = {10.1016/j.engappai.2025.110687}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110687}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep content and contrastive perception learning for automatic fetal nuchal translucency image quality assessment}, volume = {152}, year = {2025}, }
• (2025). Evaluation of detrended fluctuation analysis applied to audio signals from drum cymbals in different machine learning classification contexts. EAAI, 152, 110683. (www)

  @article{ , abstract = {Drum cymbals are musical instruments whose acoustic complexity is built from the union of various factors, from their constituent material to the final touch in the manufacturing process. The use of machine learning techniques to classify cymbals has been investigated in the literature over the years. However, most of these studies use a simplified strategy based on extracting features from signals. With this in mind, this paper explores the effectiveness of applying the combined triggering and Detrended Fluctuation Analysis (DFA) pre-processing techniques compared to using a set of time-domain information retrieved from the signals. To carry out this task, two contexts for classifying cymbals were considered: classification (i) by their bronze alloys (3 classes) and (ii) by each cymbal individually (4 classes). In addition, five machine learning models were evaluated. The Karhunen-Loève transformation based Classifier (KLTbC) performed well in both the three-class and the four-class classification schemes, achieving high accuracy with small standard deviations. In addition, the computational experiments showed that a trigger was useful to identify the initial moment of signal, and that the DFA could be applied as a pre-processing step, reducing data with a smaller number of points than the original signals, indicating that this is a promising strategy for the problem.}, archive = {J_EAAI}, author = {Tales Boratto and Elineudo Pinho de Moura and Douglas Fonseca and Alexandre Cury and Leonardo Goliatt}, doi = {10.1016/j.engappai.2025.110683}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110683}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evaluation of detrended fluctuation analysis applied to audio signals from drum cymbals in different machine learning classification contexts}, volume = {152}, year = {2025}, }
• (2025). Analytical propagation of errors and uncertainties in deep neural networks in the domain of physical sciences. EAAI, 152, 110682. (www)

  @article{ , abstract = {In recent years, neural networks have gained popularity for making predictions in various applications. However, these predictions come with inherent uncertainties due to factors such as data acquisition, parameter choices, network architecture, and the training process. These uncertainties, both epistemic and aleatory, require careful quantification and propagation for a comprehensive understanding of the predicted results. To address these requirements, we propose an algorithm called Analytical Propagation of Uncertainties (APU), which is capable of quantifying and assessing the uncertainties associated with predictions made by Multilayer Perceptron neural networks in the domain of physical sciences. APU, with its modular architecture, can provide model, data, and prediction uncertainties for pre-trained Multilayer Perceptron networks. While generally offering lower computational cost compared to Bayesian methods, the efficiency and effectiveness of APU may vary depending on the specific application and model complexity. To the best of our knowledge, this is the first method able to propagate measurement uncertainty, originating from the calibration of measuring instruments, in the training process. To evaluate the effectiveness of data uncertainty estimation, we compared our approach against Monte Carlo Dropout in the Airfoil Self-Noise dataset. The analysis revealed that APU is more realistic than Monte Carlo Dropout and satisfactorily matched Monte Carlo simulations. Thus, the APU method becomes promising under real-world conditions, especially in the field of physical sciences, where low computational cost and immediate response are essential. This is significant in edge computing scenarios, real-time monitoring, addressing concerns such as latency, bandwidth, and data privacy. The study contributes to advancing the application of deep learning in physical sciences. The source code is available at https://github.com/Geduardomello/Dissertation .}, archive = {J_EAAI}, author = {Gerson Eduardo de Mello and Vitor Camargo Nardelli and Rodrigo da Rosa Righi and Luiz Schirmer and Gabriel de Oliveira Ramos}, doi = {10.1016/j.engappai.2025.110682}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110682}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Analytical propagation of errors and uncertainties in deep neural networks in the domain of physical sciences}, volume = {152}, year = {2025}, }
• (2025). A unified solution for replacing position embedding in vision transformer for object detection. EAAI, 152, 110679. (www)

  @article{ , abstract = {The traditional Vision Transformer (ViT) demonstrating outstanding performance in various computer vision tasks. However, this high performance relies on substantial training on large datasets, for position embedding which is the key part of ViT, requires extensive data to maximize their effectiveness. What is more, in diverse vision task scenarios, position embedding may not perform well and even yield redundant information, which limits the overall model’s flexibility and robustness. To tackle this problem, we propose a novel position-free embedding model HV-SwinViT , which embed both Horizontal and Vertical features by using the self-attention mechanism with the position-free functions, replacing the positional embedding. To achieve this, firstly, we propose a module Horizontal and Vertical orientation Swin Transformer block ( HVblock ), which generate feature maps containing both horizontal and vertical information by adopting fully connected layers, then we design two hybrid sub-network:HVblock in Backbone ( HV-Swin-B ) and Channel Fusion 2 with HVblock ( Cf2-HV ) which leverage convolution layer and HVblock to tackle positional relationship information. Also we defined a learnable nonlinear activation function to increase the sensitivity to nonlinear position features. Experimental results demonstrate that the proposed HV-SwinViT model achieves an improvement of over 0.1% in Average Precision (AP) compared to state-of-the-art methods on the MS-COCO2017 dataset. Additionally, our model outperforms several network architectures designed specifically for aerial photography of targets, attaining an A P score of 35.6% and 32.1% on the VisDrone2019 and AI-TOD , respectively. These results confirm that the HV-SwinViT model can be a unified solution and highlight the stability and robustness of our approach across diverse scenarios.}, archive = {J_EAAI}, author = {XueZhuan Zhao and JiaWei Wang and LingLing Li and XiaoYan Shao and KeXin Zhang}, doi = {10.1016/j.engappai.2025.110679}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110679}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A unified solution for replacing position embedding in vision transformer for object detection}, volume = {152}, year = {2025}, }
• (2025). A goal-conditioned offline reinforcement learning algorithm and its application to quad-rotors. EAAI, 152, 110678. (www)

  @article{ , abstract = {Goal-conditioned reinforcement learning has garnered significant academic interest due to its ability to accomplish reinforcement learning tasks with different goals. Nevertheless, the requirement for extensive online interaction with the environment makes it exceedingly dangerous for practical implementations. This paper proposes a goal-conditional offline reinforcement learning algorithm called Goal-Conditional Twin Delayed Deep Deterministic with V-based Behavioral Cloning(GC-TD3VBC) and applies it to the navigation task of a quad-rotor. In this study, offline reinforcement learning algorithm Twin Delayed Deep Deterministic Behavioral Cloning(TD3BC) is extended to deal with goal-conditioned tasks. Specifically, goal information is explicitly incorporated into TD3BC’s value function network and policy network. Furthermore, to address the training inefficiency caused by sparse rewards with goal condition, this study integrates TD3BC with Hindsight Experience Replay(HER). Additionally, it devises composite advantage action weights to guide the agent in selecting advantageous actions based on the advantage function during the policy network update. The proposed approach significantly improves the overall performance of the algorithm. Finally, the algorithm’s practicality and performance were evaluated using an offline reinforcement learning benchmark dataset. Moreover, the algorithm was implemented on the target navigation task of a quad-rotor. The experimental results demonstrate that GC-TD3VBC is capable of efficiently addressing target navigation tasks of the quad-rotor under goal conditions.}, archive = {J_EAAI}, author = {Haojun Zhong and Zhenlei Wang and Yuzhe Hao}, doi = {10.1016/j.engappai.2025.110678}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110678}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A goal-conditioned offline reinforcement learning algorithm and its application to quad-rotors}, volume = {152}, year = {2025}, }
• (2025). Advancing interpretable cardiac disease diagnosis via a transformer-convolutional hybrid network on electrocardiograms. EAAI, 152, 110675. (www)

  @article{ , abstract = {Manual heart disease diagnosis with the electrocardiogram (ECG) is intractable due to the intertwined signal features and lengthy diagnosis procedure, especially for the 24-hour dynamic ECG signals. Consequently, even experienced cardiologists may face difficulty in producing all accurate ECG reports. In recent years, Artificial Intelligence (AI), particularly neural network-based automatic ECG diagnosis methods have exhibited promising performance, suggesting a potential alternative to the labor-intensive examination conducted by cardiologists. However, many existing approaches failed to adequately consider the temporal and channel dimensions when assembling features and ignored interpretability. And clinical theory underscores the necessity of prolonged signal observations for diagnosing certain ECG conditions such as tachycardia. Moreover, specific heart diseases manifest primarily through distinct ECG leads represented as channels. In response to these challenges, this paper introduces a novel neural network architecture for ECG classification (diagnosis). The proposed model incorporates Lead Fusing blocks, transformer-XL (meaning extra long) encoder-based Encoder modules, and hierarchical temporal attentions. Importantly, this classifier operates directly on raw ECG time-series signals rather than cardiac cycles. Signal integration begins with the Lead Fusing blocks, followed by the Encoder modules and hierarchical temporal attentions, enabling the extraction of long-dependent features. Furthermore, existing convolution-based methods have been argued to compromise interpretability, whereas the proposed neural network provides improved clarity in this regard. Experimental evaluations on a comprehensive public dataset confirm the superiority of the proposed classifier over state-of-the-art methods. Moreover, a visualization method was employed to generate a location map that demonstrates the areas of the signal emphasized by the model, thereby enhancing interpretability.}, archive = {J_EAAI}, author = {Xiaoqiang Liu and Yinlong Xu and Hongxia Xu and Liang He and Siyu Long and Yisen Huang and Yubin Wang and Yingzhou Lu and Yingxuan Huang and Jian Wu and Honghao Gao and Xiaobo Liu}, doi = {10.1016/j.engappai.2025.110675}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110675}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advancing interpretable cardiac disease diagnosis via a transformer-convolutional hybrid network on electrocardiograms}, volume = {152}, year = {2025}, }
• (2025). Explainable time series features for hard disk drive failure prediction. EAAI, 152, 110674. (www)

  @article{ , abstract = {Reliable data storage is crucial for industry digitalization and cloud infrastructure. To prevent data loss and improve maintenance efficiency in data centers, timely replacement of Hard Disk Drives (HDDs) is critical. HDDs are equipped with Self-Monitoring, Analysis, and Reporting Technology (SMART) to track key performance indicators with empirically established thresholds. In recent years, various machine learning models have utilized SMART time series data for early HDD failure prediction. However, decision-makers need greater transparency and explainability to trust and implement these data-driven models. In this work, we proposed a framework that extracts explainable features from the SMART time series and visualizes the feature impact on short-term HDD failure prediction using SHapley Additive exPlanations (SHAP) analysis. We trained an eXtreme Gradient Boosting (XGBoost) model with information-rich features and evaluated the failure detection rate and false alarm rate. We demonstrated the effectiveness of the proposed approach on Backblaze data for Quarter 1 and Quarter 2 of 2022. The model provided a 74.7% failure detection rate with only a 0.73% false alarm rate on the test data from Quarter 3 of 2022, outperforming an existing explainable model benchmark of a 54.68% failure detection rate and an 11.85% false alarm rate. In addition, the sensitivity analysis optimizes the signal length and the lead time to improve prediction accuracy and inform predictive maintenance policies. The results demonstrate the potential of the proposed framework for effective HDD failure prediction with explainable features. The proposed framework is also applicable to other sensor-based industrial equipment monitoring applications.}, archive = {J_EAAI}, author = {Wei Li and Haozhou Zhou and Srinivasan Radhakrishnan and Sagar Kamarthi}, doi = {10.1016/j.engappai.2025.110674}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110674}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable time series features for hard disk drive failure prediction}, volume = {152}, year = {2025}, }
• (2025). Cross-domain human activity recognition based on deviation-graph constrained non-negative matrix factorization. EAAI, 152, 110661. (www)

  @article{ , abstract = {Sensor-based human activity recognition (HAR) often faces challenges associated with domain shift and limited labeled samples. Many studies have been done to address these cross-domain issues in HAR. Nonetheless, there are two main limitations of the existing methods on the cross-person HAR problem. First, the influence of noise or outliers within the original feature space is often overlooked, potentially adversely affecting the classification accuracy of target samples. Second, the local graph constructed from the source domain sample labels may contain redundancies, which can lead to the transfer of incorrect knowledge to the target domain. To cope with these two limitations, we propose cross-domain human activity recognition approach based on Deviation-Graph Constrained Non-Negative Matrix Factorization. We introduce a deviation constraint strategy to mitigate the partial impact of noise and outliers on overall feature space. We also create a novel local graph to avoid redundant information from the source domain affecting local manifold structures, thereby enhancing the effectiveness of knowledge transfer. Then, we perform joint non-negative matrix factorization on the samples from both domains. Under the constraint of the new local manifold structure, samples of the same category from different domains can iteratively approach, thereby enhancing the performance of correct knowledge transfer. We conducted comprehensive experiments on Daily and Sports Activities Dataset. The results demonstrate that our method outperforms comparison methods by 3.41%–24.49% in single-source tasks and 4.62%–42.34% in multi-source tasks.}, archive = {J_EAAI}, author = {Yue Li and Yuxing Zhi and Huaijun Wang and Yufan Guo and Kan Wang and Lei Yu and Rong Fei and Junhuai Li}, doi = {10.1016/j.engappai.2025.110661}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110661}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-domain human activity recognition based on deviation-graph constrained non-negative matrix factorization}, volume = {152}, year = {2025}, }
• (2025). Hierarchical cascaded networks with multi-task balanced loss for fine-grained hashing. EAAI, 152, 110659. (www)

  @article{ , abstract = {Fine-grained image retrieval has seen significant advancements with the rise of deep hashing methods. However, these methods often prioritize high-level features, which may lead to the loss of important low-level details in hash code representations. Additionally, balancing the classification and hashing tasks remains a challenge. To address these issues, we propose a Hierarchical Cascaded Network (HCN) with a multi-task balanced loss function tailored for fine-grained hashing. Our model captures detailed information from different feature levels through a hierarchical backbone network and utilizes a cascaded representation learning module to enhance and fuse features using attention mechanisms. An adaptive loss function ensures a balanced contribution from both classification and hashing tasks during training. Extensive experiments on benchmark datasets demonstrate that HCN outperforms state-of-the-art methods, achieving a promising improvement across five datasets and multiple hash code lengths. These results highlight the effectiveness of HCN in enhancing fine-grained image retrieval, with potential applications in areas requiring both high accuracy and efficient retrieval.}, archive = {J_EAAI}, author = {Shun Liu and Yanjun Zheng and Xianxian Zeng and Jun Yuan and Jiawen Li and Rongjun Chen}, doi = {10.1016/j.engappai.2025.110659}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110659}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical cascaded networks with multi-task balanced loss for fine-grained hashing}, volume = {152}, year = {2025}, }
• (2025). Feature-topology cascade perturbation for graph neural network. EAAI, 152, 110657. (www)

  @article{ , abstract = {Graph Neural Network (GNN) has gained great popularity in tackling various analytics tasks focusing on graph data. Data perturbation, particularly feature perturbation, as a promising solution for augmenting graph data, enables GNN to learn powerful representation. However, most feature perturbation strategies excessively emphasize the global perspective, which neglects the contributions of influential nodes from a local perspective. Additionally, the transformed topology corresponding to feature perturbation is insufficiently involved in building networks. To address these issues, we propose a novel plug-and-play architecture, termed Feature-Topology Cascade Perturbation (FTCP) for GNN, which consists of two perturbation stages: celebrity-guided feature perturbation and cascaded topology perturbation. Specifically, on the feature level, we perturb nodes by recognizing celebrities in view of multi-hop structure naturally existing in original topology. This is because figuring out celebrities would be of great help in the representation power of GNN. On the node relationship level, we further track the topology induced by perturbed features via a polarized view, which then assists the original topology to capture richer structure information. Extensive experiments conducted on both regular graph-structure and multi-view data illustrate that our architecture FTCP consistently yields performance improvement when applied to various GNN models.}, archive = {J_EAAI}, author = {Hui Cong and Xibei Yang and Keyu Liu and Qihang Guo}, doi = {10.1016/j.engappai.2025.110657}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110657}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature-topology cascade perturbation for graph neural network}, volume = {152}, year = {2025}, }
• (2025). Community detection in multi-layer networks by regularized debiased spectral clustering. EAAI, 152, 110627. (www)

  @article{ , abstract = {Community detection is a crucial problem in the analysis of multi-layer networks. While regularized spectral clustering methods using the classical regularized Laplacian matrix have shown great potential in handling sparse single-layer networks, to our knowledge, their potential in multi-layer network community detection remains unexplored. To address this gap, in this work, we introduce a new method, called regularized debiased sum of squared adjacency matrices (RDSoS), to detect communities in multi-layer networks. RDSoS is developed based on a novel regularized Laplacian matrix that regularizes the debiased sum of squared adjacency matrices. In contrast, the classical regularized Laplacian matrix typically regularizes the adjacency matrix of a single-layer network. Therefore, at a high level, our regularized Laplacian matrix extends the classical one to multi-layer networks. We establish the consistency property of RDSoS under the multi-layer stochastic block model (MLSBM) and further extend RDSoS and its theoretical results to the degree-corrected version of the MLSBM model. Additionally, we introduce a sum of squared adjacency matrices modularity (SoS-modularity) to measure the quality of community partitions in multi-layer networks and estimate the number of communities by maximizing this metric. Our methods offer promising applications for predicting gene functions, improving recommender systems, detecting medical insurance fraud, and facilitating link prediction. Experimental results demonstrate that our methods exhibit insensitivity to the selection of the regularizer, generally outperform state-of-the-art techniques, uncover the assortative property of real networks, and that our SoS-modularity provides a more accurate assessment of community quality compared to the average of the Newman-Girvan modularity across layers.}, archive = {J_EAAI}, author = {Huan Qing}, doi = {10.1016/j.engappai.2025.110627}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110627}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Community detection in multi-layer networks by regularized debiased spectral clustering}, volume = {152}, year = {2025}, }
• (2025). Enhancing the hydraulic engineering circular-18 pier scour equation with knowledge-guided symbolic regression and field data augmentation. EAAI, 152, 110614. (www)

  @article{ , abstract = {Bridge scour is a top cause of bridge failures. However, many mainstream pier scour design equations still perform less satisfactorily, i.e., excessive overestimation or even risky underestimation can occur during practice. To improve the robustness of the widely used Hydraulic Engineering Circular-18 (HEC-18) equation, this work proposes a computational framework empowered by expressive symbolic regression and assorted datasets. In particular, a knowledge-guided genetic algorithm compensates for a sedimentary gap between lab and field conditions, and a multi-objective optimization advocates a balance between safety and budget. Furthermore, the authors modify the field data to alleviate its uncertain maturity and employ the K-means clustering to select statistically representative training data based on hydraulic, sedimentary and structural input parameters. Evaluated by multiple independent lab and field datasets, the enhanced equation can offer more economical and consistent design suggestions compared to multiple popular benchmarks, if not equivalently competitively. In particular, when evaluated by unseen lab data, its relative error is 23.14% compared with 21.70 ∼ 45.24% of benchmarks. While as safe as the other equations, the proposed equation is at least 5 ∼ 37% closer to field measurements, alleviating over-estimation issues. When predicting new field data, the standard deviation of the proposed equation’s errors is about 0.07, indicating a competitively consistent prediction capability. A post-analysis also investigates the enhanced equation’s applicability and the value of embedded knowledge.}, archive = {J_EAAI}, author = {Wen Xiong and Qilin Yang and Zhao Chen and Rongzhao Zhang}, doi = {10.1016/j.engappai.2025.110614}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110614}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing the hydraulic engineering circular-18 pier scour equation with knowledge-guided symbolic regression and field data augmentation}, volume = {152}, year = {2025}, }
• (2025). Layer-wise feature refinement for accurate three-dimensional lane detection with enhanced bird’s eye view transformation. EAAI, 152, 110585. (www)

  @article{ , abstract = {Three-dimensional (3D) lane line detection from images is a fundamental yet challenging problem in autonomous driving, with existing methods often lacking scene robustness and computational efficiency. We propose a new method for 3D lane line detection using a simple and efficient view transformation module and layer-wise refined bird’s eye view (BEV) features. Our approach introduces a dual-branch perspective transformation module combining deformable convolution and perspective relationship modules to enhance robustness across diverse scenes. Additionally, we design a cross-attention-based view transformation module with spatial position encoding and BEV spatial query to improve detail learning and transformation effectiveness. Our method further refines BEV features layer by layer to fully exploit multi-level information. Experimental results on two datasets demonstrate the superiority of our approach, showing a significant increase in F1-score compared to existing methods.}, archive = {J_EAAI}, author = {Hao Ren and Mingwei Wang and Yanyang Deng and Wenping Li and Chen Liu}, doi = {10.1016/j.engappai.2025.110585}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110585}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Layer-wise feature refinement for accurate three-dimensional lane detection with enhanced bird’s eye view transformation}, volume = {152}, year = {2025}, }
• (2025). Candidate ratio guided proximal policy optimization. EAAI, 152, 110576. (www)

  @article{ , abstract = {Proximal Policy Optimization (PPO) is a well-studied policy search method that improves policies monotonically by enforcing the probability ratio to remain within a clipping range. The probability ratios used to measure the policy distance change dynamically during policy optimization. However, the clipping range remains fixed throughout the policy training stage. A fixed clipping range may lead to sample inefficiency or aggressive policy updates, negatively affecting policy performance. In this paper, we propose a candidate-ratio-guided Proximal Policy Optimization method with self-adaptive clipping ratios to design the clipping range for improved sample efficiency and monotonic policy improvement. The clipping ratio is adjusted based on the average candidate ratios derived from actions sampled around the policy suggested actions. Increasing the clipping ratio allows more collected data to be used for policy optimization, while decreasing it effectively enforces probability ratios within the clipping range for monotonic policy improvement. Experimental results on MuJoCo tasks show that our method achieves more stable performance compared to existing baselines.}, archive = {J_EAAI}, author = {Yao Li and XiaoYang Tan}, doi = {10.1016/j.engappai.2025.110576}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110576}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Candidate ratio guided proximal policy optimization}, volume = {152}, year = {2025}, }
• (2025). A novel malware detection method based on audit logs and graph neural network. EAAI, 152, 110524. (www)

  @article{ , abstract = {Malicious programs pose a significant threat to cyberspace security, making practical and low-cost malware detection a pressing need. To address this problem, we propose a novel malware detection method based on audit logs and graph neural networks. This method first performs fine-grained parsing of the logs for obtaining the process event sequence and process invocation relationship. Then, we employ a graph convolutional network to generate an embedding vector representation for each extracted process event, effectively capturing both local and global co-occurrence information. Next, the process structure and event semantic information are used to construct an event relationship graph for each log sample. Based on the event relationship graphs, we leverage an attention gated graph neural network (AGGNN) for malware detection. The evaluation shows that our approach can detect malware effectively with explainable results, and it outperforms the recent malware detection methods based on audit logs.}, archive = {J_EAAI}, author = {Yewei Zhen and Donghai Tian and Xiaohu Fu and Changzhen Hu}, doi = {10.1016/j.engappai.2025.110524}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110524}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel malware detection method based on audit logs and graph neural network}, volume = {152}, year = {2025}, }
• (2025). Neural hyper flow: Parameterized prediction for control systems. EAAI, 152, 110413. (www)

  @article{ , abstract = {Neural operators for systems of differential equations have been used as surrogates instead of potentially computationally expensive solution algorithms. However, the application of these operators is unsuitable as a model for control loop design. The main reason is that most operators consider a specific initial condition for each training. In particular, the control application needs to update its initial condition and recompute the control law. This paper aims to provide a novel description of the neural operator that could be seamlessly applied as a surrogate with different initial conditions. By utilizing hypernetworks and selecting appropriate parameters, this work can be applied to a future control application.}, archive = {J_EAAI}, author = {Mateus Mussi Brugnolli and Bruno Augusto Angélico}, doi = {10.1016/j.engappai.2025.110413}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110413}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural hyper flow: Parameterized prediction for control systems}, volume = {152}, year = {2025}, }
• (2025). A feasibility study of machine learning-based model predictive control for commercial buildings in cooling season. EAAI, 151, 110831. (www)

  @article{ , abstract = {This study explores the feasibility of model-based predictive control (MPC) with a machine learning (ML) approach in commercial buildings. Grey-box and ML-based models were developed using experimental data from a test facility. Three different models were considered for ML-based model development: artificial neural network, gauss process regression, and support vector regression. In MPC simulations, the optimal solution for the grey-box model was achieved by applying linear programming, assuming a linear time-invariant model. On the other hand, the proposed ML-based method utilized predefined setpoint trajectories to achieve cost savings by load shifting from on-to off-peak. The estimated trajectory yielding the minimum cost was identified as the optimal trajectory, which was then input to the grey-box model to ensure a fair comparison of the performance of both MPCs against that of optimal feedback control. An average saving performance of 27.5 % and 23.7 % was achieved using the MPC with grey-box and ML approach over optimal feedback control. Near-optimal performance was achieved with ML approach without running the optimization. The comparable performance of the proposed method implies that the engineering cost in a typical MPC using a grey-box model can be significantly reduced by using the ML method with minimal engineering, which is easy to implement and scalable to other buildings.}, archive = {J_EAAI}, author = {Abu Talib and Semi Park and Piljae Im and Jaewan Joe}, doi = {10.1016/j.engappai.2025.110831}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110831}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A feasibility study of machine learning-based model predictive control for commercial buildings in cooling season}, volume = {151}, year = {2025}, }
• (2025). Design of a novel multiband terahertz metamaterial absorber using a deep learning framework. EAAI, 151, 110801. (www)

  @article{ , abstract = {Current strategies for developing multiband electromagnetic metamaterial absorbers in the terahertz range face significant challenges, primarily because of their dependence on inefficient trial-and-error design methods. To address these limitations, in this work, we introduced a deep learning framework incorporating two bidirectional spectrum networks, and build the dataset through the inherent physical mechanisms of multiband terahertz absorbers. This framework established a bidirectional mapping relationship between the structural parameters and the electromagnetic responses. These networks predicted the reflection coefficient and absorption rate of electromagnetic metamaterials in the terahertz range based solely on known structural parameters through forward prediction paths. Analysis of the forward prediction results demonstrated that the deep learning model effectively captured the relationships between the structural parameters of electromagnetic metamaterials and their electromagnetic responses. Furthermore, we evaluated the capability of the inverse design path, which is the process of designing structural parameters to meet specific absorption rate requirements, for spectral curves with two, three, and four peaks generated randomly within the special frequency band. The designed electromagnetic metamaterial not only exhibited reasonable alignment with the peak absorption rates and the designated frequency bands of the target spectral curves but also involved relatively low structural complexity, being based on a prototype structural unit. This finding demonstrated the on-demand inverse design capability of the framework, which allowed for the tailored design of electromagnetic metamaterials with specific absorption peak numbers, thus highlighting its efficiency and flexibility in the design of multiband absorbers.}, archive = {J_EAAI}, author = {Yue Luo and Kepeng Qiu and Taqing Xu and Fuli Zhang and Peng Yue}, doi = {10.1016/j.engappai.2025.110801}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110801}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Design of a novel multiband terahertz metamaterial absorber using a deep learning framework}, volume = {151}, year = {2025}, }
• (2025). Environment adaptive deep reinforcement learning for intelligent fault diagnosis. EAAI, 151, 110783. (www)

  @article{ , abstract = {Aiming at the challenge of implementing trained deep reinforcement learning (DRL) model in unseen working environment, a novel DRL model with environmental adaptability is proposed for intelligent fault diagnosis of mechanical equipment. In this model, the weighted normalized mutual information feature selection method is first utilized to enhance the dependency between the fault states and the environment observations and to suppress redundancy among the environment observations. Subsequently, a domain adversarial generalization neural network is incorporated into dueling doubling deep Q network as an agent to automatically learn the domain-invariant association between relevant observations with fault states. It effectively enhances the generalization ability of DRL process in unknown environments. Finally, the quantitative reward matrix based on multi-condition fault pattern spacing and the environment adversarial reward of domain discrimination are designed to guide the agent be sensitive to hard-to-identify classes under variable working conditions, resulting in a parameterized policy which can be shared in unseen environments. Analytical results of two experiments show that the proposed method has higher diagnostic accuracy and better environmental adaptability than the existing popular reinforcement learning methods.}, archive = {J_EAAI}, author = {Xiaofeng Liu and Zheng Zhao and Fan Yang and Fuyuan Liang and Lin Bo}, doi = {10.1016/j.engappai.2025.110783}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110783}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Environment adaptive deep reinforcement learning for intelligent fault diagnosis}, volume = {151}, year = {2025}, }
• (2025). Tracking the weld seam under strong interference in laser-arc hybrid welding via a novel local-add U-net. EAAI, 151, 110778. (www)

  @article{ , abstract = {Strong inference from light, smoke and spatters often increases errors of seam tracking based on structured light vision in robotic welding. Deep learning methods including the semantic segmentation networks make progress in extracting laser strips from the interference accurately. However, previous studies did not consider the match between the semantic complexity and network structure. Besides, the tracking accuracy was generally focused while the computational costs were neglected. A novel local-add U-net (LAU-net) is developed in this work based on the original U-net to achieve both low tracking errors and low computational costs. The backbone of the U-net was first adjusted to match the semantic complexity of laser stripe images. Secondly, the weighted-loss function was adopted to solve the imbalance between positive and negative pixels. The weight factor was optimized to avoid degression of network performance on structured light datasets. Finally, a new local-add connect was proposed to make full use of local information in feature maps. The segmentation accuracy was significantly improved by the local-add connect at the price of only a 1.1 ms increase in computational latency. The seam tracking system based on the LAU-net possessed good adaptability and robustness. The average tracking error was controlled within 1.48 pixels and the tracking speed reached 45 fps, which was competitive compared with other traditional or deep-learning seam tracking methods.}, archive = {J_EAAI}, author = {Biao Yang and Guoqing Chen and Lei Tian and Caiwang Tan and Fuyun Liu and Laijun Wu and Bo Chen and Xiaoguo Song}, doi = {10.1016/j.engappai.2025.110778}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110778}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Tracking the weld seam under strong interference in laser-arc hybrid welding via a novel local-add U-net}, volume = {151}, year = {2025}, }
• (2025). A lightweight neural network search algorithm based on in-place distillation and performance prediction for hardware-aware optimization. EAAI, 151, 110775. (www)

  @article{ , abstract = {Due to the limited computing resources of edge devices, traditional object detection algorithms struggle to meet the efficiency and accuracy requirements of autonomous driving. Consequently, designing a neural network model that balances hardware resource requirements, operating speed, and accuracy is crucial. To address this, by integrating algorithm with hardware characteristics, we propose a lightweight neural network architecture search algorithm based on in-place distillation and performance predictor (LNIP). Initially, we focus on optimizing the operators of the you only look once version 8 nano (YOLOv8n) and dynamically adjust its network structure. Then, we trained a super-network using a progressive shrinking strategy, the sandwich rule, and in-place distillation. Subsequently, we employed a Gaussian process to model the relationship between network architecture and accuracy, utilizing encoding methods and custom kernel function to develop high-performance predictor. Finally, during the search process, we introduce a reward function based on Pareto optimality to balance the performance of the model with hardware constraints. Building upon this foundation, we design an efficient search algorithm based on the performance predictor to progressively explore the optimal network structure tailored to hardware characteristics. We compared our lightweight network with state-of-the-art methods on the BDD100K, COCO, and PASCAL VOC datasets and deployed it on the Black Sesame A1000 and NVIDIA Xavier for comprehensive evaluation. On the NVIDIA Xavier, the lightweight network achieves a latency of 11.81 ms and an edge precision of 46.1 %. These experimental results demonstrate that our method outperforms existing methods in balancing hardware constraints and model performance.}, archive = {J_EAAI}, author = {Siyuan Kang and Yinghao Sun and Shuguang Li and Yaozong Xu and Yuke Li and Guangjie Chen and Fei Xue}, doi = {10.1016/j.engappai.2025.110775}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110775}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight neural network search algorithm based on in-place distillation and performance prediction for hardware-aware optimization}, volume = {151}, year = {2025}, }
• (2025). Dynamic multi-objective optimization method for production index of cement clinker firing process based on collaborative prediction strategy. EAAI, 151, 110774. (www)

  @article{ , abstract = {The cement clinker firing system is complex, with interdependent indicators, making it challenging to optimize decision-making. Furthermore, the traditional static single-objective or multi-objective optimization methods are inadequate in adapting to the dynamic changes of complex working conditions. To address these issues, this paper proposes a dynamic multi-objective optimization method for the production index of the cement clinker firing process. The way first establishes a data-driven dynamic multi-objective optimization model, the content of free calcium oxide (f-CaO) in clinker and the coal consumption of the sintering system as optimization objectives, numerous operational indicators as decision variables, considers dynamic factors and constraints during manufacturing. Then a dynamic multi-objective evolutionary algorithm based on a collaborative prediction strategy (CPS-DMOEA) is designed to solve the model. Experimental results for some benchmark test problems show a significant improvement in dynamic optimization performance with CPS-DMOEA. Furthermore, experiments using actual cement production data show that the proposed method outperforms traditional static multi-objective optimization methods.}, archive = {J_EAAI}, author = {Gang Liu and Shangjian Xie and Xiaochen Hao and Mengke Yang and Xunian Yang and Xingxing Xu and Huan Guo}, doi = {10.1016/j.engappai.2025.110774}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110774}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic multi-objective optimization method for production index of cement clinker firing process based on collaborative prediction strategy}, volume = {151}, year = {2025}, }
• (2025). An integrated multi-scale context-aware network for efficient desnowing. EAAI, 151, 110769. (www)

  @article{ , abstract = {Adverse snow conditions, significantly degrade visual data quality, resulting in substantial performance drops in computer vision systems. We propose an advanced neural network architecture designed for desnowing tasks, leveraging multi-scale features and context-aware attention mechanisms. Our novel network integrates a series of enhancements including ghost encoders for efficient feature representation and extraction through the utilization of linear transformations, and the multi-scale transformer bottleneck which combines the atrous spatial pyramid pooling with a vision transformer in the bottleneck to strengthen multi-scale feature extraction and captures local and global dependencies. Additionally, the SimGC attention module combines the simple parameter-free attention and global context attention, to enhance local and global spatial features. Lastly, the feature enhancement module is used to refine the final reconstruction. Our experimental results demonstrate that the proposed network significantly outperforms state-of-the-art methods on certain metrics and confirms the robustness and accuracy of our approach.}, archive = {J_EAAI}, author = {Samuel Akwasi Agyemang and Haobin Shi and Xuan Nie and Nana Yaw Asabere}, doi = {10.1016/j.engappai.2025.110769}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110769}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated multi-scale context-aware network for efficient desnowing}, volume = {151}, year = {2025}, }
• (2025). Improved you only look once for weed detection in soybean field under complex background. EAAI, 151, 110762. (www)

  @article{ , abstract = {Precision agriculture technology is an essential means to improve the quality of crop production. The intelligent weeding robot is a new method of regulating agriculture. However, while balancing accuracy and model size, it is still a big challenge to solve the detection difficulties caused by the similar texture of soybean and weed leaves, mutual covering, light and shadow, and other factors. To address this, an improved You Only Look Once version 8 (YOLOv8n) is proposed in this paper. Firstly, the focusing diffusion pyramid network (FDPN) is used to optimize the structure and enhance the model detection ability. Then, a lightweight Contextual Continuous Convolution for FasterBlock with an efficient multiscale attention (C2f-F-EMA) structure is designed to simplify the model size and increase the model precision. At the same time, a lightweight detection head (LDH) is devised to cut the number of model parameters and make the model more lightweight. Finally, the Inner Complete Intersection over Union (Inner-CIoU) loss function is introduced to expedite the bounding box regression; it can improve the regression precision and training velocity. According to the article, the proposed algorithm performs well in weed classification, identification, and detection, and the precision, recall, and Mean Average Precision (mAP 0.5 ) are 94.5 %, 92.6 %, and 97.2 %. The proposed model showed the best overall performance for other mainstream weed recognition algorithms, ensuring identification accuracy and lightweight performance.}, archive = {J_EAAI}, author = {Wenxu Zhang and Xiaowei Shi and Minlan Jiang and Azhi Zhang and Lingguo Zeng and Mohammed A.A. Al-qaness}, doi = {10.1016/j.engappai.2025.110762}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110762}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improved you only look once for weed detection in soybean field under complex background}, volume = {151}, year = {2025}, }
• (2025). An enhanced generative adversarial network for longer vibration time data generation under variable operating conditions for imbalanced bearing fault diagnosis. EAAI, 151, 110760. (www)

  @article{ , abstract = {As a typical data augmentation method, a generative adversarial network is widely applied to solve data scarcity problems for imbalanced bearing fault diagnosis. However, these methods still face challenges in generating longer data due to the risk of mode collapse and instability during training. To address this issue, an enhanced generative adversarial network is proposed for generating longer vibration time data to improve imbalanced bearing fault diagnosis under variable operating conditions. Firstly, a dual cross-frequency attention block is integrated into the discriminator to adaptively extract intra-component and inter-component features across low and high frequency components decomposed using Wavelet, thereby facilitating generator to generate longer synthetic time data with higher frequency resolution. Furthermore, the sequence information block is introduced to generate synthetic time data under variable operating conditions by incorporating specific operating condition information into the generator. To expedite the synthetic data generation process under variable operating conditions, healthy data corresponding to these conditions are used as input for the generator, replacing random noise. Finally, the superiority of the proposed method is validated through experiments on two bearing datasets for imbalanced bearing fault diagnosis. Experimental results on these two datasets demonstrate that the proposed method with 2048-length synthetic data achieves the highest accuracy of 98.75 % and 96.88 %, respectively, outperforming state-of-the-art methods. Therefore, the proposed method can effectively address the challenge of generating longer vibration time data, improving diagnostic accuracy in imbalanced bearing fault diagnosis under variable operating conditions.}, archive = {J_EAAI}, author = {Teng Wang and Zhi Chao Ong and Shin Yee Khoo and Pei Yi Siow and Jinlai Zhang and Tao Wang}, doi = {10.1016/j.engappai.2025.110760}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110760}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An enhanced generative adversarial network for longer vibration time data generation under variable operating conditions for imbalanced bearing fault diagnosis}, volume = {151}, year = {2025}, }
• (2025). RCSD-UAV: An object detection dataset for unmanned aerial vehicles in realistic complex scenarios. EAAI, 151, 110748. (www)

  @article{ , abstract = {Unmanned Aerial Vehicle (UAV) detection based on visible light plays an important role in urban low-altitude defense, public safety and other fields. However, the current dataset is limited by single scene, large object and other factors deviated from the actual application scene, making it difficult to meet the needs of sample-driven deep learning optical image UAV detection. Therefore, this paper proposed a novel realistic complex scenarios UAV object dataset (RCSD-UAV) to provide training data for UAV detection models based on artificial intelligence technology. All data were obtained from ordinary cameras or mobile phones in the real world, covering various commonly used UAV types and natural scenes. The dataset is classified according to the scene and the object size, and we evaluated several models and gave benchmarks. From the experimental results, it can be concluded that the detection of UAVs is challenging due to small size and complex background. The two-stage model has good detection effect but poor real-time performance. The one-stage model can better balance the detection effect and real-time performance.}, archive = {J_EAAI}, author = {Wanxuan Geng and Junfan Yi and Ning Li and Chen Ji and Yu Cong and Liang Cheng}, doi = {10.1016/j.engappai.2025.110748}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110748}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {RCSD-UAV: An object detection dataset for unmanned aerial vehicles in realistic complex scenarios}, volume = {151}, year = {2025}, }
• (2025). A novel data-driven machine learning approach for improved strain rate control in thermomechanical testing of sheet metals. EAAI, 151, 110746. (www)

  @article{ , abstract = {Thermomechanical tests on sheet metals are commonly conducted using Gleeble systems to investigate their viscoplastic behaviour. Accurate strain rate control is crucial in these tests to accurately determine the material's thermomechanical properties. However, the well-known temperature gradient along the gauge length makes accurate strain rate control using the conventional approach challenging. In this study, to improve strain rate control in thermomechanical testing, a novel data-driven Machine Learning (ML) approach has been developed, for the first time, based on the analysis of the characteristics of the experimental data obtained using the conventional approach. This novel approach utilises Long Short-Term Memory (LSTM) networks, with inputs of target deformation temperature and strain rate, and outputs of strain distributions along the gauge length throughout the deformation process. The output strain distributions are subsequently integrated to calculate the time-dependent displacement required to stretch the specimen under the target conditions. Uniaxial tensile tests on an aluminium alloy under hot stamping conditions were conducted using both conventional and novel approaches for strain rate control. The experimental data obtained using the conventional approach were adopted to train and test the novel approach. The results show that this novel approach can significantly improve strain rate control in thermomechanical tests. Compared to the conventional approach which results in a Percentage Error (PE) in strain rate of up to ∼770% and a Mean Absolute Percentage Error (MAPE) in strain rate of up to ∼120%, the novel approach significantly reduces both PE and MAPE values by over 88%. This novel ML approach provides an effective solution for controlling strain rate in thermomechanical testing, enabling accurate determination of thermomechanical properties of sheet metals.}, archive = {J_EAAI}, author = {James Dear and Ruiqiang Zhang and Zhusheng Shi and Jianguo Lin}, doi = {10.1016/j.engappai.2025.110746}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110746}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel data-driven machine learning approach for improved strain rate control in thermomechanical testing of sheet metals}, volume = {151}, year = {2025}, }
• (2025). Semi-supervised deep learning framework based on modified pyramid scene parsing network for multi-label fine-grained classification and diagnosis of apple leaf diseases. EAAI, 151, 110743. (www)

  @article{ , abstract = {Apple leaf diseases have a significant impact on both yield and quality of apples, traditionally requiring specialists for detection. However, labor shortages in large orchards can reduce the detection efficiency. Current diagnosis models using semantic segmentation technology require a large amount of annotated data and encounter difficulties in extracting detailed features. This study introduces a semi-supervised method using a modified Pyramid Scene Parsing Network (PSPNet) for segmenting apple leaves. When only 1/2, 1/4, and 1/8 of the annotated data were used, the Mean Intersection over Union (MioU) indicators reached 0.975, 0.974, and 0.965, respectively. This enhanced method outperformed the original model in edge segmentation, showing semi-supervised model performance comparable to fully supervised models. Additionally, a transformer-based fine-grained multi-label framework was developed for classifying apple leaf diseases. The F1-scores (the harmonic mean of Precision and Recall) for Alternaria blotch, brown spot, grey spot, mosaic, and rust were 0.855, 0.903, 0.919, 0.921, and 0.895, respectively. This framework reduces computational complexity compared with conventional detection techniques. The combination of semi-supervised segmentation and multi-label classification provides support for the development of phenotyping platforms for precise pesticide application and disease-resistant variety selection.}, archive = {J_EAAI}, author = {Ke-Jun Fan and Bo-Yuan Liu and Wen-Hao Su and Yankun Peng}, doi = {10.1016/j.engappai.2025.110743}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110743}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised deep learning framework based on modified pyramid scene parsing network for multi-label fine-grained classification and diagnosis of apple leaf diseases}, volume = {151}, year = {2025}, }
• (2025). Bayesian-regularized cascaded neural networks for fractional asymmetric carbon-thermal nutrient-plankton dynamics under global warming and climatic perturbations. EAAI, 151, 110739. (www)

  @article{ , abstract = {In this study, design of cascaded feedforward neural network optimized with Bayesian Regularization algorithm is deployed to study the nonlinear fractional order system of carbon-thermal nutrient-plankton model asserting the profound ramifications of global warming and climatic alterations on the planktonic populations within the marine biosphere. The fractional differential asymmetric multifactor system is reflected by the carbon dioxide atmospheric concentration, atmospheric temperature, nutrient concentration and planktonic population spatiotemporal dynamics in the ocean. This system is exacted by an efficacious variant of the fractional Adams-Bashforth-Moulton predictor-corrector method across distinct scenarios and fractional orders. These numerically treated datasets are subsequently partitioned into training and testing subsets to model the intricate dynamics of the fractional Plankton system using the cascaded intelligent computational paradigm. The stupendous knacks of the cascaded network framework are evaluated on the sundry fractional nonlinear carbon-thermal nutrient-plankton model scenarios using mean squared error convergence pattern, input-error correlation and error autocorrelation statistics, regression metric, time series response curves, and error histogram studies. The efficacy of the designed cascaded feedforward networks modelling is substantiated by diminutive training mean square error discrepancies, spanning from 10 −11 to 10 −15 across all the complex nonlinear fractional plankton cases. Furthermore, comparative analysis and absolute error visualizations reveal that the designed neural strategy effectively delineates the carbon-thermal nutrient-plankton dynamics as evidenced by the errors spanning from 10 −05 to 10 −08 . These empirical results showcase that the neurocomputing strategy is robust in characterizing the intricate and complex nutrient-plankton dynamics in marine ecosystems under global warming and climatic perturbations.}, archive = {J_EAAI}, author = {Muhammad Junaid Ali Asif Raja and Adil Sultan and Chuan-Yu Chang and Chi-Min Shu and Adiqa Kausar Kiani and Muhammad Asif Zahoor Raja}, doi = {10.1016/j.engappai.2025.110739}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110739}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bayesian-regularized cascaded neural networks for fractional asymmetric carbon-thermal nutrient-plankton dynamics under global warming and climatic perturbations}, volume = {151}, year = {2025}, }
• (2025). Inverse design of non-parametric acoustic metamaterials via transfer-learned dual variational autoencoder with latent space-based data augmentation. EAAI, 151, 110735. (www)

  @article{ , abstract = {A ventilated acoustic resonator (VAR), a class of acoustic metamaterial, offers an effective noise reduction solution in urban environments, where ventilation is essential. The significant non-linearity between the VAR structure and the corresponding acoustic response poses substantial challenges to the inverse design of VAR through analytical approaches. Although deep learning has been employed to address the inverse design of such structures, conventional deep learning-based inverse design methods are constrained to parameter-defined structures, which exhibit limited design flexibility, thereby hindering the achievement of highly accurate inverse design. To address this challenge, we propose a novel inverse design framework for non-parametric VAR, composed of a dual-variational autoencoder (Dual-VAE) and iterative transfer learning method. Dual-VAE consists of structure variational autoencoder and acoustic response variational autoencoder with their latent space aligned to facilitate accurate inverse design of non-parametric VAR structures exhibiting the target acoustic response. The iterative transfer learning method is employed to enhance the inverse design performance of the Dual-VAE by progressively augmenting the initial training dataset, which consists solely of parametric VAR structures, with generated non-parametric VAR structures. The transfer-learned Dual-VAE demonstrated approximately a 32.27 % reduction in mean squared error with the target acoustic response compared to the Dual-VAE trained solely on the initial parametric VAR dataset. We present a novel approach to the inverse design of complex structures with high non-linearity by introducing the inverse design framework, demonstrating exceptional performance in generating non-parametric structures that achieve target performance.}, archive = {J_EAAI}, author = {Keon Ko and Min Woo Cho and Kyungjun Song and Dong Yong Park and Sang Min Park}, doi = {10.1016/j.engappai.2025.110735}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110735}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Inverse design of non-parametric acoustic metamaterials via transfer-learned dual variational autoencoder with latent space-based data augmentation}, volume = {151}, year = {2025}, }
• (2025). Susceptibility risk assessment of oil and gas pipeline geological hazards in mountainous areas based on data-driven model. EAAI, 151, 110732. (www)

  @article{ , abstract = {Geological hazards are recognized as causing significant damage to oil and gas pipelines, often resulting in catastrophic loss of life and property and hindering societal progress. In this study, a data-driven evaluation model is developed by integrating the Information Value method (IVM) with a Back Propagation Neural Network (BPNN) to assess the susceptibility of geological hazards in mountainous oil and gas pipelines. The IVM is used to identify non-hazardous areas, optimizing sample selection and reducing training errors, while the BPNN is employed to determine the weights of evaluation indicators, enhancing accuracy. First, an evaluation index system is proposed that comprehensively considers the natural geographical conditions and main disaster types. Next, non-disaster areas are located using the IVM and combined with disaster-prone areas to form the sample data. The sample data is then input into a BPNN for training, and the weights of each evaluation index are obtained from the trained network. Finally, a susceptibility risk assessment model is developed based on the derived weights and information values to accurately evaluate the susceptibility of pipeline geological hazards. A pipeline in China's Zhejiang Province's mountainous region is used as an illustration. Compared to the single IVM model and the single BPNN model, the receiver operator characteristic curve shows that the proposed method achieves significant improvements in the area under the curve by 9.8 % and 11.2 %, respectively, indicating a high level of evaluation accuracy. This study provides a reliable approach for assessing geological hazard susceptibility, offering scientific support for pipeline planning and hazard mitigation in oil and gas operations.}, archive = {J_EAAI}, author = {Yuxiang Yang and Benji Wang and Xiao Cen and Bowen Shao and Baikang Zhu and Jin Yang and Bingyuan Hong}, doi = {10.1016/j.engappai.2025.110732}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110732}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Susceptibility risk assessment of oil and gas pipeline geological hazards in mountainous areas based on data-driven model}, volume = {151}, year = {2025}, }
• (2025). Application of multi-criteria decision-making method based on improved grade Z-number in site selection of new energy vehicles charging stations. EAAI, 151, 110731. (www)

  @article{ , abstract = {Due to the imbalance of supply and demand in charging infrastructure, as well as unreasonable layout and other issues, the further development of New Energy Vehicles is restricted. However, many investment decision-makers lack effective theoretical guidance. Therefore, this paper proposes a Criteria Importance Though Intercrieria Correlation-Combinative Distance-based Assessment multi-criteria decision making framework based on an improved Grade Z-number to select the optimal charging station. Sometimes experts may not be able to provide an accurate value of Z-number, and it is more practical to require them to give evaluations in the form of grades, reflecting the reliability of the evaluations in various life scenarios. At the same time, due to the influence of an expert's personal educational background, knowledge, and personality, this paper overlook the individual differences among experts in our calculations, resulting in each expert being assigned the same decision-making weight, which does not reflect the complexity of actual decision-making. Hence, this paper introduces an objective method for calculating the weight of experts based on the evaluation information they provide, making the process more scientific and reasonable, and the final ranking results more convincing. Through comparative analysis, the proposed framework demonstrates its applicability to solving practical problems and its ability to produce reasonable and effective outcomes.}, archive = {J_EAAI}, author = {Jianping Fan and Xinyue Du and Meiqin Wu}, doi = {10.1016/j.engappai.2025.110731}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110731}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Application of multi-criteria decision-making method based on improved grade Z-number in site selection of new energy vehicles charging stations}, volume = {151}, year = {2025}, }
• (2025). From ensemble to knowledge distillation: Improving large-scale food recognition. EAAI, 151, 110727. (www)

  @article{ , abstract = {Food recognition on a large scale presents significant challenges due to high intra-category similarity and inter-category variability. Addressing these challenges is crucial for developing robust and accurate food recognition systems, which have applications in health monitoring, dietary assessment, and automated food logging. This study aims to tackle these issues by employing ensemble learning and knowledge distillation. We use ensemble learning to effectively combine the local perception capability of convolutional neural networks (CNNs) and the global modeling capability of Vision Transformers. The synergistic ensemble enhances the model's ability to discern subtle differences within categories and capture a spectrum of diverse patterns across various categories. To reduce the number of base models in an ensemble, we employed a method combining knowledge distillation and re-ensembling. Specifically, we used the collective knowledge of four base models to guide the re-learning process of student models. Subsequently, we re-ensembled these distilled models, significantly enhancing the recognition performance of the ensemble while maintaining the same computational efficiency. Finally, we fine-tuned the optimal ensemble weights to further boost the recognition performance of the ensemble model. We conducted extensive experiments on the large-scale food datasets Food2k and CNFood241, achieving state-of-the-art performance. Specifically, on the Food2k dataset, our method achieved a top-1 accuracy of 86.22 % with 131.56M parameters, outperforming the state-of-the-art algorithms by 2.1 %, demonstrating its effectiveness.}, archive = {J_EAAI}, author = {Liming Nong and Guohao Peng and Tianyang Xu and Jinlin Zhu}, doi = {10.1016/j.engappai.2025.110727}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110727}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {From ensemble to knowledge distillation: Improving large-scale food recognition}, volume = {151}, year = {2025}, }
• (2025). Advanced multi-layer deep learning model for accurate estimation of heat transfer and flow designing parameters across diverse dataset configurations. EAAI, 151, 110723. (www)

  @article{ , abstract = {This study introduces a hybrid deep learning (DL) model for accurately predicting key parameters of Advanced plate heat exchanger (APHE) performance, including the Nusselt number ( Nu ), friction factor (f), and thermal performance parameter (P). The hybrid deep learning (DL) model integrates multiple deep learning (DL) models—Feedforward Neural Network (FNN), Cascade Forward Neural Network (CFNN), and Fitting Network (FN)—in a multi-layered design to optimize predictive accuracy. Evaluations were conducted using five diverse datasets, each divided into three scenarios targeting Nu , f, and P, respectively, thereby capturing a broad range of thermal system behaviors. The experimental analysis reveals that the proposed hybrid DL model delivers outstanding predictive accuracy, as reflected in the following results. For Nu prediction, the Root Mean Square Error (RMSE) of the top models in the second layer ranged from 0.0003 to 0.00305 across various datasets. In the case of predicting f, the RMSE spanned from 0.0002 to 0.0041, while for predicting P, it varied between 0.0001 and 0.0004. Notably, the third layer retained CFNN as the most effective architecture, achieving remarkably low RMSE values of 8.1 × 10 −6 , 4.8 × 10 −6 , and 4.86 × 10 −6 for Nu , f, and P, respectively. Mean Absolute Error (MAE) and Nash-Sutcliffe Efficiency (NSE) metrics further validated the model's robustness, consistently yielding NSE scores near 0.9999. These models effectively minimized errors, particularly in high-variability regions. In the third layer, the hybrid DL framework selected CFNN as the optimal DL model for final predictions across all scenarios. This refinement achieved near-perfect accuracy, with Coefficient of Determination (R 2 ) values surpassing 0.9999 and RMSE values reduced to less than 0.00003, showcasing exceptional generalization and reliability. The hybrid DL model consistently outperformed standalone DL models by combining their strengths and leveraging a hierarchical architecture. These results highlight the potential of hybrid DL approaches in addressing complex regression tasks and advancing predictive modeling in engineering applications. This study provides a significant contribution to optimizing heat exchanger performance and demonstrates the utility of hybrid DL models in real-world thermal system challenges.}, archive = {J_EAAI}, author = {Ali A.H. Karah Bash and Ahmad Aboul Khail}, doi = {10.1016/j.engappai.2025.110723}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110723}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advanced multi-layer deep learning model for accurate estimation of heat transfer and flow designing parameters across diverse dataset configurations}, volume = {151}, year = {2025}, }
• (2025). EFR-ACENet: Small object detection for remote sensing images based on explicit feature reconstruction and adaptive context enhancement. EAAI, 151, 110722. (www)

  @article{ , abstract = {Small object detection in remote sensing images has been a major challenge in the research field of remote sensing technology. Traditional deep learning models perform well in detecting multiple objects, but lack in detecting small object detection, resulting in decreased detection accuracy. In order to reduce the information loss of small objects in object detection as well as to effectively manage and utilize different features in multiple receiver domains, this study proposes an architecture for a small object detection network for remote sensing images based on explicit feature reconstruction and adaptive context enhancement (EFR-ACENet). An explicit feature reconstruction module is designed to preserve feature details in the image and reduce the loss of small object information. Also, an adaptive context enhancement module is introduced, which is capable of detecting objects more comprehensively by integrating features such as contextual environment. And the features of perception fields of different sizes are extracted by the adaptive perception capability. We evaluate EFR-ACENet against ten state-of-the-art algorithms on multiple datasets such as the dataset for Tiny Object Detection in Aerial Images (AI-TOD) and the dataset for object Detection in Optical Remote sensing images (DIOR). The experimental results show that EFR-ACENet improves the average precision by3.3 % and 13.6 % on the AI-TOD and DIOR datasets respectively, which fully demonstrates its effectiveness in small object detection tasks. This result signals the great potential of EFR-ACENet for future airborne real-time applications.}, archive = {J_EAAI}, author = {Jingyu Ji and Yuefei Zhao and Aihua Li and Xiaolin Ma and Changlong Wang and Zhilong Lin}, doi = {10.1016/j.engappai.2025.110722}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110722}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {EFR-ACENet: Small object detection for remote sensing images based on explicit feature reconstruction and adaptive context enhancement}, volume = {151}, year = {2025}, }
• (2025). Fast and intelligent measurement of the ventilation resistance coefficient for the whole mine based on sparse measurement points. EAAI, 151, 110713. (www)

  @article{ , abstract = {Artificial intelligence is playing an important role in mine ventilation engineering, especially in ensuring safe mine production. The mine ventilation resistance coefficient (MVRC) is the core and basic parameter of a mine ventilation system. It is crucial to quickly and accurately obtain the ventilation resistance coefficient (VRC) of the whole mine for the scientific, safe, and intelligent management of the mine ventilation system. To solve the time-consuming and laborious problem of the traditional mine ventilation resistance measurement method, we propose a fast and intelligent measurement method to obtain the whole mine's VRC based on an artificial intelligence differential evolution algorithm and sparse measurement points. The VRC was experimentally measured to verify the validity of the intelligent measurement method and the reliability of the model. The relative error of the air volume at the observation points of the solved results is less than 6 %. The fast intelligent measurement of the MVRC of the Longshou mine was carried out. The results were applied to develop an emergency plan for addressing the insufficient air supply in the ventilation system caused by the collapse of the mine's main blind return shaft and validated through engineering practice. After field practice, the relative error between the predicted and tested air return volume of the 10-row inclined shaft was 3.23 %. It is verified that the results obtained using this method can solve mine ventilation system problems with relatively high accuracy, significantly reducing both the testing workload and time required for the mine ventilation resistance measurements.}, archive = {J_EAAI}, author = {Dong Wang and Jian Liu and Lijun Deng and Peng Cao and Li Liu}, doi = {10.1016/j.engappai.2025.110713}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110713}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast and intelligent measurement of the ventilation resistance coefficient for the whole mine based on sparse measurement points}, volume = {151}, year = {2025}, }
• (2025). A convolutional neural network based model for daily motor activities recognition: A contribution for parkinson's disease. EAAI, 151, 110706. (www)

  @article{ , abstract = {Daily motor activities are affected by motor disabilities caused by Parkinson's disease (PD). Monitoring motor disabilities frequently observed in PD is difficult for physicians, as they are limited to the information observed or self-reported during routine consultations, resulting in subjectivity and limited assessment. Thus, it is necessary to evaluate more frequently and objectively, ideally in a continuous manner including daily life tasks. While wearable sensory devices, such as inertial sensors, and their applications are steadily growing, the advances in artificial intelligence studies have revolutionized their ability to extract deeply hidden information for accurate detection and interpretation of motor activities. However, further studies are required, mainly focused on PD. This study aimed to implement a deep learning (DL) based model for recognizing daily motor activities based on inertial data to contribute to PD. The model relied on a convolutional neural network (CNN) architecture, trained and tested on a created custom dataset. We further benchmarked our model against other popular DL frameworks. The dataset included inertial data captured from 18 patients while performing trivial quotidian tasks, such as walking, turning, sitting, and lying. We hypothesized that a DL model based on a CNN architecture could be an appropriate solution for modeling daily motor non-steady and steady-state tasks from a single inertial sensor. We measured an F1 score of 0.906 and an accuracy of 91.1 % on final testing with our optimized CNN model, being the tasks of standing and walking the most accurately recognized by the model. Future challenges should cover exploring attention-based models and increasing the dataset.}, archive = {J_EAAI}, author = {Helena R. Gonçalves and Luís Martins and Nuno Ferrete Ribeiro and Luís Abreu and Ana Margarida Rodrigues and Cristina P. Santos}, doi = {10.1016/j.engappai.2025.110706}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110706}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A convolutional neural network based model for daily motor activities recognition: A contribution for parkinson's disease}, volume = {151}, year = {2025}, }
• (2025). An effective method for prospective scheduling of tasks in cloud-fog computing with an energy consumption management approach based on Q-learning. EAAI, 151, 110705. (www)

  @article{ , abstract = {The increasing energy consumption in cloud computing data centers has become a significant concern due to the expanding scale of computational demands. Efficient task scheduling is crucial to optimizing resource utilization while reducing operational costs and energy consumption. This study proposes a M ulti- A gent R einforcement L earning (MARL)-based scheduling framework that enhances system efficiency by dynamically allocating tasks based on environmental variations and workload fluctuations. Unlike conventional methods, MARL allows multiple intelligent agents to collaboratively optimize scheduling decisions, leading to superior adaptability and performance. The proposed approach consists of two steps: first, a centralized task dispatcher assigns incoming tasks to cloud servers using a queuing model. Second, an MARL-based scheduler on each server prioritizes and allocates tasks to virtual machines while continuously updating scheduling policies to maximize efficiency. The framework is evaluated using a CloudSim-based simulation environment to ensure a realistic and controlled assessment. Experimental results demonstrate that the proposed method reduces energy consumption by an average of 51.34 %, improves CPU utilization efficiency, and decreases response time by 44.35 % compared to traditional scheduling techniques, including First In-First Out (FIFO), Greedy, and Queue-based Scheduling (Q-sch). By leveraging MARL, the scheduler effectively minimizes waiting times and optimizes task completion rates, ensuring a balance between energy efficiency and system performance. This work highlights the advantages of reinforcement learning in cloud-fog computing and underscores its potential for intelligent resource management.}, archive = {J_EAAI}, author = {Yan Jin}, doi = {10.1016/j.engappai.2025.110705}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110705}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An effective method for prospective scheduling of tasks in cloud-fog computing with an energy consumption management approach based on Q-learning}, volume = {151}, year = {2025}, }
• (2025). An adaptive integrated learning-based virtual sensing framework for temperature prediction in aircraft brake monitoring engineering. EAAI, 151, 110704. (www)

  @article{ , abstract = {In commercial aviation, effective deceleration relies on accurate brake temperature trends. Traditional thermocouple sensors exhibit discontinuous responses, while machine learning-based methods often struggle with abnormal fluctuations due to high-dimensional signal overlap during low-speed braking. This study proposes an Adaptive Thermal Damping Integrated (ATDI) virtual sensor based on integrated learning to enhance brake temperature monitoring accuracy and address these challenges. A novel feature selection method is proposed that includes extracting baseline variables from five brake-related physical models and identifying significant latent variables using Shapley values, which are calculated by integrating three correlation indices based on linear, nonlinear, and rank consistency differences from aircraft flight records. The ATDI framework employs a time-lagged loss function and self-attention mechanism for dynamic weight assignment in the feature space, capturing critical temporal correlations to determine the thermal trend. A digital approach, adaptive to aircraft kinetic energy and called virtual thermal damping, is incorporated at the end of the prediction to mitigate fluctuations and ensure compliance with thermal conduction principles in braking systems. Experimental results demonstrate that the ATDI framework outperforms comparison networks in temperature value continuity and anomaly fluctuation suppression, especially with Long Short-Term Memory (LSTM) as the meta-learner, across three evaluation metrics. Additionally, application strategies for the ATDI framework in aircraft operation and maintenance are proposed.}, archive = {J_EAAI}, author = {Lin Lin and Yin Chen and Song Fu and Hao Zhang and Jinlei Wu}, doi = {10.1016/j.engappai.2025.110704}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110704}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive integrated learning-based virtual sensing framework for temperature prediction in aircraft brake monitoring engineering}, volume = {151}, year = {2025}, }
• (2025). A universal parameter-efficient fine-tuning approach for stereo image super-resolution. EAAI, 151, 110703. (www)

  @article{ , abstract = {Despite advances in the use of the strategy of pre-training then fine-tuning in low-level vision tasks, the increasing size of models presents significant challenges for this paradigm, particularly in terms of training time and memory consumption. In addition, unsatisfactory results may occur when pre-trained single-image models are directly applied to a multi-image domain. In this paper, we propose an efficient method for transferring a pre-trained single-image super-resolution transformer network to the domain of stereo image super-resolution (SteISR) using a parameter-efficient fine-tuning approach. Specifically, the concept of stereo adapters and spatial adapters are introduced, which are incorporated into the pre-trained single-image super-resolution transformer network. Subsequently, only the inserted adapters are trained on stereo datasets. Compared with the classical full fine-tuning paradigm, our method can effectively reduce training time and memory consumption by 57% and 15%, respectively. Moreover, this method allows us to train only 4.8% of the original model parameters, achieving state-of-the-art performance on four commonly used SteISR benchmarks. This technology is expected to improve stereo image resolution in various fields such as medical imaging and autonomous driving, thereby indirectly enhancing the accuracy of depth estimation and object recognition tasks.}, archive = {J_EAAI}, author = {Yuanbo Zhou and Yuyang Xue and Xinlin Zhang and Wei Deng and Tao Wang and Tao Tan and Qinquan Gao and Tong Tong}, doi = {10.1016/j.engappai.2025.110703}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110703}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A universal parameter-efficient fine-tuning approach for stereo image super-resolution}, volume = {151}, year = {2025}, }
• (2025). Deep reinforcement learning for optimal design of compliant mechanisms based on digitized cell structures. EAAI, 151, 110702. (www)

  @article{ , abstract = {Metamaterial mechanisms are micro-architectured compliant structures that operate through the elastic deformation of specially designed flexible members. This study introduces an efficient design methodology for compliant metamaterial mechanisms using deep reinforcement learning (RL). In this approach, design domains are digitized into finite cells with various hinge connections, reformulating the design problem as a combinatorial optimization problem. To tackle this intricated optimization problem, we unfold the domain to transform the design problem into a Markov decision process where the deformation behaviors of the designed compliant mechanisms are computed through finite element analysis (FEA). The digitized cell structures are modeled using 1-dimensional (1D) beam elements, significantly reducing the computational load of FEA. The FEA results are utilized in the deep RL framework to optimize compliant mechanism designs based on specific functional requirements. This methodology is applied to the design of compliant gripper and door-latch mechanisms, exploring the effects of cell tiling direction and penalization strategies for disconnected hinges. The optimized designs generated by deep RL outperform human-guided designs, achieving a 56.3% improvement in rotational compliance for the gripper mechanism and a 2.7-fold improvement in linear compliance for the door-latch mechanism, compared to human-guided designs. The optimized compliant mechanisms are fabricated using additive manufacturing, and their performance as compliant mechanisms is experimentally validated. These findings highlight the potential of RL-based design optimization using digitized cell structures, demonstrating its capability to efficiently design high-performance compliant metamaterial mechanisms while maintaining computational efficiency.}, archive = {J_EAAI}, author = {Yejun Choi and Yeoneung Kim and Keun Park}, doi = {10.1016/j.engappai.2025.110702}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110702}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep reinforcement learning for optimal design of compliant mechanisms based on digitized cell structures}, volume = {151}, year = {2025}, }
• (2025). A dual path neural network based on spatial correlation and feature weighting for predicting remaining useful life of aero-engine. EAAI, 151, 110700. (www)

  @article{ , abstract = {Accurate prediction of the remaining useful life (RUL) of aero-engine is crucial for flight safety, the reduction of maintenance costs and the extension of engine service life. Despite that deep learning methods have achieved favorable outcomes in the field of RUL prediction in recent years, the spatial correlations among feature information have not been adequately extracted and insufficient attention has been given to the variations in the importance of different input features across various time steps. To tackle these problems, a novel hybrid dual path network prediction method based on Multi-scale Convolution Neural Network (MS-CNN) and Gated Recurrent Unit (GRU) is proposed in this article. On the one hand, an innovative high-dimensional feature correlation extraction method is developed, multiple graph structures are constructed to capture complex spatial relationships among different feature maps formed by MS-CNN. On the other hand, under the consideration of the impact of the leaning of previous data in GRU on subsequent input features, feature attention mechanism is integrated with the hidden state information in GRU network to achieve differential update of input features at different time steps. The effectiveness of proposed model is validated on two well-known datasets and compared to other methods with similar neural network structures. And the experimental results show that the proposed framework could achieve more accurate RUL prediction than existing methods, especially under complex conditions and multiple fault modes.}, archive = {J_EAAI}, author = {Mi Zhou and Wanlin Liu and Ba-Yi Cheng}, doi = {10.1016/j.engappai.2025.110700}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110700}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual path neural network based on spatial correlation and feature weighting for predicting remaining useful life of aero-engine}, volume = {151}, year = {2025}, }
• (2025). An online training strategy based on the transfer learning concept for treating contaminants in surface electromyography. EAAI, 151, 110696. (www)

  @article{ , abstract = {Movement classification systems based on surface electromyography (sEMG) signals are sensitive to noise and other artifacts. Reducing noise and artifacts is critical to achieving efficient movement classification systems. However, existing methods for identifying and/or minimizing specific contaminants typically require offline training, which can be limiting. In response, our paper proposes a method based on Transfer Learning for the online training of the contaminant identification algorithm proposed in our previous work (Actor-Critic Reinforcement Learning model with a Fuzzy Inference System (FIS)-based reward function - FIS-ACRL). This method, which involves training an initial standard model for the agent and the FIS using data from some database volunteers, serves as the starting point for the online training of other subjects. The proposed algorithm was tested in the pre-processing stage of a movement classification system with an sEMG signal contaminated with four interferences (motion artifact, electrocardiography, powerline interference, and additive white Gaussian noise). It achieved results close to the offline implementation of the method (optimal condition) with a difference of only 1.5 % in the median accuracy of the gesture classifier. Importantly, for most channels, only five episodes (or less) of online training were needed for the accuracy to match that of offline learning. The experiments presented in this work underscore the potential of our method to overcome the challenges that are fundamental for the applicability of online learning, offering a promising avenue for future research and development.}, archive = {J_EAAI}, author = {Maurício Cagliari Tosin and Tiago Oliveira Weber and Alexandre Balbinot}, doi = {10.1016/j.engappai.2025.110696}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110696}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An online training strategy based on the transfer learning concept for treating contaminants in surface electromyography}, volume = {151}, year = {2025}, }
• (2025). Reinforcement learning-dijkstra-genetic algorithm for debris removal problem under different scenarios: An earthquake case study. EAAI, 151, 110693. (www)

  @article{ , abstract = {Road reconstruction during post-earthquake crises is vital for search operations, relief efforts, the movement of individuals from affected areas, and the sending of food and relief items. Blockage of the roads may prevent the delivery of critical goods or cause a lack of cars and ambulances access to the affected areas to evacuate injured people. In many studies, road debris removal is a sophisticated issue, so only a few roads were considered or aimed at finding a route through which all affected areas can be visited. Since roads have limited capacity for vehicle traffic during severe earthquakes, a lack of full debris removal may lead to delayed movement of injured people from affected areas, and debris left after the crisis may result in environmental and psychological harm to affected people. To solve this problem, a two-step model-free approach based on reinforcement learning methods is developed for full debris removal from damaged roads. In the first step, the damaged area is initially assessed using an unmanned aerial vehicle (UAV) and a state-action-reward-state-action (SARSA) algorithm to estimate the damage. In the second step, debris removal groups remove debris from all blocked roads based on the information reported by the UAV. For this purpose, a heuristic algorithm is developed based on debris removal groups, the divergence of the genetic algorithm is ensured and the agent training matrix is regulated using the neural network. The proposed solution methods are examined under three scenarios of severe, moderate, and weak earthquakes in Rudbar city as an artificial intelligence application. The results indicate that the debris removal operation is completed in less than 80 h in the severe case if the proposed method is used, whereas the algorithm achieves the proper solution in less than 45 min.}, archive = {J_EAAI}, author = {Fatemeh Kheildar and Parvaneh Samouei and Muharram Mansoorizadeh}, doi = {10.1016/j.engappai.2025.110693}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110693}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reinforcement learning-dijkstra-genetic algorithm for debris removal problem under different scenarios: An earthquake case study}, volume = {151}, year = {2025}, }
• (2025). An incorporation of metaheuristic algorithm and two-stage deep learnings for fault classified framework for diesel generator maintenance. EAAI, 151, 110688. (www)

  @article{ , abstract = {Diesel generators play a vital role in providing reliable power to ensure uninterrupted power supply. However, effective fault classification in these systems is challenging due to their complexity. This paper proposes a two-stage framework that combines deep learning with metaheuristic optimization for fault classification of diesel generators in Artificial Internet of Things (AIoT) systems. The first stage involves employing a Long short-term memory convolutional neural network (LSTM-CNN) model for accurate feature extraction and fault detection. The improved particle swarm optimization (IPSO) algorithm is employed to optimize the hyperparameters of the LSTM-CNN model, resulting in an enhanced IPSO-LSTM-CNN framework. A comprehensive performance evaluation is conducted by comparing the developed algorithm with other models, including recurrent neural networks (RNN), CNN, gated recurrent units (GRU), LSTM, and CNN-LSTM. The IPSO-LSTM-CNN obtains the most significant gains of many evaluation benchmarks when compared to other state-of-the-art algorithms. In terms of fault classification accuracy and robustness, the created model performs better than the alternative methods, confirming its usefulness in enhancing the operational efficiency and dependability of diesel generators in AIoT frameworks. This research provides a completed IPSO-LSTM-CNN framework in AI application for failure diagnosis of industrial machinery in maintenance service.}, archive = {J_EAAI}, author = {Thanh-Phuong Nguyen}, doi = {10.1016/j.engappai.2025.110688}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110688}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An incorporation of metaheuristic algorithm and two-stage deep learnings for fault classified framework for diesel generator maintenance}, volume = {151}, year = {2025}, }
• (2025). Multiple aerial/ground vehicles coordinated spraying using reinforcement learning. EAAI, 151, 110686. (www)

  @article{ , abstract = {Investments in unmanned aerial vehicles (UAVs) have recently surged in precision agriculture. However, multi-UAV missions can face limitations due to weather conditions, highlighting the need for effective spray coverage. A novel system tailored for spraying in windy conditions to tackle this challenge is proposed. Instead of directly controlling sprayed drops, the location of spraying UAVs based on real-time wind data is adjusted. Our proposed methodology consists of three stages: Firstly, on-policy reinforcement learning (RL) with Proximal Policy Optimization (PPO) is utilized to optimize path planning. In the second stage, another PPO iteration to correct wind drift is employed, leveraging the latest wind data to enhance spray mission efficiency. Lastly, a novel algorithm is introduced to improve efficiency in narrow areas by substituting unmanned aerial vehicles with unmanned ground vehicles. To evaluate the efficiency of the proposed aerial spraying system, we conducted a simulation and reported the corresponding results.}, archive = {J_EAAI}, author = {Ali Moltajaei Farid and Jafar Roshanian and Malek Mouhoub}, doi = {10.1016/j.engappai.2025.110686}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110686}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multiple aerial/ground vehicles coordinated spraying using reinforcement learning}, volume = {151}, year = {2025}, }
• (2025). Understanding of air entrainment process in the sewage drop structure using physically-based symbolic intelligence models. EAAI, 151, 110684. (www)

  @article{ , abstract = {The vortex drop structure is employed to transport water through subterranean channels in urban sewer and drainage networks. As the water descends, it incorporates a significant amount of air, which is subsequently discharged from the drop shaft further downstream. Quantifying the volume of entrained air is generally a challenging task. A physical prototype was built as part of this study to comprehend the process of air movement within a vortex structure. The researchers conducted experiments to examine the effects of different variables on the air circulation. The outcomes of the dimensional analysis revealed that certain effective factors, such as the approach flow Froude number ( F r), the ratio of the total drop height to the diameter of the shaft ( L / D ), and the ratio of the sump depth to the diameter of the shaft ( H s / D ), had a substantial impact on the relative air discharge ( β ). In order to understand effects of independent variables on the relative air discharge ( β ), 144 experimental observations were used to develop four robust symbolic intelligence models (i.e., Model Tree [MT], Evolutionary Polynomial Regression [EPR], Multivariate Adaptive Regression Spline [MARS], and Gene-Expression Programming [GEP]) in order to provide with mathematical expressions for evaluation of vortex structures under operation. The performance of symbolic intelligence models were evaluated by using various setting parameters of the intelligent models. In addition, the results demonstrated MT (Index of Agreement [IOA] = 0.9713, Root Mean Square Error [RMSE] = 0.0160, and Scatter Index [SI] = 0.1094) provided the most accurate prediction of the relative air discharge and followed by MARS (IOA = 0.9561, RMSE = 0.0199, and SI = 0.1357), EPR (IOA = 0.9455, RMSE = 0.0221, and SI = 0.1513), and GEP (IOA = 0.9145, RMSE = 0.0277, and SI = 0.1885). Moreover, the performance of empirical equation, given by Central Composite Face-entered Design (RSM-CCFD) methodology, demonstrated the minimum level of precision (IOA = 0.8931, RMSE = 0.0310, and SI = 2016) in comparison with all symbolic intelligence models. Furthermore, as found in the experimental observations, the performance of mathematical expressions proved that β ratio displayed an upward trend as both Fr and L / D ratio increased.}, archive = {J_EAAI}, author = {Mohammad Najafzadeh and Mohammad Mahmoudi-Rad}, doi = {10.1016/j.engappai.2025.110684}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110684}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Understanding of air entrainment process in the sewage drop structure using physically-based symbolic intelligence models}, volume = {151}, year = {2025}, }
• (2025). Navigation path extraction for farmland headlands via red-green-blue and depth multimodal fusion based on an improved DeepLabv3+ model. EAAI, 151, 110681. (www)

  @article{ , abstract = {Accurately detecting farmland boundary lines is crucial for performing cross-row operations with autonomous agricultural vehicles. In dryland farming environments, complex and variable conditions, including weed interference, lighting variations, and difficult-to-identify headland area field ridges pose significant challenges for accurately detecting headland boundaries. Although the existing detection methods demonstrate good performance across various scenarios, considering the complex variations found in dryland farming environments could improve them. Therefore, an efficient and accurate DeepLabv3+-based headland boundary detection algorithm is developed. The method employs a lightweight MobileNet architecture, using depth information as auxiliary features to better perceive terrain variations. A multihead dual-feature attention mechanism (M-DAFM) adaptively fuses red-green-blue (RGB) and depth information. An improved dense residual atrous spatial pyramid pooling (DR-ASPP) structure provides enhanced feature extraction capabilities while maintaining computational efficiency. Multistage processing-based fusion and polynomial fitting are subsequently used to extract boundary lines from headland contours, and multiorder Bézier curves integrated with boundary curvature variation points are used to calculate autonomous turning paths for agricultural robots. Experimental results demonstrate that the improved DeepLabv3+ model achieves a mean intersection-over-union (mIoU) value of 93.9 % in headland boundary detection tasks with an inference speed of 23.1 frames per second (FPS). Final navigation experiments confirm that the robot can successfully perform autonomous cross-row turning operations with a 0.09- meter maximum deviation, demonstrating precise navigation control in complex agricultural environments. The proposed approach advances the theoretical agricultural robot navigation framework and provides a practical solution for implementing challenging autonomous cross-row operations in field conditions.}, archive = {J_EAAI}, author = {Tianlun Wu and Hui Guo and Wen Zhou and Guomin Gao and Xiang Wang and Chuntian Yang}, doi = {10.1016/j.engappai.2025.110681}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110681}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Navigation path extraction for farmland headlands via red-green-blue and depth multimodal fusion based on an improved DeepLabv3+ model}, volume = {151}, year = {2025}, }
• (2025). Peridynamic-driven feature-enhanced vision transformer for predicting defects and heterogeneous materials locations: Applications of deep learning in inverse problems. EAAI, 151, 110677. (www)

  @article{ , abstract = {To achieve efficient and real-time evaluation of internal compositions in heterogeneous structures, this study proposes a peridynamic-driven feature-enhanced deep learning (DL) framework by integrating the Context Anchored Attention (CAA) mechanism and Kolmogorov-Arnold Networks (KANs) into a Vision Transformer (CK-ViT) architecture. This framework effectively captures deep structural features while leveraging a self-attention DropKey regularization technique to improve generalization. Three types of heterogeneous structure datasets were constructed based on peridynamic theory, enabling transfer learning with the CK-ViT model. Comprehensive evaluations, including epoch-wise and multi-metric assessments, demonstrate that CK-ViT achieves superior training efficiency and inference accuracy compared to conventional models. Specifically, the coefficient of determination ( R 2 ) reaches 88.23 %, 93.19 %, and 90.69 %, while the accuracy achieves 97.99 %, 98.72 %, and 98.22 %, respectively, for the three problem types. These results indicate that CK-ViT can accurately predict the locations of defects and heterogeneous materials by leveraging the global strain distribution of structures. Ablation studies further confirm that the introduction of CAA and KANs enhances model performance by 7.8 % and decreases the parameter storage overhead by 51.0 %, underscoring their effectiveness. The Gradient-weighted Class Activation Mapping (Grad-CAM) technique was employed to conduct an in-depth analysis of the inference process and decision-making mechanisms, revealing the underlying principles that drive model predictions. Additionally, an error analysis was conducted to identify key sources of prediction uncertainty and potential optimization directions. This study establishes a robust and interpretable framework for real-time defect location prediction and material distribution estimation. It provides valuable references for structural health monitoring and material engineering.}, archive = {J_EAAI}, author = {Yalin Li and Jinwei Guan and Li Guo}, doi = {10.1016/j.engappai.2025.110677}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110677}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Peridynamic-driven feature-enhanced vision transformer for predicting defects and heterogeneous materials locations: Applications of deep learning in inverse problems}, volume = {151}, year = {2025}, }
• (2025). Continuous reinforcement learning via advantage value difference reward shaping: A proximal policy optimization perspective. EAAI, 151, 110676. (www)

  @article{ , abstract = {Deep reinforcement learning has shown great promise in industrial applications. However, these algorithms suffer from low learning efficiency because of sparse reward signals in continuous control tasks. Reward shaping addresses this issue by transforming sparse rewards into more informative signals, but some designs that rely on domain experts or heuristic rules can introduce cognitive biases, leading to suboptimal solutions. To overcome this challenge, this paper proposes the advantage value difference (AVD), a generalized potential-based end-to-end exploration reward function. The main contribution of this paper is to improve the agent’s exploration efficiency, accelerate the learning process, and prevent premature convergence to local optima. The method leverages the temporal difference error to estimate the potential of states and uses the advantage function to guide the learning process toward more effective strategies. In the context of engineering applications, this paper proves the superiority of AVD in continuous control tasks within the multi-joint dynamics with contact (MuJoCo) environment. Specifically, the proposed method achieves an average increase of 23.5% in episode rewards for the Hopper, Swimmer, and Humanoid tasks compared with the state-of-the-art approaches. The results demonstrate the significant improvement in learning efficiency achieved by AVD for industrial robotic systems.}, archive = {J_EAAI}, author = {Jiawei Lin and Xuekai Wei and Weizhi Xian and Jielu Yan and Leong Hou U and Yong Feng and Zhaowei Shang and Mingliang Zhou}, doi = {10.1016/j.engappai.2025.110676}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110676}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Continuous reinforcement learning via advantage value difference reward shaping: A proximal policy optimization perspective}, volume = {151}, year = {2025}, }
• (2025). Personalised healthcare and exercise rehabilitation based on upper-limb metrics. EAAI, 151, 110673. (www)

  @article{ , abstract = {Rapid and accurate upper-limb motion analysis and metrics are essential for enhancing exercise medicine knowledge graph to drive personalised medicine. However, current studies face limitations in understanding multi-timescale and multi-target, compounded by discrepancies in physiological structure. This study proposes Upper-Limb Dynamic Warping (UP-Ldw), which effectively responds to motion discrepancies and adapts to the physiological characteristics. UP-Ldw constructs a geometric model by parameterizing features and incorporating physiological structure parameters to accommodate individual variations. Dynamic temporal regularization is integrated to accommodate motion sequences across multiple time scales. Ultimately, the motion similarity among various targets is outputted to facilitate comparison and metrics. Furthermore, two datasets are developed: Upper-Limb 3-Dimensional Dataset (UP-L-3D), and Upper-Limb Geometric Modeling Dataset (UP-L-GM), both utilized for validation. Comparison experiments employed convolutional neural network (CNN), principal component analysis (PCA), and random forests. Results demonstrate that UP-Ldw achieves the highest accuracy of 97.92 % using metrics 20 as the discriminant criterion, with a short running time of 1–8 ms. UP-Ldw aligned with CNN confusion matrices and the PCA downscaling, validating its precise motion analysis. The Random Forest model attained an average accuracy of 91.1 %, confirming the validity of the geometric model. A generalization experiment was conducted using the public dataset Arm-CODA, further validating UP-Ldw's ability to adapt to physiological structures and effectively metricize upper-limb motion. Overall, UP-Ldw employs artificial intelligence to metricize motion, facilitating mirror rehabilitation. This advancement contributes significantly to the engineering applications of personalised healthcare and exercise rehabilitation.}, archive = {J_EAAI}, author = {Honggang Wang and Yisu Wang and Zengmin He and Xuzhi Li and Yufeng Yao}, doi = {10.1016/j.engappai.2025.110673}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110673}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Personalised healthcare and exercise rehabilitation based on upper-limb metrics}, volume = {151}, year = {2025}, }
• (2025). Cancer–drug response prediction via feature aggregation and association graph learning. EAAI, 151, 110671. (www)

  @article{ , abstract = {Predicting cancer–drug response (CDR) not only guides the development of anticancer drugs but also provides personalized treatment plans for cancer patients, driving the advancement of precision medicine. Currently, numerous computational methods are utilized for CDR prediction and achieve great achievements. However, there are still some issues need to be addressed. From the aspect of data using, many studies primarily use single nucleotide variations (SNV) as features of cell lines, yet their acquisition from clinical data is often incomplete and inefficient, making it challenging to extend SNV-based CDR models from in vitro cell lines to clinical data. From the aspect of modeling, most works focus excessively on learning the intrinsic characteristics of drugs and cell lines, while neglecting to fully capture their interaction information and network-level knowledge. In this paper, we introduce a feature aggregation and association graph learning network (F A 2 GL) for CDR, which aggregates intrinsic and interaction feature representations of drugs and cell lines, and obtains their network-level knowledge from the association graph. In this network, F A 2 GL initially employs the self-attention block (SAB) to capture drug intrinsic features, with a focus on the relationships between substructures within the same drug, and utilizes the multilayer perceptron (MLP) to obtain cell line intrinsic features. Subsequently, it constructs the cross-attention block (CAB) to extract interaction information between drugs and cell lines, followed by the utilization of a fusion coefficient for aggregating these features. Additionally, F A 2 GL also constructs an association graph by introducing pipe nodes and similarity networks, bridging the feature gap between drugs and cell lines and enhancing predictions of similar substances, respectively. Then, a graph neural network (GNN) is applied to this graph to capture their network-level information. By integrating fine-grained and network-level features, F A 2 GL demonstrates competitiveness compared to state-of-the-art methods on Genomics of Drug Sensitivity in Cancer (GDSC) dataset which contains various cancers such as acute lymphoblastic leukemia, diffuse large B-Cell lymphoma, and multiple myeloma, while avoiding the use of large-scale SNV data. The codes and data for F A 2 GL are available at https://github.com/Dtdtxuky/FA-2GL/tree/main .}, archive = {J_EAAI}, author = {Kaiyi Xu and Minhui Wang and Xin Zou and Chengfu Ji and Hua Zhou and Chang Tang}, doi = {10.1016/j.engappai.2025.110671}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110671}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cancer–drug response prediction via feature aggregation and association graph learning}, volume = {151}, year = {2025}, }
• (2025). Fault diagnosis of real-scenario battery systems in electric vehicles using the multi-scale fuzzy entropy algorithm. EAAI, 151, 110670. (www)

  @article{ , abstract = {The entropy algorithm, as a widely used and efficient fault diagnosis technique, holds significant potential for ensuring battery safety in electric vehicles. This study investigates the effectiveness of Shannon entropy and multi-scale fuzzy entropy in battery safety applications, with a focus on verifying the reliability, sensitivity, and robustness of fuzzy entropy. By comparing various battery cell inconsistency evaluation indices, the study highlights the superior application value of fuzzy entropy in battery fault diagnosis. Furthermore, leveraging its advantages in inconsistency evaluation, fuzzy entropy is employed to analyze long-term operational data from actual vehicles. Based on these findings, a robust and generalizable multi-level diagnosis strategy is proposed, validated using real vehicle fault data. Notably, this work introduces a novel entropy algorithm tailored for electric vehicle applications, offering new insights and development directions for entropy-based fault diagnosis methods. This study bridges artificial intelligence and engineering by demonstrating the practical application of the proposed method to enhance battery safety in electric vehicles.}, archive = {J_EAAI}, author = {Fengwei Liang and Jichao Hong and Xinyang Zhang}, doi = {10.1016/j.engappai.2025.110670}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110670}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis of real-scenario battery systems in electric vehicles using the multi-scale fuzzy entropy algorithm}, volume = {151}, year = {2025}, }
• (2025). Computer-aided detection systems based on ballistocardiography signals: A review. EAAI, 151, 110669. (www)

  @article{ , abstract = {This comprehensive review paper seeks to provide an in-depth survey of the technologies and methodologies employed in decision support systems for ballistocardiography. The paper extensively covers the biometric information embedded in the measured ballistocardiography signals. The presented exploration of ballistocardiography covers various pivotal stages, including signal measurement techniques, pre-processing methodologies, feature extraction approaches, classification techniques, and evaluation methods. Within the scope of this study, a systematic review has been performed, bringing together notable strategies employed in ballistocardiography from its initial stages to its current state of advancement. The efficacy of these systems in estimating ballistocardiography-based biometrics has demonstrated noteworthy proximity to acceptable levels. The utilization of ballistocardiography signals holds significant promise as an evolving field of research. This paper concludes by addressing the limitations inherent in the current state of research, outlining the potential directions for future investigations and real-world applications, and discussing the crucial aspect of explainability, which represents one of the new trends in computer-aided detection requirements.}, archive = {J_EAAI}, author = {Dalibor Cimr and Damian Busovsky and Hamido Fujita and Filip Studnicka and Richard Cimler}, doi = {10.1016/j.engappai.2025.110669}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110669}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Computer-aided detection systems based on ballistocardiography signals: A review}, volume = {151}, year = {2025}, }
• (2025). Paying attention to the minute details: Supervised keypoint detection on dense, complex point clouds. EAAI, 151, 110668. (www)

  @article{ , abstract = {Keypoints in three-dimensional (3D) objects are crucial for applications such as shape description, registration, and clinical diagnosis. However, current keypoint detection methods for 3D objects often depend on multiview two-dimensional (2D) images or unsupervised learning, largely due to the limited availability of annotated datasets. This limitation restricts their ability to detect anatomically significant keypoints directly on 3D models. To overcome this challenge, we propose a supervised keypoint detection framework designed for dense and complex 3D point clouds. Our framework combines a specialized segmentation network to identify potential keypoint regions, followed by precise keypoint localization using a PointNet encoder with residual modules and a novel “double SoftMax” mechanism. While keypoint detection accuracy benefits from restricting potential regions to smaller areas, this can leave some regions undetected in dense point clouds. To address this, we introduce a “Penalty Dice Loss” into the segmentation network, which effectively reduces undetected regions. Experiments on in-house skull and tibia datasets show mean radial errors of 1.43 mm and 1.54 mm, and success detection rates of 76.00% and 76.40%, respectively, within the clinically accepted 2 mm range. These results are competitive with state-of-the-art 2D cephalometric X-ray landmark detection methods and demonstrate the potential of our framework for clinical applications, such as preoperative planning, intraoperative navigation, and postoperative evaluation.}, archive = {J_EAAI}, author = {Qiuyang Chen and Shenghui Liao and Xiaoyan Kui and Ziyang Hu and Jianda Zhou}, doi = {10.1016/j.engappai.2025.110668}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110668}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Paying attention to the minute details: Supervised keypoint detection on dense, complex point clouds}, volume = {151}, year = {2025}, }
• (2025). Corrigendum to “Image–text sentiment analysis based on hierarchical interaction fusion and contrast learning enhanced” [Eng. appl. artific. intellig. 146 (2025) 110262]. EAAI, 151, 110667. (www)

  @article{ , archive = {J_EAAI}, author = {Hongbin Wang and Qifei Du and Yan Xiang}, doi = {10.1016/j.engappai.2025.110667}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110667}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Corrigendum to “Image–text sentiment analysis based on hierarchical interaction fusion and contrast learning enhanced” [Eng. appl. artific. intellig. 146 (2025) 110262]}, volume = {151}, year = {2025}, }
• (2025). Leveraging deep reinforcement learning to optimize linear quadratic regulator parameters for leader-follower formation control. EAAI, 151, 110666. (www)

  @article{ , abstract = {The widespread application of robotic formations in various fields, from exploration missions to precision agricultural operations, has highlighted the urgent need for studying and optimizing formation control techniques. This paper aims to enhance the full-process control of a leader-follower formation system, encompassing both formation establishment and maintenance. Firstly, an algorithm is designed to autonomously assign team roles to follower robots in a distributed fashion, thereby generating a virtual goal pose for each follower. Secondly, a combination of a Linear Quadratic Regulator (LQR) controller, based on the Ackermann model, and a Proportional-Derivative (PD) controller is developed for lateral and longitudinal distance control of each robot. Subsequently, an improved Asynchronous Advantage Actor-Critic (A3C) algorithm is utilized to expedite the training process and enable real-time online parameter optimization of the LQR controller. Finally, the Robot Operating System 2 (ROS2)-based formation system and experimental setup are presented, along with a detailed description of the A3C training process. According to both simulation and real-vehicle experiments, the formation control method proposed in this paper demonstrates stable operation and high accuracy in formation maintenance, surpassing other similar methods and the traditional LQR approach.}, archive = {J_EAAI}, author = {Zhi Wang and Yun Ling and Min Ma and TingTing Chen}, doi = {10.1016/j.engappai.2025.110666}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110666}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Leveraging deep reinforcement learning to optimize linear quadratic regulator parameters for leader-follower formation control}, volume = {151}, year = {2025}, }
• (2025). Interpretable spatio-temporal prediction using deep neural network - Local interpretable model-agnostic explanations: A case study on leptospirosis outbreaks in malaysia. EAAI, 151, 110665. (www)

  @article{ , abstract = {Leptospirosis is a widespread zoonotic disease with complex spatio-temporal dynamics. This study investigates the use of Deep Neural Network (DNN) in combination with Local Interpretable Model-Agnostic Explanations (LIME) for weekly spatio-temporal predictions of leptospirosis occurrence. The predictive model integrates hydroclimatic and environmental data to assess its effectiveness in predicting leptospirosis cases and quantifying key input variables in Negeri Sembilan, Malaysia. Using a DNN architecture with hyperparameter tuning via grid search, we developed a globally trained model that achieved an overall prediction accuracy of 70.5% across 214 pixels. We identified acidic soil and a higher presence of rubber plantations as strong predictors of leptospirosis occurrence. Additionally, mean temperature and minimum rainfall emerged as important hydroclimatic contributors. These insights enable public health authorities to proactively identify and prioritize high-risk areas for targeted interventions, improving disease mitigation strategies. Furthermore, the methodology is adaptable to other regions with similar environmental and socio-economic conditions, strengthening early warning systems and enhancing preparedness against future leptospirosis outbreaks. While demonstrated on leptospirosis prediction, the proposed DNN-LIME framework is adaptable to spatio-temporal challenges in diverse domains such as supply chain optimization, urban planning, and industrial risk management. The integration of interpretability via LIME ensures actionable insights for stakeholders beyond public health, bridging the gap between complex models and real-world decision-making.}, archive = {J_EAAI}, author = {Fariq Rahmat and Zed Zulkafli and Asnor Juraiza Ishak and Ribhan Zafira Abdul Rahman and Wardah Tahir and Jamalludin Ab Rahman and Veianthan Jayaramu and Simon De Stercke and Salwa Ibrahim and Muhamad Ismail}, doi = {10.1016/j.engappai.2025.110665}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110665}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpretable spatio-temporal prediction using deep neural network - Local interpretable model-agnostic explanations: A case study on leptospirosis outbreaks in malaysia}, volume = {151}, year = {2025}, }
• (2025). From data scarcity to diagnostic precision: A novel data augmentation and fault diagnosis framework for district heating substations. EAAI, 151, 110662. (www)

  @article{ , abstract = {This study introduces FLAME (Fault Localization using Augmented Model Enhancement), a novel fault diagnosis framework for District Heating (DH) substations. Automated Fault Detection and Diagnosis (FDD) has become imperative as many DH substations perform sub-optimal due to faults. The main challenges complicating accurate fault diagnosis are increasing operational complexities and a scarcity of labelled data. FLAME integrates a hybrid Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) model with an attention mechanism and introduces the Fault Augmentation Signature Technique (FAST). FAST overcomes the limitations of traditional stochastic data augmentation methods by leveraging pattern mixing of the time series. The FLAME framework uses transfer learning, initially trained on augmented data using FAST and fine-tuned using original substation data. Experimental results reveal that FLAME outperforms conventional methods, obtaining F1 scores of approximately 0.95 and 0.92 on lab-simulated and real-world datasets, respectively. Additionally, the research found the importance of the temperature difference measurement ( Δ T ) and median-based sampling strategies for optimal fault pattern identification. These findings establish FLAME as a new benchmark in DH system diagnostics, offering a robust framework to enhance fault diagnosis accuracy and operational efficiency of DH substations.}, archive = {J_EAAI}, author = {Jonne van Dreven and Abbas Cheddad and Sadi Alawadi and Ahmad Nauman Ghazi and Jad Al Koussa and Dirk Vanhoudt}, doi = {10.1016/j.engappai.2025.110662}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110662}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {From data scarcity to diagnostic precision: A novel data augmentation and fault diagnosis framework for district heating substations}, volume = {151}, year = {2025}, }
• (2025). Improving zero-shot cross-domain slot filling via machine reading comprehension prompt template. EAAI, 151, 110660. (www)

  @article{ , abstract = {In recent years, task-oriented dialogue systems have gained significant attention and seen rapid development in applications such as intelligent customer service, voice assistants, and online consultations. Slot filling, as a fundamental task in task-oriented dialogue systems, plays a crucial role in extracting relevant information to help these systems fulfill users’ specific tasks. However, in practical applications, models often face data from unseen domains, which presents a challenge to their generalization ability. To address this, cross-domain slot filling has been introduced to enhance the robustness and adaptability of multi-domain dialogue systems. Previous works normalized cross-domain slot filling to a reading comprehension task. Nevertheless, due to inherent discrepancies between slot filling and machine reading comprehension, this approach fails to fully exploit the wealth of knowledge embedded in pre-trained reading comprehension model. To that end, we proposed a R eading C omprehension P rompt T emplate for S lot F illing (RCPTSF). It contains two components: C ontextual A ttention T emplate (CAT) and S lot A ttention T emplate (SAT). To tackle the challenge of confusion encountered by traditional methods across different domain slots, the slot attention template employs the slot description from a specific domain to reconstruct a passage for model input. Additionally, the contextual attention template is introduced to reconcile the conflict arising from the robust contextual reasoning ability of pre-trained models in reading comprehension task and the conceptual comprehension ability required by slot-filling task. Consequently, we redesign the query statement by incorporating contextual information. Experiments show that our method achieves a large improvement (14.6%) on the SNIPS dataset, achieving state-of-the-art performance.}, archive = {J_EAAI}, author = {Fuping Liu and Peijie Huang and Yuhong Xu and Guotai Huang}, doi = {10.1016/j.engappai.2025.110660}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110660}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving zero-shot cross-domain slot filling via machine reading comprehension prompt template}, volume = {151}, year = {2025}, }
• (2025). Kernel-free quadratic proximal support vector machine with lp-norm regularization. EAAI, 151, 110658. (www)

  @article{ , abstract = {For binary classification problems, real-world data often exhibit class imbalance, noise, and outliers, and their complex distribution requires the use of kernel functions to separate the data non-linearly. In this paper, we propose a novel nonlinear classifier, called the kernel-free quadratic proximal support vector machine with an arbitrary L p -norm regularization (L p -QPSVM), where p > 0 . The goal of our L p -QPSVM is to find two quadratic hypersurfaces to non-linearly classify. By introducing the L p -norm regularization term in our method, L p -QPSVM allows for flexible adjustment of p , enhancing its robustness and generalization. To strengthen practical applications, a simplified version of L p -QPSVM is proposed. Additionally, we transform the two optimization problems of L p -QPSVM into the convex quadratic programming problems, and design an iterative algorithm to solve them. The convergence, interpretability and computational complexity of L p -QPSVM are provided. Numerical experiments on the artificial and benchmark datasets validate the effectiveness of our L p -QPSVM, demonstrating its state-of-the-art performance compared with the representative classification methods.}, archive = {J_EAAI}, author = {Xue Yang and ZhiXia Yang and JunYou Ye and YuanYuan Chen}, doi = {10.1016/j.engappai.2025.110658}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110658}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Kernel-free quadratic proximal support vector machine with lp-norm regularization}, volume = {151}, year = {2025}, }
• (2025). Surface defect detection using distributed features. EAAI, 151, 110655. (www)

  @article{ , abstract = {Surface defect detection is crucial in manufacturing and has attracted increasing attention due to its significant impact on product quality and operational efficiency. With advancements in artificial intelligence (AI), deep learning based networks have become widely utilized for automated surface defect detection. However, most existing networks rely on a standard convolutional layer (Conv), which mainly extracts local information due to their limited learning region defined by kernel size. To address this limitation, we propose a surface defect detection network utilizing distributed features (DF-SDD), which consists of a distributed feature extraction block (DFEB) for feature extraction and a spatial attention gate (SAG) for feature fusion. DFEB effectively captures local and long-range information by introducing a distributed convolutional layer (DConv) and a distributed depthwise convolutional layer (DDWConv), which offer a broader learning area while maintaining low complexity. SAG merges features from different levels to enhance spatial information across multiple scales. Experimental results show that DF-SDD outperforms state-of-the-art networks on surface defect datasets in terms of both performance and complexity.}, archive = {J_EAAI}, author = {The Van Le and Jordan Daniel Joshua and Taehwan Kim and Jinhyuk Lee and Seong Han Kim and Jin Young Lee}, doi = {10.1016/j.engappai.2025.110655}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110655}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Surface defect detection using distributed features}, volume = {151}, year = {2025}, }
• (2025). Integration of neural networks in brain–computer interface applications: Research frontiers and trend analysis based on python. EAAI, 151, 110654. (www)

  @article{ , abstract = {Brain–computer interfaces (BCIs) establish a connection between the human brain and external devices, facilitating novel forms of communication and control. Although BCIs possess significant potential, there is a need for performance improvements. Integrating neural networks into BCIs is crucial for enhancing functionality and promoting broader adoption. This study examines 1,867 articles from the Web of Science core database, covering the period from 1996 to 2024, to identify contemporary hotspots and trends in the application of neural networks within BCIs. This study employs bibliometric methods and Python for analysis, examining collaborative relationships, citation networks, keyword bursts, and clustering with visual representations of the findings. The results indicated that current study hotspots predominantly center on “P300,” “Long Short-Term Memory,” “Motor Imagery,” “Epilepsy,” “Emotion Recognition,” “Feature Extraction,” and “Transfer Learning.” Future development directions encompass: (1) the establishment of public BCI datasets; (2) the exploration of diverse feature extraction and fusion methods; (3) the enhancement of machine learning and deep learning integration for improved performance and real-time processing; (4) the expansion of application scenarios and the development of portable devices; (5) the optimization of transfer learning algorithms to mitigate performance challenges arising from individual differences. This study provides an overview of the current research landscape and identifies potential future research directions. Furthermore, it assists practitioners in recognizing additional business opportunities and acts as a resource for the formulation of government policy.}, archive = {J_EAAI}, author = {Junming Chen and Hongyu Yin and Kai Zhang and Yangzhi Ren and Hui Zeng}, doi = {10.1016/j.engappai.2025.110654}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110654}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integration of neural networks in brain–computer interface applications: Research frontiers and trend analysis based on python}, volume = {151}, year = {2025}, }
• (2025). Event-triggered fractional-order fuzzy sliding mode control using online reinforcement learning for uncertain nonlinear systems: Practical validation. EAAI, 151, 110653. (www)

  @article{ , abstract = {In this paper, a novel event-triggered control strategy is proposed for uncertain nonlinear systems by developing a fractional-order fuzzy sliding mode controller based on a fractional-order actor–critic network. The proposed approach offers several key features. First, a sigma-point Kalman filter is employed to accurately estimate unmeasured states. Second, a fractional-order sliding mode controller with an event-triggered mechanism is designed to achieve practical sliding mode control while preventing the Zeno phenomenon. Third, to reduce chattering in sliding mode control, a fractional-order actor–critic recurrent neural network is proposed, effectively approximating the switching control stage and enhancing system performance while reducing event triggers. The fractional-order actor–critic network incorporates fuzzy rules defined by a generalized Gaussian function with the Mittag-Leffler function, and a critic network approximates the value function, further enhancing performance. Parameter learning is guided by a fractional-order Gauss–Newton method. Stability analysis is performed using the Lyapunov method. Finally, the efficacy of the proposed method is demonstrated via experimental validation on a real inverted pendulum system.}, archive = {J_EAAI}, author = {Tarek A. Mahmoud and Mohammad El-Hossainy and Belal Abo-Zalam and Raafat Shalaby}, doi = {10.1016/j.engappai.2025.110653}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110653}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Event-triggered fractional-order fuzzy sliding mode control using online reinforcement learning for uncertain nonlinear systems: Practical validation}, volume = {151}, year = {2025}, }
• (2025). Robust underwater imaging model and automatic parameter optimization for underwater image restoration. EAAI, 151, 110651. (www)

  @article{ , abstract = {Existing prior-driven restoration methods generally apply traditional underwater imaging model to restore image. The traditional underwater imaging model does not take non-uniform illumination and noise into consideration, resulting in low brightness and amplified noise in restored images. Moreover, the configuration of parameters in existing methods relies heavily on empirical strategies, generating suboptimal results. To address these issues, a novel robust underwater imaging model is formulated by separating incident illumination and reflectance from real underwater scene and adding a noise term in the traditional model. Subsequently, a variational framework is established based on the proposed robust underwater imaging model, and novel regularization terms are explored to suppress noise and calculate the incident illumination and reflectance components simultaneously. After that, two post-processing methods are performed to improve clarity of the reflectance and correct color deviation of the incident illumination, respectively. Finally, a parameter optimization strategy is designed to search for optimal parameters for the restoration method. Comprehensive experiments demonstrate the excellent performance of the proposed method in improving visibility, correcting color and suppressing noise of underwater images.}, archive = {J_EAAI}, author = {Chenggang Dai and Dongnian Li and Chengjun Chen and Zhengxu Zhao and Mingxing Lin}, doi = {10.1016/j.engappai.2025.110651}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110651}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust underwater imaging model and automatic parameter optimization for underwater image restoration}, volume = {151}, year = {2025}, }
• (2025). Chance constrained programming for sustainable four dimensional fuzzy-rough transportation problem with rest period of drivers and time window constraints. EAAI, 151, 110648. (www)

  @article{ , abstract = {The rest period of driver plays a critical role in ensuring both safety and efficiency, and hence the success rate of a transportation system. Fatigued drivers are more prone to accidents and errors, making it essential to incorporate their rest time into the transportation planning. Additionally, time window constraints, which define specific time frames for deliveries, play a significant role in the efficiency of transportation systems. Despite their importance, existing research has yet to integrate both driver’s rest period and time window constraints into transportation models. To address these gaps and improve operational performance, this study introduces a novel multi-objective, multi-item four-dimensional green transportation model that incorporates both driver’s rest period and time window constraints. Given the complexities of predicting market demand and other transportation-related parameters within specific time frames, the model’s parameters are represented as trapezoidal fuzzy-rough numbers. A new methodology, “Neutrosophic Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS)”, based on neutrosophic programming, is proposed to find an optimal compromise solution. The practicality of this approach is demonstrated by solving a real-world industrial problem. A comparative analysis shows that the Neutrosophic TOPSIS method yields the most effective Pareto-optimal solution. The results reveal a reduction of 10.85 h in transportation time and 21.36 kg in carbon emissions compared to existing methods. Additionally, the findings reveal that excluding the driver’s rest period reduces transportation time by 15.9 h but increases carbon emissions by 591 kg. Lastly, the possible avenues for future research are outlined.}, archive = {J_EAAI}, author = {Shivani and Deepika Rani and Gourav Gupta}, doi = {10.1016/j.engappai.2025.110648}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110648}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Chance constrained programming for sustainable four dimensional fuzzy-rough transportation problem with rest period of drivers and time window constraints}, volume = {151}, year = {2025}, }
• (2025). Fast fusion net: Defect detection and fault identification methods for high-voltage overhead power lines. EAAI, 151, 110646. (www)

  @article{ , abstract = {The integrity of the power line's mechanical structure directly impacts the stable operation of the transmission and distribution system. Defects in the power line are the leading cause of transmission failures. However, in the intelligent detection of power line defects using artificial intelligence algorithms, no relevant research has achieved a balance between prediction accuracy and computational efficiency in model design. Additionally, an intelligent algorithm that is resistant to background noise interference and suitable for the inherent characteristics of power line defect detection must be designed. Based on this, we propose an efficient network structure: Fast Fusion Net. First, by designing an efficient and lightweight backbone and head structure, the algorithm significantly reduces the computational complexity of the model during feature extraction and prediction output. Second, a Dual Attention Coordination Mechanism (DACM) and a context semantic information enhancement feature fusion module are designed. These enhance the fine-grained representation in the feature map, enabling the model to handle background noise interference and improving its overall robustness. Finally, a host computer software system for model deployment is developed in conjunction with the algorithm, and the application potential of the proposed method in semantic segmentation tasks is preliminarily explored. Experimental results show that compared with other state-of-the-art methods, our algorithm achieves more advanced SOTA (State-of-the-Art) performance in terms of inference efficiency and prediction accuracy. To further verify the portability of the method, we conducted verification experiments on the public datasets, PASCAL VOC 2007 (PASCAL Visual Object Classes Challenge, 2007), and obtained superior results. The results of other comparative and ablation experiments also prove the effectiveness of the proposed method. Our work encourages further exploration, and tasks related to the datasets can be accessed at: https://github.com/Song-zhiwei/PLDD-datasets.git .}, archive = {J_EAAI}, author = {Zhiwei Song and Yun Zhang and Xinbo Huang ( Senior Member, Ieee ) and Ye Zhang}, doi = {10.1016/j.engappai.2025.110646}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110646}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast fusion net: Defect detection and fault identification methods for high-voltage overhead power lines}, volume = {151}, year = {2025}, }
• (2025). Research on multi-objective point path planning for mobile inspection robot based on multi-informed-rapidly exploring random tree∗. EAAI, 151, 110645. (www)

  @article{ , abstract = {Daily inspection as a top priority to ensure the safe production of intelligent workshops, can directly respond to the degree of intelligence of the factory workshop, which often needs robots to inspect multiple target devices. To enhance the inspection quality of intelligent workshop inspection robots, this thesis is based on the Informed-Rapidly Exploring Random Tree (RRT) algorithm and introduces the concept of simultaneous expansion of multiple trees. This paper proposes a strategy involving random tree connectivity to the Multi-Informed-RRT∗ (MI-RRT∗) global path planning algorithm, which is suitable for multi-objective point planning. It is verified through simulation that in solving the same planning problem for eight objective points, compared with the basic Informed-RRT∗ algorithm, the MI-RRT∗ algorithm's running time is reduced by 79.56 %, and the final generated path length is reduced by 25.26 %; compared with the Multi-RRT∗ algorithm, which is also used for solving the multi-objective point planning problem, the running time of the algorithm is reduced by 48.48 % and the final generated path length is reduced by 7.45 %. In conclusion, this study proposes a multi-objective point path planning algorithm tailored for the dynamic environment of intelligent workshops, achieved through the integration of MI-RRT∗ and an enhanced Dynamic Window Approach (DWA) algorithm. Experimental evidence substantiates that the path generated by the MI-RRT∗ algorithm, as presented in this paper, is superior, resulting in shorter robot running times and a substantial enhancement in inspection efficiency within the workshop. Simultaneously, it mitigates the issue of traditional inspection equipment leakage, thereby possessing tangible practical value.}, archive = {J_EAAI}, author = {Hongjuan Yang and Xu Luo and Chao Duan and Peishen Wang and Kaidi Zhu and Xingqiao Deng and Wenjuan Ren}, doi = {10.1016/j.engappai.2025.110645}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110645}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on multi-objective point path planning for mobile inspection robot based on multi-informed-rapidly exploring random tree∗}, volume = {151}, year = {2025}, }
• (2025). Three-level-fused attribute reductions based on three-view uncertainty measures of three-view weighted neighborhood rough sets. EAAI, 151, 110644. (www)

  @article{ , abstract = {Attribute reductions facilitate classification learning, and they rely on uncertainty measures related to knowledge granulations of various rough sets. Weighted neighborhood rough sets (WNRSs) introduce attribute weights to optimize granular structures and improve neighborhood models, but their current attribute reductions and corresponding algorithms utilize algebra weights and dependency degrees from only a single algebraic perspective. For enriching WNRSs and their attribute reductions, three-view attribute weights and three-view granulation measures are constructed from algebraic, informational, and algebra-information-fused viewpoints, and thus 3 × 3 = 9 attribute reductions are systematically proposed to generate heuristic algorithms on three-level fusion. First based on existing algebraic weights, informational weights are supplemented by information entropy, while fused weights are constructed by algebra-information integration; thus, three-view weights induce three-view weighted neighborhood granulations and three-view WNRSs, and the latter’s granulation monotonicity is achieved. Then based on WNRSs and dependency degrees, decision entropies are supplemented by information function on roughness, while fused measures are constructed by multiplication operation; thus, three-view measures are formulated to acquire granulation monotonicity. Furthermore, three-view weights and measures motivate attribute reductions of WNRSs, and 3 × 3 = 9 heuristic reduction algorithms are systematically designed, thus extending and improving the existing algorithm; in terms of fusion strength, these 9 algorithms exhibit a three-level structure of 4 + 4 + 1 on non-fusion, single-fusion, double-fusion. Finally through data experiments, relevant uncertainty measures and attribute reductions are validated; all 9 reduction algorithms are comprehensively compared in classification learning, the new algorithms generally outperform the contrastive algorithm, and the double-fused algorithm on fused weights and measures acquires the best performance.}, archive = {J_EAAI}, author = {Jiang Chen and Xianyong Zhang and Xiao Tang and Xiaoling Yang and Zhiying Lv}, doi = {10.1016/j.engappai.2025.110644}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110644}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-level-fused attribute reductions based on three-view uncertainty measures of three-view weighted neighborhood rough sets}, volume = {151}, year = {2025}, }
• (2025). Green and efficient path planning framework based on multi-granularity view fusion for amphibious unmanned aerial vehicles. EAAI, 151, 110643. (www)

  @article{ , abstract = {Unmanned aerial vehicles (UAVs) equipped with land-air mobility are intelligent amphibious systems that are extensively applied in low-altitude transportation, reconnaissance, and emergency rescue. However, the unique multimodal movement characteristics of amphibious UAVs, along with the complexity of three-dimensional (3D) environments, render traditional path planning mechanisms computationally expensive and energy-intensive, thereby resulting in suboptimal efficiency and sustainability. In response, this study proposes an innovative green and efficient path planning scheme based on multi-granularity view fusion. First, a fine-grained view generation mechanism is introduced to construct multiple two-dimensional (2D) views, thereby ensuring global consistency and capturing local details to adapt to varying environments. Subsequently, an improved path search algorithm is proposed, and parallel computation is utilized to enhance search efficiency. Furthermore, a path fusion module is introduced that integrates paths from different views, ultimately generating 3D paths that combine land-based driving and aerial flight. Finally, to validate the effectiveness and superiority of our method, extensive experiments are conducted across various environments. The experimental results demonstrate that our approach effectively integrates multi-granularity views with an enhanced path planning strategy, achieving reductions in path length, energy consumption, and time complexity. Compared to existing path planning methods for amphibious robots, the proposed approach exhibits improvements in the average green index by 1.96%, 16.98%, 8.29%, and 43.97%, respectively. This lays a solid foundation for the application of green and intelligent technologies in the development of land-air amphibious unmanned aerial vehicles.}, archive = {J_EAAI}, author = {Xinghong Yang and Lizhen Shao}, doi = {10.1016/j.engappai.2025.110643}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110643}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Green and efficient path planning framework based on multi-granularity view fusion for amphibious unmanned aerial vehicles}, volume = {151}, year = {2025}, }
• (2025). The optimization-based fuzzy logic controllers for autonomous ground vehicle path tracking. EAAI, 151, 110642. (www)

  @article{ , abstract = {Nowadays, it is possible to see Autonomous Ground Vehicles (AGV) in traffic, factories and even restaurants. One of the key elements of these vehicles is the path tracking controller. Path tracking performance of the controller has a direct effect on the ride experience of the vehicle. This work proposes Fuzzy Logic Controllers (FLC) for steering of an autonomous vehicle. A fuzzy controller does not need for a certain model to steer the vehicle. However, the rules and membership functions should be determined with the aid of an expert. Here, with the purpose of getting one step close to the automatically tunable fuzzy control systems, the Artificial Bee Colony (ABC) algorithm is utilized on tuning the output membership functions of the FLC. Three different fuzzy controllers are developed for an autonomous vehicle. The first one is provided from the literature, two others are developed with the help of ABC algorithm. The base widths of the output membership functions of two other fuzzy controllers are determined by using cost functions of the Integral of Absolute Error (IAE) and the Integral of Time Absolute Error (ITAE). The performances of the developed controllers are evaluated through the results of various simulation studies, keeping in view performance indicators such as overshoot, settling time and Mean Squared Error (MSE). The results demonstrate superiority of the controllers proposed in this work to the reference one (the first fuzzy controller). Finally, this work reveals that the parameters of the fuzzy controller devised to steer an AGV can be tuned by using metaheuristic optimization algorithms.}, archive = {J_EAAI}, author = {Ibrahim Aliskan}, doi = {10.1016/j.engappai.2025.110642}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110642}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The optimization-based fuzzy logic controllers for autonomous ground vehicle path tracking}, volume = {151}, year = {2025}, }
• (2025). Classification of power quality disturbances using residual networks with channel attention mechanism. EAAI, 151, 110641. (www)

  @article{ , abstract = {The detection and classification of power quality disturbances (PQDs) carries significant importance for power systems. In response to this imperative, numerous intelligent diagnostic methods have been developed. However, existing identification methods usually concentrate on single-type signals or on complex signals with two types, rendering them susceptible to noisy labels and environmental effects. This study proposes a novel method for the classification of PQDs, named Stockwell transform-Grouped convolution Squeeze-and-excitation Residual Network (ST-GSResNet), which utilizes the Stockwell transform (S-Transform) and an improved residual network (ResNet) with a channel attention mechanism. The ST-GSResNet approach initially uses the S-Transform to transform a time-series signal into a two-dimensional (2D) time–frequency image for feature enhancement. Then, an improved ResNet model is introduced, which employs grouped convolution instead of the traditional convolution operation. This improvement aims to facilitate learning with a block-diagonal structured sparsity on the channel dimension, the highly-correlated filters are learned in a more structured way in the networks with filter groups. By reducing the number of parameters in the network in this significant manner, the model becomes less prone to overfitting. Furthermore, the channel attention mechanism (squeeze-and-excitation module) concentrates on primary components, which enhances the model’s robustness in recognition and immunity to noise. Experimental results demonstrate that, compared to existing deep learning models, our approach has advantages in computational efficiency and classification accuracy.}, archive = {J_EAAI}, author = {Su Pan and Xingyang Nie and Xiaoyu Zhai and Cheng He and Zhenping Ding}, doi = {10.1016/j.engappai.2025.110641}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110641}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Classification of power quality disturbances using residual networks with channel attention mechanism}, volume = {151}, year = {2025}, }
• (2025). Fuzzification-back propagation neural network-based model prediction for robotic arm positioning error reduction. EAAI, 151, 110639. (www)

  @article{ , abstract = {Robotic arms are pervasively used in critical manufacturing fields, but its absolute positioning accuracy cannot be controlled because of random errors. Although some researches using back propagation neural network to predict the robotic arm's absolute positioning error have been proved, they suffer from poor convergence and low prediction accuracy in that input parameters contain unavoidable measurement errors. This paper proposed an error prediction model based on fuzzification and back propagation neural network. The rotation angle and direction of robotic joints are employed as training samples for the back propagation neural network, and converted into error contributions using the fuzzification to eliminate the influence of measurement errors in the input parameters. The input parameters are simplified, which enables the training process of the back propagation neural network to be optimized. Experimental results showed that the training time of the model was reduced by two times or more, and the mean square error was decreased by roughly 2.94 %. Meanwhile, the average absolute positioning error of the robotic arm with the prediction model was reduced by 59.22 %. The model can be easily transplanted into embedded systems to provide a methodology for new design of robotic arm error compensators.}, archive = {J_EAAI}, author = {Jiang Liu and Jianwei Wu and Jiansheng Pan and Pengyue Zhao}, doi = {10.1016/j.engappai.2025.110639}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110639}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzification-back propagation neural network-based model prediction for robotic arm positioning error reduction}, volume = {151}, year = {2025}, }
• (2025). Synthetic data-augmented explainable vision transformer for colorectal cancer diagnosis via surface tactile imaging. EAAI, 151, 110633. (www)

  @article{ , abstract = {In this work, we present a synthetic data augmented explainable Vision Transformer (ViT) framework designed for the informed and intuitive early diagnosis of colorectal cancer (CRC) polyps. The framework uses textural images — generated by our recently developed vision-based tactile sensor (called HySenSe) and augmented by synthetically generated images from a diffusion model pipeline, to output class-based probabilities of potential CRC polyp types. Additionally, it provides local relevancy-based heatmaps to assist clinicians by highlighting key areas of interest in the tactile images representing CRC polyp textures. We benchmark each aspect of this framework through: (i) Inception Scores for the synthetic images generated by the diffusion pipeline, (ii) Performance evaluation and sensitivity analyses on the effects of synthetic data addition on model generalizability compared with other state-of-the-art architectures, (iii) Dimensionality reduction techniques to confirm the suitability of synthetically generated images, and (iv) Comparison of two independent approaches visualizing explainability.}, archive = {J_EAAI}, author = {Siddhartha Kapuria and Naruhiko Ikoma and Sandeep Chinchali and Farshid Alambeigi}, doi = {10.1016/j.engappai.2025.110633}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110633}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Synthetic data-augmented explainable vision transformer for colorectal cancer diagnosis via surface tactile imaging}, volume = {151}, year = {2025}, }
• (2025). Aspect-based sentiment analysis for software requirements elicitation using fine-tuned bidirectional encoder representations from transformers and explainable artificial intelligence. EAAI, 151, 110632. (www)

  @article{ , abstract = {Aspect-Based Sentiment Analysis (ABSA) of app reviews allows a better understanding of user preferences regarding specific product features and helps the development team elicit requirements effectively. The existing literature faces challenges such as limited focus on the automation of Requirement Elicitation (RE), insufficient task-specific fine-tuning of models such as Bidirectional Encoder Representations from Transformers (BERT), and lack of interpretability owing to the black-box nature of these models. Therefore, our work makes the following significant contributions to address these challenges: (1) development and evaluation of a robust method based on ABSA for the automation of the RE process; (2) optimization of ABSA using BERT fine-tuning for enhanced performance, which includes conducting a comprehensive ablation study to obtain the best hyperparameters that guarantee the best model performance and robustness; and (3) integration of Explainable Artificial Intelligence (XAI) techniques for enhanced BERT model interpretability. Our work was evaluated on the ABSA Warehouse of Apps REviews (AWARE) dataset, a specifically tailored dataset for the RE process. Our study outperformed baseline models such as the Support Vector Machine (SVM), Convolutional Neural Network (CNN), and BERT, and achieved an average F1-Score of 0.83 for the Aspect Category Detection (ACD) task and 0.94 for the Aspect Category Polarity (ACP) task. In addition, we employed XAI using Locally Interpretable Model-Agnostic Explanations (LIME) to explain the BERT model prediction results, which aids in the improved visualization and interpretability of the app review analysis for the automated RE process.}, archive = {J_EAAI}, author = {Soonh Taj and Sher Muhammad Daudpota and Ali Shariq Imran and Zenun Kastrati}, doi = {10.1016/j.engappai.2025.110632}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110632}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Aspect-based sentiment analysis for software requirements elicitation using fine-tuned bidirectional encoder representations from transformers and explainable artificial intelligence}, volume = {151}, year = {2025}, }
• (2025). Advanced code time complexity prediction approach using contrastive learning. EAAI, 151, 110631. (www)

  @article{ , abstract = {It is a crucial task to predict the algorithmic time complexity for estimating the efficiency of a software code. Since the problem is known to be undecidable in theory, there is no 100% accurate tools to solve the problem. Even humans often make mistakes when analyzing the time complexity of code, and this process requires considerable effort and time to thoroughly examine the code. Therefore, we aim to develop an automated method for analyzing code time complexity. We observe that solution codes submitted for coding problems in competitive programming contests tend to have similar time complexities due to constraints such as time limits and functional requirements of the problems. Based on this observation, we propose a contrastive learning-based training strategy that aligns solution codes for the same competitive programming problem. Our training strategy clusters codes with similar time complexities by using both natural language problem descriptions and a single reference code per problem as anchors. This design enables the model to capture core algorithmic features such as loops and recursion more accurately. Experiments in three scenarios – in-dataset, cross-dataset, and cross-language – demonstrate substantial gains on pre-trained code models, consistently surpassing existing methods in both accuracy and generalizability. Our proposed training strategy yields an average 12.54% improvement over cross-entropy-based training, and an 8.01% improvement over data augmentation-based contrastive learning.}, archive = {J_EAAI}, author = {Shinwoo Park and Joonghyuk Hahn and Elizabeth Orwig and Sang-Ki Ko and Yo-Sub Han}, doi = {10.1016/j.engappai.2025.110631}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110631}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advanced code time complexity prediction approach using contrastive learning}, volume = {151}, year = {2025}, }
• (2025). Automatic selection of inducing points in sparse gaussian process for approximations of finite element analyses. EAAI, 151, 110629. (www)

  @article{ , abstract = {Gaussian process regression (GPR) is a widely used regression model, but it has poor scalability. Sparse approximation methods improve scalability by using inducing points to approximate the GPR, but determining the optimal number and placement of these points is challenging. Increasing the number of inducing points generally improves the predictive accuracy, but it comes at a computational cost. This article presents a method to estimate the necessary number of inducing points for accurate predictions of finite element method (FEM) analyses using approximate GPR. The approach leverages the proper orthogonal decomposition (POD) technique, using its modes to determine the inducing points. Results demonstrate that the proposed method identifies a sufficient number of inducing points for approximate GPR to achieve predictive accuracy comparable to full GPR, but with half the training time. This approach ensures computational efficiency without significant loss in accuracy, making it a valuable tool for scalable regression in engineering applications. POD has previously been combined with GPR to provide computationally efficient predictions for the full solution field across unseen variable combinations, treating spatial components separately via reduced basis functions. However, this work treats the spatial component as a variable within the GPR approximation, allowing continuous spatial predictions. This ensures that the covariance in the spatial dimension is captured by a single GPR. The method is applied to simulations of a three-span, post-tensioned concrete girder bridge.}, archive = {J_EAAI}, author = {Heine Havneraas Røstum and Sebastien Gros and Ketil Aas-Jakobsen and Joseph Morlier}, doi = {10.1016/j.engappai.2025.110629}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110629}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic selection of inducing points in sparse gaussian process for approximations of finite element analyses}, volume = {151}, year = {2025}, }
• (2025). Integrating cumulative binomial probability into artificial bee colony algorithm for global optimization in mechanical engineering design. EAAI, 151, 110628. (www)

  @article{ , abstract = {The artificial bee colony (ABC) algorithm has gained significant attention in engineering design optimization for its simplicity and robustness. However, it suffers from low convergence accuracy and imbalanced exploration and exploitation, particularly in non-convex and multi-modal problems. To alleviate these shortcomings, we present a novel ABC algorithm that combines cumulative binomial probability (CBABC), which contains two versions, single-dimensional evolution (CBACB_S) and multi-dimensional evolution (CBABC_M). According to the success and failure evolution experience, we first introduce a scaling factor based on cumulative binomial probability. Sequentially new movement equations with different characteristics are designed in the onlooker bee phase to balance local and global search capabilities. Then, a novel abandoned solution update mechanism is defined during scout bee phase to partly improve the solution accuracy. Finally, our approaches achieve a minimum winning rate of 68.19% and a maximum winning rate of 95.45% against its 9 outstanding ABC variants across the 22 functions. In addition, the proposed algorithms maintain a winning rate within [57.89%, 100%] compared to several state-of-the-art evolutionary algorithms for tackling 19 real-world mechanical engineering problems from Congress on Evolutionary Computation 2020 (CEC2020) suite.}, archive = {J_EAAI}, author = {Xiangyu Kong and Pengpeng Shang and Chunfeng Wang and Lixia Liu}, doi = {10.1016/j.engappai.2025.110628}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110628}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating cumulative binomial probability into artificial bee colony algorithm for global optimization in mechanical engineering design}, volume = {151}, year = {2025}, }
• (2025). Control oriented fast optimisation with SHapley additive explanations assisted two-stage training of input convex neural network. EAAI, 151, 110625. (www)

  @article{ , abstract = {Advancement of technology has enabled operation optimisation control of complex systems, which involves solving an optimisation problem. However, there are still challenges to harness the power of such technologies. This is especially true when it is required to solve a complex optimisation problem within prescribed time and resource limits. This paper introduces a novel fast optimisation framework by utilising input convex neural networks to approximate time-consuming optimal control algorithms. A two-stage training structure, consisting of offline data training and online fine-tuning, is put forward to ensure the effectiveness of the trained model. Moreover, an expansion data collection method and a roaming-style data augmentation strategy are proposed to improve the quality of training samples. A SHapley additive explanations assisted data selection algorithm is then developed to identify the most useful samples for training the model. Thus, reducing the training complexity. Further, a predictive safety filter is employed to guarantee that the control inputs produced by the trained model do not violate any constraints. Experiments on an inverted pendulum problem under four types of disturbances demonstrated that the trained input convex neural network controller exhibits superior control performance in comparison to a model predictive controller. Experimental results indicate that, in benchmark inverted pendulum on a cart control tests encompassing eight distinct scenarios, the trained input convex neural network controller achieves similar control performance comparable to that of the model predictive controller. Compared with the time model predictive controller, the trained input convex neural network controller demonstrates superior performance in scenarios with large initial angles and disturbances, though it performs slightly worse under disturbance-free conditions. Furthermore, the trained input convex neural network controller significantly outperforms both the model predictive controller and time model predictive controller in terms of computational efficiency, reducing computation time by 49.42% and 94.71%, respectively.}, archive = {J_EAAI}, author = {Chuang Wang and Lijun Zhang}, doi = {10.1016/j.engappai.2025.110625}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110625}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Control oriented fast optimisation with SHapley additive explanations assisted two-stage training of input convex neural network}, volume = {151}, year = {2025}, }
• (2025). Improving one-dimensional-based environmental sound classification models with progressive projection knowledge distillation from two-dimensional-based models. EAAI, 151, 110622. (www)

  @article{ , abstract = {Currently, state-of-the-art solutions for environmental sound classification (ESC) are achieved by predominantly using two-dimensional (2D) spectrograms with large deep neural networks. In contrast, the more favorable one-dimensional (1D)-based ESC models based on raw audio samples are lightweight, but their performances are relatively poor. Here, for the first time, we explore the potential of knowledge distillation (KD) to enhance lightweight 1D-based ESC models with powerful 2D-based ESC models. Specifically, we meticulously design two techniques to tackle the challenge in KD arising from different input modalities/model architectures and the large performance gap between teacher and student models. First, a projection KD introduces additional exit branches into the student that project the extracted intermediate features into the logits space, and align them with the logits output of the teacher; Second, progressive KD exploits the intermediate checkpoints of the teacher to construct a smooth training trajectory for the student to gradually learn. With the various combination pairs of representative teacher and student models, we conduct extensive experiments to evaluate the effectiveness of our proposed framework on two benchmark ESC datasets. With our proposed KD framework, 1D-based student ESC models consistently achieve significant performance improvements, with a maximum absolute gain of 1 . 6 % − 2 . 8 % and 2 . 2 % − 2 . 3 % , respectively. We hope that our research will bring a new perspective to knowledge distillation for sound-related applications and can also be applied to other tasks, such as audio-based cough detection and acoustic monitoring of biodiversity for efficient deployment.}, archive = {J_EAAI}, author = {Huaxing Xu and Haoyuan Chen and Xudong Liu and Haichuan Ren}, doi = {10.1016/j.engappai.2025.110622}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110622}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving one-dimensional-based environmental sound classification models with progressive projection knowledge distillation from two-dimensional-based models}, volume = {151}, year = {2025}, }
• (2025). Advancing greenhouse gas emission reduction strategies: Integrating multi-criteria decision-making with complex q-rung picture fuzzy Sugeno–Weber operators. EAAI, 151, 110621. (www)

  @article{ , abstract = {The urgent need to mitigate greenhouse gas emissions amidst escalating global climate change demands innovative and comprehensive strategies. This paper addresses this challenge by proposing a novel Multi-Criteria Decision-Making (MCDM) framework that integrates Sugeno–Weber t-norm and t-conorm within the Complex q-Rung Picture Fuzzy Set (Cq-RPFS) paradigm. Our approach introduces three new aggregation operators: Sugeno–Weber Weighted Average, Sugeno–Weber Ordered Weighted Average, and Sugeno–Weber Hybrid Weighted Average under the environment of Cq-RPFS. These operators enhance fuzzy information aggregation by addressing complex criterion interrelationships, offering a more nuanced evaluation of emission reduction strategies. We validate our framework through a detailed case study on emissions in India, focusing on aligning reduction strategies with sustainable development goals. The proposed method demonstrates superior effectiveness in managing multidimensional data compared to traditional aggregation techniques. Comprehensive sensitivity analysis and comparative studies further reveal the robustness and adaptability of our approach, showcasing its potential to provide more accurate and balanced assessments in real-world scenarios. This research fills a critical gap in decision-making methodologies and offers valuable insights for policymakers and researchers engaged in sustainable climate action.}, archive = {J_EAAI}, author = {Subramanian Petchimuthu and Fathima Banu M. and S. Thiruvazhimarba Pillai and Tapan Senapati}, doi = {10.1016/j.engappai.2025.110621}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110621}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advancing greenhouse gas emission reduction strategies: Integrating multi-criteria decision-making with complex q-rung picture fuzzy Sugeno–Weber operators}, volume = {151}, year = {2025}, }
• (2025). Sparse large-scale multi-objective optimization algorithm based on impact factor assistance. EAAI, 151, 110615. (www)

  @article{ , abstract = {In the real world, there exists a special category of multi-objective optimization problems with more than 1000 decision variables. However, only a few decision variables play a crucial role in optimizing the objective functions. Such problems are defined as sparse large-scale multi-objective optimization problems (SLSMOPs). Due to the difficulty in effectively identifying the non-zero positions of decision variables, traditional evolutionary optimization algorithms suffer from slow convergence speed and poor convergence effect, which means it is unable to efficiently obtain the Pareto optimal solution set. To address this challenge, the Impact Factor Assisted Algorithm (IFA) is proposed, which adopts a novel initial population strategy to generate sparse populations. Meanwhile, the impact factor of each decision variable is calculated, serving as a key basis for measuring the importance of each decision variable. During the algorithm’s operation, the impact factors are iteratively updated to rationally group decision variables and guide population evolution. This approach can accurately identify the positions of non-zero decision variables. The experimental results on eight benchmark and real-world problems indicate that the algorithm outperforms several existing sparse large-scale multi-objective optimization algorithms (SLSMOEAs).}, archive = {J_EAAI}, author = {Ziyu Hu and Xuetao Nie and Hao Sun and Lixin Wei and Jinlu Zhang and Cong Wang}, doi = {10.1016/j.engappai.2025.110615}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110615}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sparse large-scale multi-objective optimization algorithm based on impact factor assistance}, volume = {151}, year = {2025}, }
• (2025). Neural network-based metamodel of synthetic seismograms: Application for uncertainty quantification. EAAI, 151, 110613. (www)

  @article{ , abstract = {We developed time and frequency domains neural network surrogate models of synthetic seismograms stemming from the resolution of the three-dimensional equation of motion. The surrogates predict the time or frequency-series knowing the input variables of the physical model. The surrogate models are used to quantify epistemic uncertainty in ground motion prediction through global sensitivity analysis. We first generate a dataset of time domain synthetic seismograms in a seven-dimensions uncertain space using the spectral-element method. To validate the surrogate model, we evaluate its ability to reproduce at least 80% of the bootstrap resamples. Additionally, the R 2 regression coefficient between the simulations generated by the spectral-element method code and those predicted by the neural network is 0.94 for the validation set, confirming the accuracy of the surrogate model. These surrogates allow fast predictions of velocity time-series or Fourier amplitude spectra where spectral-element simulations are not done (neural networks compute about 100,000 surrogates per second, while a single spectral-element simulation longs approximately 7 h on 48 cores). To quantify the uncertainty of the physical system under study, a global sensitivity analysis is undertaken to better understand how uncertain parameters affect the predicted state of the system. We present two sampling-based estimation methods, the so-called “pick-freeze” Sobol method and the Li and Mahadevan method, to quantify this uncertainty. The Sobol method requires approximately 500,000 simulations to achieve stability, whereas the Li and Mahadevan method requires only 30,000 simulations. Using the metamodel, both methods require only a few seconds to produce results, although the Li and Mahadevan method analyzes 10,000 simulations per second, compared to 3,000 for the Sobol method. The results indicate that the shear wave velocity in the physical system’s layer is the most influential parameter affecting the ground speed on the physical system. In contrast, at a reference station (without considering the geological properties of the physical system), the results show that the shear wave velocity in the first and third deep layers are the most influential parameters.}, archive = {J_EAAI}, author = {Mohamad Ali Noureddine and Florent De Martin and Rani El Meouche and Muhammad Ali Sammuneh and Fakhreddine Ababsa and Mickael Beaufils}, doi = {10.1016/j.engappai.2025.110613}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110613}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural network-based metamodel of synthetic seismograms: Application for uncertainty quantification}, volume = {151}, year = {2025}, }
• (2025). Partial convolution-simple attention mechanism-SegFormer: An accurate and robust model for landslide identification. EAAI, 151, 110612. (www)

  @article{ , abstract = {To achieve accurate and robust landslide identification, this study presents an advanced model based on the SegFormer architecture, named Partial Convolution-Simple Attention Mechanism-SegFormer (PConv-simAM-SegFormer). The model integrates partial convolution (PConv) to enhance spatial feature extraction capabilities and employs a simple attention mechanism (simAM), inspired by neuroscience, to optimize attention allocation. Additionally, a novel reverse transfer learning strategy is introduced to leverage features from older landslides, thereby improving the detection of new landslides. Experimental results on three large-scale publicly available datasets validate the modelś effectiveness: a mean Intersection over Union (mIoU) of 76.05% on the Landslide Research for Sichuan–Tibet Transportation Corridor dataset (LRSTTC), representing a 3.04% improvement over the baseline; an mIoU of 90.35% on the Bijie Landslide dataset (Bijie), with a 2.88% enhancement; and an mIoU of 97.34% on the Tibetan Plateau Lakes dataset (TPL), with a 0.81% enhancement. These results demonstrate the modelś high accuracy and robustness across diverse datasets and application scenarios, showcasing its significant potential for practical applications in the field of landslide detection.}, archive = {J_EAAI}, author = {Shuang Yang and Yuzhu Wang and Kai Zhao and Xiaocai Liu and Jingqin Mu and Xupeng Zhao}, doi = {10.1016/j.engappai.2025.110612}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110612}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Partial convolution-simple attention mechanism-SegFormer: An accurate and robust model for landslide identification}, volume = {151}, year = {2025}, }
• (2025). Deep learning for lameness level detection in dairy cows. EAAI, 151, 110611. (www)

  @article{ , abstract = {Lameness detection using raw sensor data is a very challenging task, as the data are devoid of specific information regarding predictors such as gait distribution, weight among legs, etc. We have addressed this challenge using a deep learning technique, named LLP-Cow (Lameness level predictor for Cow), which is an application of artificial intelligence (AI). For objective comparison, LLP-Cow is validated using CowScreeningDB, an unbalanced public dataset composed of sensor data. This dataset is recorded during the normal life of dairy cows. Hence, LLP-Cow models the normal behaviour of cows and consists of feature extraction, application-specific deep network and a voting system. The technique presented is able to model the behaviour of a cow for both binary and multiclass classification. The precision and specificity reported by our technique stand at 0.94 and 0.98 for multiclass and 0.91 and 0.90 for binary protocols for the best case scenario. Moreover, F1 measure, Matthews correlation coefficient and Kappa are 0.94, 0.91, and 0.91, respectively. The technique introduced provides a margin for human intervention through the use of a voting system at the classification stage. The technique presented is therefore an implemented AI system for cow lameness detection that offers room for exploration in terms of real time implementation.}, archive = {J_EAAI}, author = {Shahid Ismail and Moises Diaz and Miguel Angel Ferrer}, doi = {10.1016/j.engappai.2025.110611}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110611}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning for lameness level detection in dairy cows}, volume = {151}, year = {2025}, }
• (2025). Stereo matching on epipolar plane image for light field depth estimation via oriented structure. EAAI, 151, 110608. (www)

  @article{ , abstract = {Depth estimation plays a pivotal role in civil engineering such as road surface defect detection, as it serves as a valuable tool, offering high-precision and critical information about scene surface geometry. The Light Field captures both spatial and angular information of a scene, enabling precise depth estimation. The Epipolar Plane Image represents a specific 2-dimensional slice of the light field and is characterized by multiple depth-related lines. Previous epipolar plane image-based methods typically estimate depth maps by extracting the optimal slope for each line; however, they often neglect the visual relationships within this representation, leading to inaccuracies. In this paper, we explore the visual relationship of it and propose a novel visual feature, termed Oriented Structure, which can be utilized to compute scene depth. Similar to previous stereo matching-based methods, we design a new epipolar plane image-based cost volume to extract depth cues from this structure. The cost volume combines the occlusion robustness of epipolar plane image-based methods with the noise robustness of stereo matching-based methods, resulting in smoother depth maps with sharper edges. Building on the framework of existing stereo matching networks, we introduce an epipolar plane image-based stereo matching network for light field depth prediction. Finally, we conduct experiments using both synthetic and real datasets, demonstrating that our network produces higher-quality depth maps compared to previous state-of-the-art methods, ranking first (about 1.405 mean squared error) on the 4-dimensional light field benchmark. Additionally, we also apply our method to defect detection tasks, providing accurate depth information that leads to improved results.}, archive = {J_EAAI}, author = {Rongshan Chen and Hao Sheng and Ruixuan Cong and Da Yang and Zhenglong Cui and Sizhe Wang and Wei Ke}, doi = {10.1016/j.engappai.2025.110608}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110608}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Stereo matching on epipolar plane image for light field depth estimation via oriented structure}, volume = {151}, year = {2025}, }
• (2025). Dynamic momentum contrastive learning network for diabetic retinopathy grading. EAAI, 151, 110604. (www)

  @article{ , abstract = {Diabetic retinopathy is the leading cause of blindness among the global working population. Automated and accurate grading of diabetic retinopathy is crucial for the diagnosis and treatment of retinal diseases. However, challenges arise in grading due to class imbalance, inter-class similarity, intra-class variability, and the small scale of lesions in different stages of diabetic retinopathy. To address the issues of class imbalance and small lesion scales, the dynamic momentum contrastive learning network (DMCLNet) for diabetic retinopathy grading is proposed in this paper. Firstly, a new strategy is introduced for constructing positive and negative samples to enhance individual differences. Then, an encoder and a momentum encoder are used to extract features from the main image, the positive and negative samples, respectively. The dynamic balancing strategy is presented to update the multi-class feature queue and calculate the similarity matrix between the main image and each positive and negative sample for contrastive learning. A dual dimensional loss function guides the training of the proposed model to fully capture the subtle differences between images of different classes, which improves inter-class discrimination ability. Finally, channel attention and spatial attention mechanisms are applied to enhance disease-related fine-grained features and suppress irrelevant redundant information, which accurately identifies small-scale lesions and improves the precision of severity grading. Extensive comparative experiments and ablation studies on three public datasets demonstrate that DMCLNet outperforms other state-of-the-art methods. It can achieve superior diabetic retinopathy grading performance across multiple datasets without relying on pixel-level lesion annotation, with high accuracy and generalization.}, archive = {J_EAAI}, author = {Yanfei Guo and Chenglong Yang and Hangli Du and Yuanke Zhang and Fei Ma and Shasha Yuan}, doi = {10.1016/j.engappai.2025.110604}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110604}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic momentum contrastive learning network for diabetic retinopathy grading}, volume = {151}, year = {2025}, }
• (2025). Design and implementation of a soft Actor–Critic controller for a robotic arm. EAAI, 151, 110589. (www)

  @article{ , abstract = {In this study, reinforcement learning (RL) was employed to train and control a robotic arm with six degrees of freedom in a point tracking task. The movements of such a robotic arm are not intuitive because they are determined by the rotations of and lengths between its six joints. Given this consideration, RL can be employed for the robotic arm to learn from its mistakes and autonomously explore the effects of joint rotation of its end point position. This approach is mostly used in complex learning tasks, such as those in gaming, robotics, and automated driving. The proposed learning process was verified in a simulator, in which point tracking and obstacle avoidance tasks were performed using RL algorithms. The performance of the soft actor–critic (SAC), proximal policy optimization, advantage actor–critic, and trust region policy optimization algorithms was compared in the point tracking task, and the results indicated that the SAC algorithm outperformed the other algorithms in this task. Therefore, the SAC algorithm was used with an adequately designed reward function to conduct model training, enabling the robotic arm to achieve obstacle avoidance in an object picking task. The established model was verified to allow the robotic arm to learn from failed experiences, enable fast navigation to the target position, and facilitate rapid deployment in future application.}, archive = {J_EAAI}, author = {Ping-Huan Kuo and Chen-Ting Huang and Chen-Wen Chang and Po-Hsun Feng and Yu-Sian Lin}, doi = {10.1016/j.engappai.2025.110589}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110589}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Design and implementation of a soft Actor–Critic controller for a robotic arm}, volume = {151}, year = {2025}, }
• (2025). Active learning of optical path classification. EAAI, 151, 110582. (www)

  @article{ , abstract = {Creating classification models to predict whether an optical channel can provide a required level of transmission quality is a promising approach to automated path quality assessment in optical network design. The applicability of machine learning algorithms in this domain is limited, however, by the cost, effort, and time needed to collect sufficient labeled data for model creation. The necessary amount of labeled data may be substantially reduced by active learning. This is an iterative process, creating a sequence of models, where each model is used to select the most useful paths for a class labeling query which are then added to the training set for the next model. Such a learning scenario poses different challenges for machine learning algorithms than standard “passive” learning, since they have to deal with very small and often imbalanced data. This work examines how these challenges are handled by algorithms that have been found particularly useful for optical path classification by prior studies. The random forest and extreme gradient boosting algorithms are applied to active learning from a real dataset provided by a network operator, starting only from a handful of training instances. Uncertainty sampling and diversity sampling are used for query selection. Confidence-based and stability-based stopping criteria are used to determine when the process can be safely terminated. The results confirm that active learning is a useful approach to creating optical path classification models, making it possible to save more than a half of the cost needed to provide class labels for model creation.}, archive = {J_EAAI}, author = {Paweł Cichosz}, doi = {10.1016/j.engappai.2025.110582}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110582}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Active learning of optical path classification}, volume = {151}, year = {2025}, }
• (2025). Dynamic-demand vulnerability analysis of water networks using a two-index neural network algorithm. EAAI, 151, 110581. (www)

  @article{ , abstract = {User demand changes periodically throughout the day, and the water distribution network is multi-terminal, so a vulnerability assessment based on a static network cannot accurately describe the network. In view of the above two reasons, we first establish two indicators in this paper, path centrality index and node importance index, and the proposed indicators accurately quantify the importance of paths and nodes in the water distribution network. The dynamic demand water distribution network vulnerability is assessed through the proposed two-index neural network algorithm. Firstly, a novel feedforward neural network is proposed to obtain path and flow information, and the neural network structure is established according to the topology of the water distribution network. Then, based on this information, the path centrality index and node importance index are calculated, and the important node and path sets are found. Finally, according to the dynamic demand of each user, the supply–demand ratio in the faulty network is solved. Theoretical analysis proves that the time complexity of the algorithm is o ( 6 n 2 ) . Administrators can further optimize maintenance schedules based on the supply–demand ratio over time. Taking the widely studied North Marin Water District Network in California as an example, it is proved that the proposed two-index neural network algorithm has increased accuracy, faster computation than static methods. However, the proposed algorithm relies on the complete network topology, and the uncertainty of the network topology can be further explored in future research.}, archive = {J_EAAI}, author = {Baowen Zhang}, doi = {10.1016/j.engappai.2025.110581}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110581}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic-demand vulnerability analysis of water networks using a two-index neural network algorithm}, volume = {151}, year = {2025}, }
• (2025). Unlocking language boundaries: AraCLIP - Transforming arabic language and image understanding through cross-lingual models. EAAI, 151, 110577. (www)

  @article{ , abstract = {In the domain of image retrieval, the integration of text and images has been transformative, facilitating models that transcend language barriers. This paper introduces Arabic Contrastive Language-Image Pre-training (AraCLIP), an extension of the Contrastive Language-Image Pre-training (CLIP) model tailored for Arabic image retrieval. AraCLIP leverages the CLIP architecture, introducing Knowledge Distillation to transfer cross-modal knowledge from a pre-trained English model to an Arabic counterpart. Unlike existing multilingual models lacking Arabic contextual nuances, AraCLIP addresses biases in image retrieval tasks. Our methodology involves dataset preparation, incorporating a synthetic dataset of about 12.5M samples, which translated using a unique neural machine translation model for accurate English-to-Arabic translation. The training phase utilizes Knowledge Distillation, treating an English model as the teacher and a pre-trained Arabic model as the student. AraCLIP focuses on optimizing computational efficiency, maximizing cosine similarity, and minimizing Mean Squared Error. Our best model surpasses the state-of-the-art multilingual model by approximately 10% across various evaluation metrics, including Mean Reciprocal Rank (MRR) and Recall. In addition, it demonstrates competitive performance in ImageNet-based zero-shot classification tasks. We have also released several datasets to support text image-related tasks 1 . AraCLIP along with published resources will leverage enhanced capabilities for Arabic image retrieval and opens avenues for diverse applications.}, archive = {J_EAAI}, author = {Muhammad Al-Barham and Imad Afyouni and Khalid Almubarak and Ayad Turky and Ibrahim Abaker Targio Hashem and Ali Bou Nassif and Ismail Shahin and Ashraf Elnagar}, doi = {10.1016/j.engappai.2025.110577}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110577}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unlocking language boundaries: AraCLIP - Transforming arabic language and image understanding through cross-lingual models}, volume = {151}, year = {2025}, }
• (2025). Self-supervised learning of monocular depth estimators in autonomous vehicles with federated learning. EAAI, 151, 110572. (www)

  @article{ , abstract = {In recent years, artificial intelligence has been applied to improve intelligent transportation systems, with a special focus on developing autonomous vehicles. Monocular depth estimation is a crucial task in autonomous driving, offering a cost-effective alternative to binocular methods. While existing approaches for this task leverage self-supervised learning of deep neural networks, they have not taken into account key requirements for their use in autonomous vehicles, such as the need to preserve the privacy of samples collected by multiple vehicles, reduce network consumption during training, and optimize the computation cost distribution. In recent studies, the use of federated learning has shown notable benefits when addressing these requirements. Thus, we propose a novel method combining federated learning and deep self-supervision to enable training of monocular depth estimators with comparable efficacy and superior efficiency to current methods. Our evaluation experiments, using two public benchmarks of images captured by moving vehicles, show that our proposed method achieves near state-of-the-art performance, with test losses of 0.115 and 0.169 on each dataset requiring, on average, only 1k to 1.25k training steps and about 65.1% to 87.4% less data transfer per vehicle in each round than our baseline.}, archive = {J_EAAI}, author = {Elton F. de S. Soares and Carlos Alberto V. Campos}, doi = {10.1016/j.engappai.2025.110572}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110572}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-supervised learning of monocular depth estimators in autonomous vehicles with federated learning}, volume = {151}, year = {2025}, }
• (2025). Graph convolutional network for traffic incidents duration classification. EAAI, 151, 110570. (www)

  @article{ , abstract = {Traffic incidents are a primary cause of severe congestion in urban areas, making accurate forecasting of incident duration essential for effective traffic management systems. However, the inherent uncertainty associated with incidents presents significant challenges in predicting their durations. In this paper, we propose a novel deep neural network model for predicting and classifying traffic incident durations. To capture the dynamic nature of incidents, the model learns from time series data on traffic flow, speed, and occupancy. Additionally, it employs a graph neural network architecture to model the spatial relationships between sensors, while also accounting for factors such as time and incident type. By training the model with cross-entropy loss, we enable it to predict whether an incident’s duration will be long or short. Experimental results demonstrate that our model outperforms existing baselines, demonstrating the effectiveness of our proposed approach. Furthermore, we conduct a case study to visualize the impact of incidents and further validate the model’s predictive capability.}, archive = {J_EAAI}, author = {Lyuyi Zhu and Qixin Zhang and Xiangru Jian and Yu Yang}, doi = {10.1016/j.engappai.2025.110570}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110570}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph convolutional network for traffic incidents duration classification}, volume = {151}, year = {2025}, }
• (2025). Correction of medical image segmentation errors through contrast learning with multi-branch. EAAI, 151, 110564. (www)

  @article{ , abstract = {Medical image segmentation plays a pivotal role in computer-aided diagnosis. Despite the considerable advancements achieved by convolutional neural networks, segmentation errors such as under-segmentation and mis-segmentation remain significant obstacles, particularly in complex medical images where target regions often share similar features with the background. In this paper, we propose a generalized network called Correction of Medical Image Segmentation Errors through Contrast Learning with Multi-Branch Network (MEC) to address these biases and ultimately improve segmentation accuracy. MEC leverages a multi-branch architecture, where each branch is specifically designed to tackle distinct segmentation issues. The negative branch focuses on encoding mis-segmented regions, the positive branch captures accurately labeled regions, and the anchor branch refines the segmentation output from the base network. To effectively integrate the complementary information from these branches, we employ a contrastive learning strategy that aligns the semantic features of the anchor branch with those of the positive branch, while maintaining a clear distinction from the negative branch. This methodology enhances segmentation performance by capitalizing on the unique strengths of each branch. Experimental results on several medical image datasets demonstrate significant improvements in segmentation accuracy, thereby validating the effectiveness of our proposed approach.}, archive = {J_EAAI}, author = {Tianlei Gao and Lei Lyu and Nuo Wei and Tongze Liu and Minglei Shu and Yushui Geng}, doi = {10.1016/j.engappai.2025.110564}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110564}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Correction of medical image segmentation errors through contrast learning with multi-branch}, volume = {151}, year = {2025}, }
• (2025). Resilient object detection for autonomous vehicles: Integrating deep learning and sensor fusion in adverse conditions. EAAI, 151, 110563. (www)

  @article{ , abstract = {Autonomous vehicles (AVs) rely on advanced object detection systems to ensure safe navigation, especially under adverse weather conditions that can impair sensor visibility and introduce detection challenges. This manuscript provides a comprehensive analysis of state-of-the-art methodologies, focusing on deep learning frameworks, multi-sensor fusion techniques, and specialized datasets designed for AV object detection across various environmental conditions. We categorize approaches based on accuracy, computational efficiency, and resilience to challenging weather scenarios, offering insights into the strengths and limitations of each technique. Additionally, widely used datasets, such as KITTI and Waymo, along with synthetic and real-time datasets, are evaluated to assess their impact on detection accuracy in complex scenarios. While deep learning models demonstrate high accuracy, the integration of sensor fusion and transfer learning techniques further enhances robustness and adaptability. Our findings emphasize the importance of developing weather-resilient AV perception systems and provide recommendations for advancing object detection frameworks in autonomous driving applications.}, archive = {J_EAAI}, author = {Pardhu Thottempudi and Asral Bin Bahari Jambek and Vijay Kumar and Biswaranjan Acharya and Fernando Moreira}, doi = {10.1016/j.engappai.2025.110563}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110563}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Resilient object detection for autonomous vehicles: Integrating deep learning and sensor fusion in adverse conditions}, volume = {151}, year = {2025}, }
• (2025). Optimizing quadratic programming problems with intuitionistic fuzzy parameters using a neural network. EAAI, 151, 110562. (www)

  @article{ , abstract = {The integration of intuitionistic fuzzy theory in optimization problems has significantly enhanced the ability to handle complex, uncertain and imprecise scenarios. Despite these advancements, existing models have not adequately addressed the specific challenges of intuitionistic fuzzy quadratic programming problems (IFQPPs). This study fills this gap by proposing a novel recurrent neural network for IFQPPs. First of all, IFQPP is transformed into a multi-objective optimization problem using α , β -cuts. This technique allows to explore a wide range of possible solutions using various combinations of α , β -cuts. Next, the multi-objective problem is converted into a weighted problem coupled with duality theory to remodel into a single-layer recurrent neural network. To validate the proposed approach, theorems as well as lemmas have been constructed and proved at appropriate places. The proposed neural network model is illustrated using numerical examples to explain the methodology. Later, it is applied to a small-scale electrical power grid to demonstrate the practical utility and impact of the proposed approach.}, archive = {J_EAAI}, author = {Pooja Rani and Sumati Mahajan}, doi = {10.1016/j.engappai.2025.110562}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110562}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing quadratic programming problems with intuitionistic fuzzy parameters using a neural network}, volume = {151}, year = {2025}, }
• (2025). Two-sided matching optimization model for green housing technology selection based on hesitant 2-tuple linguistic rough numbers. EAAI, 151, 110559. (www)

  @article{ , abstract = {In two-sided matching decision problems, the matching objects with different knowledge, experiences, and cultures provide linguistic assessments using diverse or multi-granular sets with a factor that the information provided is hesitant in nature due to different opinions given by experts. In the proposed approach, the hesitant 2-tuple linguistic information is integrated with rough approximations to develop the two novel approaches called hesitant rough numbers and hesitant 2-tuple linguistic rough numbers. The proposed novel approximations are implemented on a two-sided matching optimization model to study hesitant multi-granular uncertainty. Firstly, the matching objects provide their evaluations in the form of hesitant multi-granular terms converted into hesitant 2-tuple linguistic rough numbers. Secondly, certain optimization models based on hesitant 2-tuple linguistic rough approximations are constructed to compute the criteria weights using incomplete information. In hesitant 2-tuple linguistic rough optimization models, the maximizing deviation technique is used to find the distance between two proposed novel coefficients. To maximize the level of satisfaction with matching objects, a hesitant 2-tuple linguistic rough optimization model is developed to evaluate the overall satisfaction degree and stability of matching objects. The significance of the proposed two-sided matching optimization model is illustrated with a case study of matching between the green building technology supply and demand. The out-performance of the proposed model is highlighted by a comparison analysis with existing approaches to analyze that it can provide hesitant multi-granular rough flexibility and deals with incomplete information regarding criterion weights.}, archive = {J_EAAI}, author = {Musavarah Sarwar and Ghous Ali and Wajeeha Gulzar and Muhammad Akram and Dragan Pamucar}, doi = {10.1016/j.engappai.2025.110559}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110559}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Two-sided matching optimization model for green housing technology selection based on hesitant 2-tuple linguistic rough numbers}, volume = {151}, year = {2025}, }
• (2025). Attention guided class activation maps for boosting weakly supervised semantic segmentation. EAAI, 151, 110556. (www)

  @article{ , abstract = {Weakly Supervised Semantic Segmentation (WSSS) has garnered significant attention for its ability to utilize weaker labels in place of expensive pixel-level annotations while maintaining commendable performance. Class Activation Maps (CAM) can still possess target localization capabilities without pixel-level annotations, and are thus widely used as pseudo labels to supervise subsequent segmentation tasks. With the continuous advancement of artificial intelligence, generating high-quality CAM by combining the strengths of Convolutional Neural Networks (CNN) and Transformer architectures has received widespread attention. As a classic architecture, Conformer adopts a parallel structure and has been applied in multiple WSSS models. We observed that the attention matrices of different levels of Transformer blocks in Conformer exhibit significant characteristic differences, and these matrices effectively capture the correlations between different regions. Based on this observation, we propose a method called Attention-Guided Class Activation Map (AG-CAM), which selectively utilizes attention matrices at different levels to enhance features for various purposes. Detailed experiments on common datasets have shown that the proposed AG-CAM method significantly improves the quality of class activation maps. Our work provides a more precise solution for WSSS, thereby demonstrating immense potential and value in real-world applications where data annotation is scarce.}, archive = {J_EAAI}, author = {Junhui Li and Lei Zhu and Wenwu Wang and Yin Gong}, doi = {10.1016/j.engappai.2025.110556}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110556}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Attention guided class activation maps for boosting weakly supervised semantic segmentation}, volume = {151}, year = {2025}, }
• (2025). Riemann–Liouville fractional exponential sampling type neural network kantorovich operators. EAAI, 151, 110477. (www)

  @article{ , abstract = {This research introduces a novel neural network operator that leverages fractional calculus and Kantorovich-type methods to enhance function approximation capabilities. The incorporation of a flexible parameter a , derived from the fractional Riemann–Liouville integral operator, significantly improves performance, achieving up to 99.5% reduction in maximum absolute error and 19.63% reduction in root mean squared errors. A robust theoretical framework, including convergence analysis and multivariate extensions, supports the operator’s effectiveness. Furthermore, we demonstrate the effectiveness of linear combinations of these operators, showcasing their enhanced approximation properties. In Example 2, a linear combination of three operators is constructed for r = 3 , resulting in significant performance gains. The proposed operators exhibit superior performance across diverse applications. In image processing, we observed substantial improvements in image quality metrics. For the ’Mona Lisa’ image, we achieved up to 5.68% increase in Peak Signal-to-Noise Ratio (PSNR) and 4.18% increase in Structural Similarity Index Measure (SSIM) compared to state-of-the-art (SOTA) methods such as nearest neighbor, bilinear, and bicubic interpolation. Similarly, for the ’Baboon’ image, improvements of up to 5.45% in PSNR and 5.25% in SSIM were achieved. Moreover, the operators effectively model real-world phenomena. We applied the operators to model the trend of COVID-19 cases and deaths in Europe, North America, and Asia using real-world data from the World Health Organization (WHO) dashboard. Our analysis, supported by visual comparisons in Figs. 12 – 17 and quantitative evaluations in Tables 18 and 19 , reveals that the proposed operators consistently outperform existing methods, including those presented by Kadak (2023b), achieving significant reductions in Root Mean Squared Error (RMSE) and absolute error for both daily cases and deaths across all regions. Notably, the improvement in predicting daily deaths is particularly substantial, with reductions in RMSE ranging from 21.68% in Asia to 46.8% in America and reductions in absolute error ranging from 24.17% in Asia to 35.09% in America. To enhance visualization and clarity, flowcharts have been included in key sections of the paper, such as Numerical Simulation and Performance Evaluation ( Fig. 8 ), COVID-19 Case and Death Modeling ( Fig. 11 ), and Image compression/decompression process ( Fig. 18 ). These findings highlight the versatility and high performance of the proposed operators across diverse applications, demonstrating their significant potential for addressing complex challenges in machine learning, image processing, and other critical domains.}, archive = {J_EAAI}, author = {Artan Berisha and Fesal Baxhaku and Purshottam Narain Agrawal and Behar Baxhaku}, doi = {10.1016/j.engappai.2025.110477}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110477}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Riemann–Liouville fractional exponential sampling type neural network kantorovich operators}, volume = {151}, year = {2025}, }
• (2025). A novel koopman-based assistant features network for long and short-term carbon emission prediction. EAAI, 151, 110366. (www)

  @article{ , abstract = {The growing threat of global warming has heightened the importance of accurately predicting carbon dioxide emissions, a key factor in climate change mitigation. Previous studies have developed carbon emission prediction models using aggregated yearbook statistics, which are limited by spatiotemporal constraints and data scarcity. These models often assume data stationarity, overlooking the dynamic nature of carbon dioxide emission, leading to less reliable predictions that fall short of informing effective urban emission reduction strategies. This paper presents a novel approach by introducing two first daily, city-specific, multi-source carbon dioxide emissions datasets. It integrates global carbon dioxide emissions grid data from CarbonMonitor-graced with concentration measurements from the global carbon column total observation network. To make better use of these data features, we introduce a novel deep learning model, the K oopman A ssistant F eature Net work (KAFNet). By incorporating carbon dioxide concentration and calendar features as additional inputs, we apply the Koopman Operator theory, which is adept at analyzing the time-varying dynamics of complex systems. This approach allows for a comprehensive consideration of the underlying dynamics within carbon emissions data, enabling end-to-end optimization of predictive targets. Our empirical analysis reveals that our model significantly outperforms the sub-optimal models it was compared against, with an average reduction of 34.5% in Mean Absolute Error and 18.8% in Mean Squared Error. This enhancement in predictive accuracy provides a robust tool for capturing the evolving trends in carbon dioxide emissions, thereby offering a solid quantitative foundation to support data-driven decision-making in urban carbon reduction policies.}, archive = {J_EAAI}, author = {Daishun Cui and Qin Jiwei and Dezhi Sun and Xizhong Qin and Fei Shi}, doi = {10.1016/j.engappai.2025.110366}, journal = {Engineering Applications of Artificial Intelligence}, month = {7}, pages = {110366}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel koopman-based assistant features network for long and short-term carbon emission prediction}, volume = {151}, year = {2025}, }
• (2025). Data-efficient federated learning for edge network intrusion detection. EAAI, 150, 110685. (www)

  @article{ , abstract = {The use of deep learning enabled Network Intrusion Detection Systems (NIDS) has enabled greater efficiency in classifying and responding to cyber threats. As networks continue to grow and increasingly become interconnected, novel NIDS platforms need to be engineered to enhance data-efficiency and data privacy so that NIDS may share what they learned with each other. To accomplish this, we developed a Federated Learning (FL) NIDS platform where clients learn from raw network traffic payloads and share their model weights with a central server. This server aggregates the received weights and redistributes to each client. Our platform demonstrates the feasibility of this approach using both realistic edge hardware and simulation environments. Here, we examine an environment where each network has heterogeneous data distributions but must keep the training data locally. The proposed neural network architectures must also be relatively simple to enable learning at the edge on low-compute devices. Through experimentation, our federated NIDS platform powered by a simple convolutional neural network utilizing the FedProx FL aggregation algorithm achieved over 82% accuracy in a multi-class setting with a Cohen’s Kappa statistic of 0.6853 after 60 rounds of FL.}, archive = {J_EAAI}, author = {David A. Bierbrauer and Sean M. Coffey and Mikal R. Willeke and John D. Beggs and Nathaniel D. Bastian}, doi = {10.1016/j.engappai.2025.110685}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110685}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-efficient federated learning for edge network intrusion detection}, volume = {150}, year = {2025}, }
• (2025). Multimodal data imputation and fusion for trustworthy fault diagnosis of mechanical systems. EAAI, 150, 110663. (www)

  @article{ , abstract = {The presence of missing values in the collected data due to sensor failure, communication interruption, or environmental interference can greatly diminishes the trustworthiness of fault diagnosis for mechanical systems. Therefore, this study proposes and evaluates a novel multimodal data imputation and fusion method to perform the trustworthy fault diagnosis of mechanical systems. First, a generative adversarial imputation network, termed as the L2 regularization temporal–spatial generative adversarial imputation network (L2-TSGAIN), is developed. This L2-TSGAIN network, based on a temporal–spatial feature extraction module and L2 regularization loss function, can comprehensively extract data features from both temporal and spatial perspectives, thus achieving high-quality imputation of anomalous sensor data. Subsequently, a multi-input single-output autoencoder (MISO-AE) is designed to extract a universal representation of the imputed data from different modalities and recover features in the fusion data. Finally, the fusion data from different health states of mechanical systems are input into a convolutional neural network classifier to perform fault diagnosis. Experiment validations, considering the presence of missing values in sensor data, have been carried out on the planetary transmission system and gearbox test bench. Compared with several mainstream data imputation methods for fault diagnosis, the optimal diagnostic accuracy of 99.68 % and 100 % on these two datasets can be obtained using the proposed method, respectively, confirming its superior performance and reliability. Thus, the proposed method can provide a trustworthy fault diagnosis tool for mechanical systems in industrial scenarios considering anomalous sensor data.}, archive = {J_EAAI}, author = {Jie Zhang and Yun Kong and Qinkai Han and Tianyang Wang and Mingming Dong and Hui Liu and Fulei Chu}, doi = {10.1016/j.engappai.2025.110663}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110663}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodal data imputation and fusion for trustworthy fault diagnosis of mechanical systems}, volume = {150}, year = {2025}, }
• (2025). Atrous spatial pyramid pooling with swin transformer model for classification of gastrointestinal tract diseases from videos with enhanced explainability. EAAI, 150, 110656. (www)

  @article{ , abstract = {Accurate and early identification of gastrointestinal (GI) lesions is crucial for treating and preventing GI diseases, including cancer. Automated computer-aided diagnosis methods can assist physicians in early and accurate detection. Video classification of GI endoscopic videos is challenging due to the complexity and variability of visual data. This research proposes a novel method for classifying GI diseases using endoscopic videos. Leveraging the public HyperKvasir dataset, we applied preprocessing algorithms to enhance GI frames by removing noise and artifacts with morphological opening and closing techniques, ensuring high-quality visuals. We addressed dataset imbalance by proposing a novel algorithm. Our hybrid model, Atrous Spatial Pyramid Pooling with Swin Transformer (ASPPST), combines advanced Convolutional Neural Networks and the Swin Transformer to classify GI videos into 30 distinct classes. We incorporated Gradient-Class Activation Mapping (Grad-CAM) in ASPPST's final layer to improve model explainability. The proposed model achieved 97.49 % accuracy in classifying 30 GI diseases, outperforming other transfer learning models and transformers by 8.04 % and 3.99 %, respectively. It also demonstrated a precision of 97.80 %, recall of 97.77 %, and an F1 score of 97.75 %, showcasing robustness across metrics. The high accuracy of ASPPST makes it suitable for real-world use, delivering fewer errors and more precise results in GI endoscopy video classification. Our approach advances artificial intelligence (AI) in computer vision and deep learning for biomedical engineering applications. Grad-CAM integration enhances transparency, boosting clinician trust and adoption of AI tools in diagnostic workflows.}, archive = {J_EAAI}, author = {Arefin Ittesafun Abian and Mohaimenul Azam Khan Raiaan and Mirjam Jonkman and Sheikh Mohammed Shariful Islam and Sami Azam}, doi = {10.1016/j.engappai.2025.110656}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110656}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Atrous spatial pyramid pooling with swin transformer model for classification of gastrointestinal tract diseases from videos with enhanced explainability}, volume = {150}, year = {2025}, }
• (2025). Solution and application of two-dimensional seismic wavefield evolution based on physics-informed neural networks. EAAI, 150, 110652. (www)

  @article{ , abstract = {Physics-Informed Neural Networks (PINN) integrate partial differential equations, initial conditions, and boundary conditions into the loss function to predict the solutions of partial differential equations, and have already demonstrated their value in solving two-dimensional (2D) seismic wavefields. However, when dealing with wave problems involving boundary conditions, the added complexity of boundary conditions can lead to imbalanced convergence rates among different loss terms, which may affect both the efficiency and accuracy of the computations. Moreover, the need to retrain the model for different problems limits the flexibility of its application. Therefore, this paper introduces an adaptive weight balancing method and presents a 2D wave simulation based on Self-Adaptive PINN (SA-PINN). This method automatically adjusts the weights in the loss function, improving the solving performance. Additionally, to improve the computational efficiency of PINN in solving similar wave problems, a transfer learning strategy is adopted. By leveraging the similarities between the PINN models of related wave problems, this strategy enhances the generalization ability of PINN when dealing with variations in source location and medium wave speed. Numerical examples in semi-infinite domains and V-shaped valleys demonstrate that this method effectively achieves intelligent and efficient simulation of 2D seismic wavefields, providing a more efficient and intelligent solution for complex seismic wave problems.}, archive = {J_EAAI}, author = {Zhihui Zhu and Zong Wang and Yang Feng and Weiqi Zheng}, doi = {10.1016/j.engappai.2025.110652}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110652}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Solution and application of two-dimensional seismic wavefield evolution based on physics-informed neural networks}, volume = {150}, year = {2025}, }
• (2025). Adaptive human in the loop system for identifying non-optimal states in natural product manufacturing process. EAAI, 150, 110650. (www)

  @article{ , abstract = {In the extraction of natural products, the identification of non-optimal production states is pivotal for ensuring consistent product quality. Currently, there is a deficiency in online, automated detection methods. This study introduces an online machine vision strategy in a real industrial setting to maintain optimal production state. Specifically, the strategy incorporates an adaptive human in the loop deep learning approach to select high-value samples. This method achieves over 90 % accuracy with fewer training samples, effectively addressing the challenges posed by the low-value density characteristic of industrial data. Additionally, a convolutional neural networks-transformer framework is employed as a classifier for video data to meet the demands of time-series data. To enhance the efficiency of processing multiple video streams, we have implemented a knowledge distillation technique to lighten the model. Finally, this model has been deployed in an actual industrial environment for online monitoring of three extraction devices. The system encapsulates the expertise of engineers to standardize the criteria for assessing production states. This integration of innovative technologies ensures a more reliable and efficient extraction process, meeting the industry's need for consistent product quality.}, archive = {J_EAAI}, author = {Qilong Xue and Yang Yu and Shixin Cen and Yequan Yan and Jiping Pang and Ping Li and Yehan Hou and Lei Wang and Zheng Li}, doi = {10.1016/j.engappai.2025.110650}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110650}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive human in the loop system for identifying non-optimal states in natural product manufacturing process}, volume = {150}, year = {2025}, }
• (2025). A chinese medical named entity recognition method considering length diversity of entities. EAAI, 150, 110649. (www)

  @article{ , abstract = {Extracting clinical entity concepts from professional medical materials is crucial for medical information analysis and knowledge extraction. Whereas, the Chinese medical named entity recognition (CMNER) task faces challenges due to the knowledge specialization and the diversity in entity lengths. To address these challenges, a novel method by considering length diversity of entities for CMNER is proposed, focusing on the integration of local information based on the predominance of large language models (LLMs). The method pre-trains a bidirectional encoder representation from transformers (BERT) based on open Chinese medical texts and designs a multi-dimensional convolutional residual module to enhance the semantic information for characters. This module effectively mines local information across various ranges and employs a local channel self-attention block to integrate this information, establishing a link between local information and entity length. Meanwhile, an adaptive optimization strategy for a learning rate is designed to improve the method's ability to search for the optimal solution. Experimental results reveal that, compared with state-of-the-art models, our approach achieves the optimal Recall and F1 , especially Recalls achieve 94.50 % (p < 0.05) and 93.51 % (p < 0.05) with effective performance in current task. The ablation results suggest that incorporating local information within 1–7 characters effectively addresses the challenges mentioned, highlighting the potential of our method to advance CMNER task.}, archive = {J_EAAI}, author = {Hongyu Zhang and Long Lyu and Weifu Chang and Yuexin Zhao and Xiaoqing Peng}, doi = {10.1016/j.engappai.2025.110649}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110649}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A chinese medical named entity recognition method considering length diversity of entities}, volume = {150}, year = {2025}, }
• (2025). Improving weak magnetic detection of ferromagnetic material defects diagnostics via transfer learning-enhanced residual networks. EAAI, 150, 110647. (www)

  @article{ , abstract = {The use of weak magnetic detection technology to identify defect sizes in ferromagnetic materials poses significant challenges due to the large volume of data and the relatively low prediction accuracy of conventional methods. Therefore, this paper proposes an improved Residual Networks (ResNet18) model that integrates transfer learning and channel attention mechanisms. Compared to traditional methods, the following improvements have been made: First, transfer learning has significantly reduced the data volume and time cost required for training from scratch, enhancing the model's generalization capability. Second, we have added a channel attention mechanism to the ResNet18 model, which involves calculating the importance of each channel through adaptive average pooling and fully connected layers, and generating channel weights using a Sigmoid function. This improvement allows the model to more accurately focus on features with higher relevance to defect sizes. Experimental results demonstrate that for grayscale images with defect lengths of 50 mm, depths of 2 mm, and widths of 1, 2, 3, 4, and 5 mm, the prediction accuracies reached 100 %, 100 %, 98.84 %, 99.58 %, and 100 %, respectively. For grayscale images with defect lengths of 50 mm, widths of 2 mm, and depths of 1, 2, 3, 4, and 5 mm, the prediction accuracies were 99.68 %, 100 %, 99.63 %, 100 %, and 99.60 %, respectively. Compared to the traditional ResNet18 model, the improved model not only enhances the accuracy of defect size prediction but also exhibits greater robustness, providing a new and effective method for defect classification in weak magnetic detection of ferromagnetic materials.}, archive = {J_EAAI}, author = {Yu Chen and Liangliang Li and Zhengxiang Ma and Xinling Wen and Jiabao Pang and Weitao Yuan}, doi = {10.1016/j.engappai.2025.110647}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110647}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving weak magnetic detection of ferromagnetic material defects diagnostics via transfer learning-enhanced residual networks}, volume = {150}, year = {2025}, }
• (2025). Self-adaptive production scheduling for discrete manufacturing workshop using multi-agent cyber physical system. EAAI, 150, 110638. (www)

  @article{ , abstract = {At present, the production control process of a discrete manufacturing workshop is characterized by high concurrency, mixed production lines and difficulty in prediction, which lead to uncertainty caused by dynamic disturbances and challenges in production control. Traditional system architectures struggle to handle these uncertainties flexibly and adaptively. To address these issues, an adaptive production scheduling system for the workshop is proposed, utilizing the Multi-agent Cyber Physical System (CPS-MAS) framework. This system integrates self-organization mechanisms and self-adaptive decision-making mechanisms to achieve cooperative optimal control of manufacturing resources. Using multi-agent technology, the resource model in the information space is encapsulated into an intelligent Cyber Physical System (CPS)-Agent model with cognitive interaction and autonomous decision-making capabilities. The improved contract network protocol (CNP) is utilized to the constructed agent, enabling their collaboration and competition to support the self-organization, negotiation, and assignment of manufacturing tasks. Based on multi-agent real-time perception and interactive negotiation, an adaptive control model of the manufacturing process is constructed based on Proportion Integration Differentiation (PID) control principle. This model is trained with the multi-layer perceptron that integrates an attention mechanism. The production strategy and parameters of the agent cooperative network are dynamically adjusted to enable dynamic decision-making optimization under disturbances. The proposed method is verified by experiments in scenarios involving machine failure, emergency order insertion and due date changes, proving its effectiveness.}, archive = {J_EAAI}, author = {Jie Chen and Zequn Zhang and Liping Wang and Dunbing Tang and Qixiang Cai and Kai Chen}, doi = {10.1016/j.engappai.2025.110638}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110638}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-adaptive production scheduling for discrete manufacturing workshop using multi-agent cyber physical system}, volume = {150}, year = {2025}, }
• (2025). A pencil lead break-triggered, adversarial autoencoder-based approach for rapid and robust rail damage detection. EAAI, 150, 110637. (www)

  @article{ , abstract = {Detecting early-stage damage is essential for railway maintenance, ruling out potential risks that could undermine railway ride comfort and safety. Ultrasonic testing methods, featuring high precision and non-destructive characteristics, have gained widespread use for on-site inspections in modern railway systems. However, current ultrasonic testing remains a highly complex technique that requires expensive ultrasonic devices and trained professionals for operation. This study presents a novel approach for rail damage detection utilizing a disposable mechanical pencil. By intentionally breaking the pencil lead on rail surface, the accumulated potential energy is released in the form of ultrasonic bursts which are acquired by sensors mounted on the rail. The rail damage diagnosis is empowered by an adversarial autoencoder (AAE) which learns representations of ultrasonic signals induced by pencil lead break (PLB). A damage-sensitive indicator is developed based on the Jensen-Shannon Divergence (JSD) between the AAE model output distributions of the baseline and an unknown signal, facilitating rapid and accurate damage diagnosis. Both laboratory experiments and on-site verifications were conducted to validate the proposed approach. The results demonstrate the effectiveness of the damage detection framework in identifying rail damage, exhibiting excellent robustness and reliability. Comparative studies are also conducted to demonstrate the adaptability and effectiveness of the proposed method against field testing environments. The research outcomes of this study will significantly contribute to the development of more efficient on-site inspection techniques for railway maintenance and sustainability.}, archive = {J_EAAI}, author = {Da-Zhi Dang and Bo-Yang Su and You-Wu Wang and Wai Kei Ao and Yi-Qing Ni}, doi = {10.1016/j.engappai.2025.110637}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110637}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A pencil lead break-triggered, adversarial autoencoder-based approach for rapid and robust rail damage detection}, volume = {150}, year = {2025}, }
• (2025). Two-dimensional temperature field prediction with in-situ data in metal additive manufacturing using physics-informed neural networks. EAAI, 150, 110636. (www)

  @article{ , abstract = {Accurately predicting the temperature field in metal additive manufacturing (AM) processes is critical for preventing overheating, adjusting process parameters, and ensuring process stability. While physics-based computational models offer precision, they are often time-consuming and unsuitable for real-time predictions. Machine learning models, on the other hand, rely heavily on high-quality datasets, which can be costly and difficult to obtain in the metal AM domain. Existing studies on physics-informed neural networks (PINNs) have made progress in integrating physics with machine learning but often lack in-situ data integration, which is essential for capturing real-time thermal dynamics. Additionally, their methodologies are typically heavily dependent on specific process characteristics, limiting their flexibility. Our work addresses these gaps by introducing a PINN-based framework specifically designed for temperature field prediction in metal AM. The framework incorporates in-situ temperature data gathered during the manufacturing process, combining it with physics-informed inputs and a custom loss function. The approach is demonstrated through two case studies. In the first case, using a small set of experimental data, the model achieves an error below 3 % with a mean absolute error (MAE) of 11 °C. In the second case, using simulation data, the model achieves an error below 1 % with an MAE of 7 °C. In addition, the framework shows promising adaptability for different metal AM scenarios with different geometries, deposition patterns, and process parameters.}, archive = {J_EAAI}, author = {Pouyan Sajadi and Mostafa Rahmani Dehaghani and Yifan Tang and G. Gary Wang}, doi = {10.1016/j.engappai.2025.110636}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110636}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Two-dimensional temperature field prediction with in-situ data in metal additive manufacturing using physics-informed neural networks}, volume = {150}, year = {2025}, }
• (2025). A real-time onboard compressor stall warning method based on attention multiple sensors fusion and lightweight network. EAAI, 150, 110635. (www)

  @article{ , abstract = {Compressor stall is one of the critical faults in aero-engines. An effective stall warning model can assist operators and aviation power systems in taking timely measures to avoid or minimize the impact of compressor stall on flight. Real-time onboard stall warning systems for compressors demand models that possess sufficient accuracy, high reliability, and fast execution speed. Therefore, this paper proposes a novel lightweight network based on attention mechanism for multiple sensors feature fusion, named attention feature fusion lightweight network (AFF-LWNet). Specifically, the network first accomplishes multi-sensor feature fusion through an attention feature fusion block. It then learns input features through a lightweight network with two lightweight feature extraction units and finally employs a multi-layer perceptron (MLP) to output stall warning signals. To verify the feasibility of this approach, the proposed method is evaluated on the aero-engine compressor stall dataset. The results demonstrate that the proposed method achieves an average testing accuracy of 99.453 % and an actual average lead time of 241.87ms on the stall dataset, which surpasses the other five competing methods. Therefore, we believe that the proposed method can effectively achieve real-time onboard stall warning for aero-engine compressors with outstanding performance.}, archive = {J_EAAI}, author = {Huijie Jin and Yong-Ping Zhao and Zhiqiang Wang}, doi = {10.1016/j.engappai.2025.110635}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110635}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A real-time onboard compressor stall warning method based on attention multiple sensors fusion and lightweight network}, volume = {150}, year = {2025}, }
• (2025). Explainable graph convolutional network based on catastrophe theory and its application to group activity recognition. EAAI, 150, 110634. (www)

  @article{ , abstract = {Graph convolutional networks (graph models for short) are crucial for understanding model decisions through mathematical white-box interpretation, which can radically improve the performance and credibility of downstream artificial intelligence applications. To address the limitations of existing interpretability of over-smoothing and over-squashing, we propose an explainable graph model based on nonlinear catastrophe theory and apply it to group activity recognition to validate the usefulness of interpretability. (1) We introduce catastrophe mathematical theory to explore the internal processes of graph models and construct the explainable dynamical equations of the graph convolutional network; (2) When graph node features lose uniqueness, leading to over-smoothing, which reduces the discriminative power of the graph model, we propose a mathematical method to predict over-smoothing; (3) In response to the over-squashing of the node feature values that is excessively compressed, we design a channel expansion unit to extend the transmission paths of graph nodes and alleviate the over-squashing in the graph structure. Finally, we apply our model to group activity recognition tasks to capture complex interactions within groups. We obtain the competitive results on five publicly available graph structure datasets (Actor, Chameleon, Texas, Cornell, Cora) and our self-built group activity dataset. Our model can effectively capture node and graph-level features with stronger generalization capabilities. For complex and diverse real-world group activity data, our model offers intuitive graph-level explanations for group activity analysis. Through the analysis of over-smoothing and over-squashing, our method extends new theoretical approaches in explainable artificial intelligence.}, archive = {J_EAAI}, author = {Junpeng Kang and Jing Zhang and Lin Chen and Hui Zhang and Li Zhuo}, doi = {10.1016/j.engappai.2025.110634}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110634}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable graph convolutional network based on catastrophe theory and its application to group activity recognition}, volume = {150}, year = {2025}, }
• (2025). A knowledge graph for the vulnerability of construction safety system in megaprojects based on accident inversion. EAAI, 150, 110630. (www)

  @article{ , abstract = {The increasing vulnerability of construction safety systems in megaprojects (CSSMs) poses significant challenges to their safety management and control. To address this obstacle, this study retrodicts the accidents based on text mining using the Bidirectional Encoder Repre-sentations from Transformers Topic (BER-Topic) model to uncover topic and topic words related to the vulnerabilities of CSSMs. The vulnerability indicator system (VIS) is established by considering the exposure, sensitivity, and adaptability of the vulnerability of CSSMs. Subsequently, an improved Decision-making Trial and Evaluation Laboratory (DEMATEL) method based on association rules is proposed to reduce the subjectivity in assigning weights to vulnerability indicators, and a topological network based on complex network is constructed to identify the characteristics of VIS. Based on this, a knowledge graph of vulnerabilities in CSSMs is developed. Finally, taking into account the occurrence probability and the actual losses incurred of vulnerability indicators, a vulnerability assessment model for CSSMs is proposed. The research findings are: 1) Based on the BER-Topic model, 32 topics and topic words related to the vulnerability of CSSMs are mined. 2) A VIS for CSSMs is constructed, including 42 indicators across three dimensions of exposure (19), sensitivity (14), and adaptability (9), involving four aspects: humans, machines, environment, and management. 3) The key points for vulnerability management and control in CSSMs are Inaccurate implementation of geological remediation plans, Rusting of connecting components, and Unlicensed personnel operating, among others, which have strong intermediary roles.}, archive = {J_EAAI}, author = {Yingliu Yang and Pengcheng Xiang}, doi = {10.1016/j.engappai.2025.110630}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110630}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A knowledge graph for the vulnerability of construction safety system in megaprojects based on accident inversion}, volume = {150}, year = {2025}, }
• (2025). Robust sparse discriminative least squares regression for image classification. EAAI, 150, 110626. (www)

  @article{ , abstract = {Discriminative Least Squares Regression (DLSR) is a method used for multi-class classification tasks that expands the distance between different classes through an ε -dragging technique. However, it also amplifies the differences in intra-class regression targets. Moreover, the samples contain a significant amount of noise, which negatively affect the classification performance. To mitigate these problems, we propose Robust Sparse Discriminative Least Squares Regression (RSDLSR) approach to enhance the model's discriminative power. Firstly, we maintain the original data structure by matrix decomposition in the label space. Secondly, the noise is fitted using sparse constrained noise matrix to enhance the model's denoising ability. Furthermore, we select important features from label space using a linear discriminant analysis criterion to minimize the influence of redundant features. Finally, l 2 , 1 norm constraint is imposed on the relaxation matrix to improve the sparsity and robustness of the model. Comparative evaluations demonstrate that our proposed method exhibits significant advantages over various existing methods across different classification tasks.}, archive = {J_EAAI}, author = {Zhangjing Yang and Dingan Wang and Pu Huang and Minghua Wan and Fanlong Zhang}, doi = {10.1016/j.engappai.2025.110626}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110626}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust sparse discriminative least squares regression for image classification}, volume = {150}, year = {2025}, }
• (2025). A physics informed convolution neural network for spatiotemporal temperature analysis of concrete dams. EAAI, 150, 110624. (www)

  @article{ , abstract = {Structural health monitoring is indispensable throughout the life cycle of dams, and the loading conditions determines the reliability of the assessment. Among them, temperature plays an important role on the behavior of arch dams, which are sparsely monitored in practice. How to use these sparsely measured data to obtain the accurate spatiotemporal temperature field becomes a critical problem. This study proposes a physics informed convolutional neural network for spatiotemporal temperature field of arch dams. A dual thread convolutional neural network considers the effects of spatiotemporal and temporal variables distinctively. The proposed model is validated using measured data from an existing arch dam. Compared with applied convolutional neural network, the proposed model improves the accuracy of temperature field reconstruction by 18 % and reduces reliance on measured data. Benefit of consideration of the continuity and heat transfer, the spatial distribution of the temperature field is more reasonable in continuity, and can retain accuracy even with limited monitoring data. The proposed model can provide the actual spatiotemporal non-uniform temperature field of the arch dam, providing basic data for the analysis and safety evaluation of arch dams throughout their life-cycle.}, archive = {J_EAAI}, author = {Jiaqi Yang and Jinting Wang and Feng Jin and Jianwen Pan}, doi = {10.1016/j.engappai.2025.110624}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110624}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A physics informed convolution neural network for spatiotemporal temperature analysis of concrete dams}, volume = {150}, year = {2025}, }
• (2025). A decision modeling approach for the development of sustainable transportation oil companies. EAAI, 150, 110623. (www)

  @article{ , abstract = {Constructing sustainable transportation infrastructure to address climate change by reducing carbon dioxide and greenhouse gas emissions is unfeasible without the involvement of international oil companies (IOCs). Identifying the most sustainable IOC can significantly enhance legitimacy, corporate image, brand value, transparency, and reputation. The environmental impact of oil transportation is crucial to the sector's long-term growth, and modeling IOCs can guide decision-making that aligns with regulatory, environmental, and societal expectations, ensuring successful project implementation. Modeling IOCs presents multiple-attribute decision-analysis (MADA) challenges. Previous research proposed a decision matrix that crossed IOC alternatives with attributes, sub-attributes, and measurement items, utilizing assessments from 483 experts across 11 IOCs based on 2 attributes, 9 sub-attributes, and 47 measurements. Despite the literature review, challenges such as score deviations in the ranking method, as well as informational vagueness, ambiguity, and uncertainty in both weighting and ranking processes, remain unsolved, and early MADA methods exhibit theoretical flaws. This study aims to formulate and develop a decision modeling approach using multi-attribute ideal-real comparative analysis (MAIRCA) and fuzzy weighted zero inconsistency (FWZIC) within a homogeneous interval-valued intuitionistic fuzzy rough set (IIFRS) environment. This solution addresses the limitations in the literature and effectively handles the complexity of the decision matrix. The findings reveal that cost leadership under a hybrid competitive strategy (HCS-CL) emerged as the most sensitive attribute, holding the highest weight, highlighting the importance of cost efficiency and competitive pricing. IOC11 ranked highest, followed by IOC3 and IOC10, providing benchmarks for other companies. Lesser-ordered companies, such as IOC4, preserve employ these comprehensions to recognize intentional gaps and embrace best attempts from extraordinary-ordered participants. Sensitivity-analysis, Spearman's-correlation, and comparative-analysis proven the robustness of the proposed approach. The study highlights the require for oil and gas administrators to spotlight cost leadership advantages, raise-efficiency, lower-production costs, and influence economies of ratio to maintain a reasonable edge and enhance-market-positioning.}, archive = {J_EAAI}, author = {Hassan A. Alsattar and Sarah Qahtan and Nahia Mourad and A.A. Zaidan and Muhammet Deveci and Dragan Pamucar and Jurgita Antucheviciene and Weiping Ding}, doi = {10.1016/j.engappai.2025.110623}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110623}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A decision modeling approach for the development of sustainable transportation oil companies}, volume = {150}, year = {2025}, }
• (2025). Improving model calibration in bone marrow cell classification through mixup and center loss fusion. EAAI, 150, 110620. (www)

  @article{ , abstract = {The analysis of bone marrow cell morphology is essential for the accurate diagnosis of hematological disorders. Traditional manual classification methods are time-consuming and labor-intensive. Although current automatic deep learning techniques mitigate these issues, they may still present significant risks in critical medical diagnostics due to overconfidence in predictions. To tackle these challenges, this paper proposes a novel calibration method called MixCL (Mix-Center Loss). MixCL combines the simple and effective data augmentation method Mixup with deep metric learning Center Loss, achieved through the design of a new loss function. By utilizing Mixup to generate mixing centers that enrich the feature sampling in the feature space, and leveraging the clustering effect of Center Loss to enhance the grouping of similar samples, MixCL combines the strengths of both methods. The effectiveness of MixCL is validated using three real bone marrow cell image datasets, demonstrating significant reductions in Expected Calibration Error (ECE) and Overconfidence Error (OE) for in-distribution samples. For example, in Shifted Windows Transformer model, ECE and OE metrics decreased across all datasets, with reductions averaging 1.72% in ECE and 2.10% in OE. The confidence Kernel Density Estimation (KDE) plot reveals that models using MixCL more effectively manage uncertainty in out-of-distribution samples, ensuring better differentiation between in-distribution and out-of-distribution samples. Thus, the proposed method effectively improves the calibration performance of the model while exhibiting better generalization performance, significantly improved when compared with current advanced bone marrow cell classification methods. Moreover, it has potential applications in various image classification fields, providing reliable confidence estimates.}, archive = {J_EAAI}, author = {Shuming Cheng and Qinghang Lu and Qianhang Guo and Yunqi Lin and Mingxin Li and Xingyu Zhao and Liang Guo and Jiaming Li and Jie Li and Qingmao Zhang and Qiongxiong Ma}, doi = {10.1016/j.engappai.2025.110620}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110620}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving model calibration in bone marrow cell classification through mixup and center loss fusion}, volume = {150}, year = {2025}, }
• (2025). Precise step counting algorithm for pedestrians using ultra-low-cost foot-mounted accelerometer. EAAI, 150, 110619. (www)

  @article{ , abstract = {Zero-velocity update (ZUPT) is one of the most widely used step counting methods in pedestrian inertial navigation. Existing accelerometer-based methods encounter issues with misjudgments of zero velocity and step counting errors in various activities. To tackle these challenges, we present a precise step counting algorithm leveraging ultra-low-cost foot-mounted accelerometer for pedestrians with extremely low complexity. This algorithm mainly contains two key points: adaptive acceleration threshold selection and state vector update, implemented by twice zero-velocity detections (ZVD, generating state vectors to depict pedestrian's status) and length comparisons. Accelerations undergo gravity correction, magnitude calculation, data smoothing, and parameters initialization. Subsequently, the zero-acceleration threshold is adaptively selected through ZVD, and the length of state intervals (LSI) is compared with the first length threshold. Then, the state vector is updated by ZVD, and the LSI is compared with the second length threshold. Finally, the number of state intervals is interpreted as step counts. Step counting experiments were conducted utilizing three diverse datasets, which comprised a self-constructed ultra-low-cost accelerometer-based dataset and two public datasets. These datasets covered various activities, such as normal walking, fast walking, running, multiple turns in a corridor, stationary stepping, and upstairs/downstairs. The proposed algorithm could achieve an accuracy of 100 % under the above situations. Compared to long short term memory (LSTM) and other algorithms, it exhibited an accuracy improvement of at least 13.05 %, with a processing time only 0.07 % of that required by LSTM. The proposed algorithm enables accurate step counting across a range of activities performed by different pedestrians with high precision, low complexity, and robust applicability.}, archive = {J_EAAI}, author = {Jingxue Bi and Jianhui Wang and Baoguo Yu and Guobiao Yao and Yunjia Wang and Hongji Cao and Lu Huang and Huaqiao Xing}, doi = {10.1016/j.engappai.2025.110619}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110619}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Precise step counting algorithm for pedestrians using ultra-low-cost foot-mounted accelerometer}, volume = {150}, year = {2025}, }
• (2025). Two-stream structure-oriented neighbor enhancement network for dental model segmentation. EAAI, 150, 110618. (www)

  @article{ , abstract = {The primary objective of digital orthodontic treatment is to achieve accurate tooth segmentation of the three-dimensional (3D) dental mesh model obtained from oral scanning equipment. However, current advanced deep learning-based methods often consolidate all features into a single vector during the feature learning process, thus overlooking the distinct information among features and subsequently weakening their complementary roles. To address this issue, we propose a two-stream structure-oriented neighbor enhancement network (TSNEN) to improve the complementary effect among different features. Specifically, TSNEN develops an input-specific two-stream structure and feature enhancement modules to emphasize the geometric disparities among various meshes and achieve the complementary enhancement of different features, respectively. Furthermore, the self-attention module is modified to fully integrate the local features derived from the branches of the two streams, which effectively balances the data differences among different features to mitigate feature confusion, and ultimately achieves accurate segmentation of the dental model. The real dental model dataset is analyzed to verify the effectiveness and capability of the proposed method which reached an overall accuracy (OA) at 96.83 % and mean over union (mIoU) at 92.06 %. Finally, the comparative analysis is implemented and the results further show that the proposed method has better performance both in prediction accuracy and robustness.}, archive = {J_EAAI}, author = {Zhihua Liu and Hao Tang and Jiutao Xue and Yuhe Liao}, doi = {10.1016/j.engappai.2025.110618}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110618}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Two-stream structure-oriented neighbor enhancement network for dental model segmentation}, volume = {150}, year = {2025}, }
• (2025). Real-time prediction of axial force in concrete-filled steel tubular columns under fire conditions using modular artificial intelligence techniques. EAAI, 150, 110617. (www)

  @article{ , abstract = {The axial force plays a critical role in assessing the functional integrity of columns within a building in fire. However, it cannot be measured directly and is influenced by factors such as temperature, load ratio, and axial restraint. This study proposes a real-time methodology to predict the axial force of restrained concrete-filled steel tubular (CFST) columns exposed to real fires, utilizing modular artificial intelligence. A module is developed that combines a convolutional neural network (CNN) and long short-term memory (LSTM) networks to predict the temperature field of CFST columns caused by fire in real time. This module estimates the current temperature field using past data and the current surface temperature, which is continuously monitored with inherent noise. It effectively mitigates noise interference, achieving an R 2 of 0.97 on the test dataset, which ensures accurate estimations. Additionally, a separate LSTM module with a skip connection is employed to predict the axial force ratio, integrating temperature predictions and real-time measurements of axial deformation. Finally, the accuracy of this modular model demonstrates better performance in predicting real-time axial force compared to the conventional integrated deep learning model, achieving an R 2 of 0.99. The proposed approach enables accurate prediction of axial force in restrained CFST columns across various fire scenarios and structural conditions, aiming at increasing the scientificity of fire rescue decisions.}, archive = {J_EAAI}, author = {Hong-Hui Qi and Guo-Qiang Li and Shaojun Zhu}, doi = {10.1016/j.engappai.2025.110617}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110617}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time prediction of axial force in concrete-filled steel tubular columns under fire conditions using modular artificial intelligence techniques}, volume = {150}, year = {2025}, }
• (2025). A real-time prediction model for instantaneous dam-break flood evolution of concrete gravity dams based on attention mechanism and spatiotemporal multiple features. EAAI, 150, 110616. (www)

  @article{ , abstract = {Simulating the flood evolution following the sudden breach of concrete gravity dams is crucial for enabling prompt emergency flood control decisions. The real-time performance and reliability of these flood propagation simulations are essential for improving the accuracy and speed of emergency responses. This study introduces a deep learning model that integrates an attention mechanism to predict flood evolution parameters in real time. Initially, parameters such as water depth and flow rate were measured under 32 distinct dam-break scenarios using a hydrodynamic model. By combining terrain data with time-series flood discharge data, we compiled a dataset containing 1984 entries, enhanced through reduced-order methods. A novel deep learning model, the Flood-Swin-Transformer, was then developed to predict the spatiotemporal evolution of dam-break floods. This model was benchmarked against 11 baseline models and four state-of-the-art deep learning models. The results indicate: (1) Baseline models accurately predict water depth but are less effective at predicting flow rate parameters; (2) Deep learning models outperform baseline models in both accuracy and classification capabilities for water depth and flow rate parameters, showing robust performance; (3) Extensive analyses, including error, classification accuracy, effectiveness, robustness, and flood parameter error mapping, demonstrate the superior performance of the proposed model; (4) The proposed model predicts flood evolution up to 43.75 times faster than traditional hydrodynamic models, facilitating real-time prediction capabilities.}, archive = {J_EAAI}, author = {Chao Wang and Yaofei Zhang and Sherong Zhang and Xiaohua Wang and Xingbo Zhou and Yishu Lai}, doi = {10.1016/j.engappai.2025.110616}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110616}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A real-time prediction model for instantaneous dam-break flood evolution of concrete gravity dams based on attention mechanism and spatiotemporal multiple features}, volume = {150}, year = {2025}, }
• (2025). Dialogue response coherency evaluation with feature sensitive negative sample using multi list-wise ranking loss. EAAI, 150, 110609. (www)

  @article{ , abstract = {Automatic evaluation of dialogue coherency is crucial for developing high-quality dialogue systems. However, traditional evaluation metrics such as Bilingual Evaluation Understudy (BLEU) and Recall-Oriented Understudy for Gisting Evaluation (ROUGE) have limitations when it comes to assessing diverse and creative responses because they heavily rely on reference responses. For learnable metrics which utilize contrastive learning, challenges are encountered due to the use of randomly selected negative samples that do not reflect conversational features (i.e. topic, emotion, intention) and the lack of granularity in assessing response appropriateness. To address these limitations, we propose the Feature sensitive Multi-Listwise Ranking (FMListR) response coherency evaluation model. This model aims to evaluate dialogue coherency in degrees while considering conversational sensitive features. This approach involves sampling feature-sensitive responses that share conversational features with ground truth responses and utilizing them as hard negative samples. The model is trained using Multi-Listwise Ranking (MListR) loss, which is designed to learn the ranking between negative samples and identify response features. The experimental results demonstrate that Feature sensitive Multi-Listwise Ranking exhibits stronger correlations with human judgment compared to other response coherency evaluation metrics. By considering conversational features and training the model using a specialized loss function, FMListR provides a more robust and accurate evaluation of dialogue coherency.}, archive = {J_EAAI}, author = {YeongJun Hwang and Dongjun Kang and JinYeong Bak}, doi = {10.1016/j.engappai.2025.110609}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110609}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dialogue response coherency evaluation with feature sensitive negative sample using multi list-wise ranking loss}, volume = {150}, year = {2025}, }
• (2025). Comprehensive assessment of failure mode and shear capacity of reinforced concrete circular columns based on data-driven machine learning methods. EAAI, 150, 110603. (www)

  @article{ , abstract = {Reinforced concrete (RC) columns are critical elements in frame structures. A common challenge in structural engineering is estimating the seismic capacity of circular RC columns, as existing methods typically require transforming circular columns into equivalent square sections due to the absence of direct formulas. To address this, this study introduces data-driven machine learning (ML) methods to directly assess both failure modes and seismic shear capacity of circular RC columns. With the help of automated machine learning (AutoML), seven ML algorithms were selected and fine-tuned, resulting in 40 distinct models that were analyzed and compared in detail. The results indicate that the Multilayer Perceptron (MLP) model outperforms others in predicting seismic failure modes, achieving a high accuracy of 88 %. For seismic shear capacity predictions, the Weighted Ensemble (WE) model achieved the best performance with Root Mean Squared Error of 53.49, Mean Absolute Error of 39.23, Coefficient of Determination of 0.88 and Mean Squared Error of 2861.18 among all the ML models. Furthermore, the results (the predicted maximum lateral force/the experimental results) of the WE model, with a mean value of 0.98, a standard deviation of 0.11, and a coefficient of variation (mean/standard deviation) of 12.8 %, surpass those of traditional theoretical and empirical models. Besides, the ML models offer fast, accurate seismic performance evaluations for circular RC columns, eliminating the need for complex and time-consuming calculations. Furthermore, SHapley Additive exPlanations (SHAP) analysis provided visual insights into parameter contributions, enhancing model transparency and trust for engineering applications.}, archive = {J_EAAI}, author = {Yue Wen and Shiqiao Zhou and Gaochuang Cai and Zhili He and Amir Si Larbi}, doi = {10.1016/j.engappai.2025.110603}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110603}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comprehensive assessment of failure mode and shear capacity of reinforced concrete circular columns based on data-driven machine learning methods}, volume = {150}, year = {2025}, }
• (2025). Leveraging large language models to examine the interaction between investor sentiment and stock performance. EAAI, 150, 110602. (www)

  @article{ , abstract = {Understanding the relationship between investor sentiment and stock performance is crucial in dynamic financial markets. Existing researches often focus on financial news and stock prices, while studies on investor sentiment typically rely on traditional machine learning models that require extensive data labeling. Additionally, most researches focus on single stock indices, overlooking the impact of brand popularity. To address these gaps, this study proposes a novel framework to analyze the interaction between investor sentiment and stock performance, using Chinese Baijiu industry stocks as a case example. It further explores how brand popularity influences this relationship, offering insights for informed investment decisions through artificial intelligence technology. In this study, we leverage Generative Pre-trained Transformer 4 (GPT-4), a state-of-the-art black-box large language model, to process vast volumes of unstructured text data from stock forums. By employing in-context learning with human-labeled examples, GPT-4 generates weak labels that are subsequently used to fine-tune Large Language Model Meta AI (LLaMA), a smaller and more efficient open-source LLM from Meta AI, thereby enabling sentiment-driven decision-making in real-world scenarios. To construct a comprehensive sentiment indicator, we integrate both direct and indirect factors influencing sentiment and use principal component analysis to combine them effectively. To examine interaction between sentiment and stock yield, we apply the Granger causality test and vector autoregression models across stocks with different brand popularity levels. The results show that our framework achieves state-of-the-art performance investor sentiment analysis. Moreover, with brand popularity significantly amplifying the interaction between investor sentiment and stock yield, it leads to bidirectional Granger causality in highly popular brands.}, archive = {J_EAAI}, author = {Yong Zhuang and Feilong Wang and Dickson K.W. Chiu and Kevin K.W. Ho}, doi = {10.1016/j.engappai.2025.110602}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110602}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Leveraging large language models to examine the interaction between investor sentiment and stock performance}, volume = {150}, year = {2025}, }
• (2025). Lane change trajectory planning based on improved model predictive control with artificial potential field for autonomous vehicles in medium-high speed scenarios. EAAI, 150, 110601. (www)

  @article{ , abstract = {This paper presents the design of a lane change control system for autonomous vehicles (AVs) in medium-high speed scenarios. A lane change trajectory planning method based on the combination of artificial potential field (APF) and model predictive control (MPC), referred to as APF-MPC, is proposed. To ensure a smooth transition at the beginning and end of the lane change, a reference trajectory based on a sinusoidal curve is designed. The obstacle potential field, constructed using a two-dimensional joint probability density function, is combined with the road potential field to enhance lane change safety. The potential field function is integrated into the MPC optimal control problem, which is constructed based on the point-mass model. This problem is then solved to obtain the planned trajectory within the predicted time domain. The lateral and longitudinal tracking controllers, designed using linear MPC, are employed to track the planned trajectory in real-time, thereby demonstrating the real-time performance of this method. The feasibility of the APF-MPC method is verified through a co-simulation platform. The simulation results indicate that the trajectories planned by the APF-MPC method meet the requirements of safety, occupant comfort, and lane change efficiency. In different scenarios, the front wheel angle of the AV during the lane change ranges from -2 to 2 degrees. Compared with non-adaptive APF and traditional MPC methods, the maximum lateral acceleration, maximum lateral velocity, and maximum front wheel steering angle of the AV during lane changing are reduced by more than 40%.}, archive = {J_EAAI}, author = {Zhaojun Zhang and Jiale Qin and Simeng Tan and Hongjie Luo}, doi = {10.1016/j.engappai.2025.110601}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110601}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lane change trajectory planning based on improved model predictive control with artificial potential field for autonomous vehicles in medium-high speed scenarios}, volume = {150}, year = {2025}, }
• (2025). Input parameterized physics informed neural networks for de noising, super-resolution, and imaging artifact mitigation in time resolved three dimensional phase-contrast magnetic resonance imaging. EAAI, 150, 110600. (www)

  @article{ , abstract = {Motivation: Hemodynamic analysis is crucial for diagnosing and predicting cardiovascular diseases. However, methods relying on fluid flow simulations or blood flow imaging are complex, time-consuming, and require specialized expertise, limiting their clinical use. Goal: This research aims to automate the enhancement of blood flow images, providing clinicians with a fast, accurate tool for hemodynamic analysis without requiring advanced expertise. Objectives: A software tool based on physics-constrained neural networks was developed to enable clinicians to easily select and process regions of interest (ROIs) in time-resolved three-dimensional phase contrast magnetic resonance imaging (4D-Flow MRI) blood flow images for quick, accurate analysis. Methods: The Input Parameterized Physics-Informed Neural Network (IP-PINN) was introduced to improve the spatio-temporal resolution of 4D-Flow MRI. IP-PINN mitigates noise, velocity aliasing, and phase errors. A convolutional neural network processes ROI data into latent vectors, which are then used to predict velocity, pressure, and spin density via a multi-layer perceptron. The method is trained with synthetic blood flow data using an innovative loss function that addresses noise and artifacts. Results: IP-PINN successfully enhanced image resolution, reducing noise and artifacts when tested on synthetic 4D-Flow MRI data derived from blood flow simulations of intracranial aneurysms. For data with 20 decibels (dB) signal-to-noise ratio, results closely matched the ground truth with less than 5.5% relative error. Processing took under two minutes. The method also has the potential to reduce data acquisition time by 25%. Conclusions: IP-PINN could significantly enhance the clinical use of 4D-Flow MRI for personalized hemodynamic analysis in cardiovascular diseases.}, archive = {J_EAAI}, author = {Amin Pashaei Kalajahi and Hunor Csala and Zayeed Bin Mamun and Sangeeta Yadav and Omid Amili and Amirhossein Arzani and Roshan M. D’Souza}, doi = {10.1016/j.engappai.2025.110600}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110600}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Input parameterized physics informed neural networks for de noising, super-resolution, and imaging artifact mitigation in time resolved three dimensional phase-contrast magnetic resonance imaging}, volume = {150}, year = {2025}, }
• (2025). Solving quay wall allocation problems based on deep reinforcement learning. EAAI, 150, 110598. (www)

  @article{ , abstract = {Quay walls and graving docks are critical production resources in shipyards. Traditionally, quay walls have not been a bottleneck resource for constructing conventional vessels, such as oil carriers and container ships. However, the growing demand for high value-added vessels requiring more complex post-stage outfitting operations has increased workloads at quay walls. Accordingly, the importance of efficient quay wall allocation has grown significantly to improve overall production efficiency and ensure timely vessel delivery. In this study, the quay wall allocation problem is modeled as a flexible job shop scheduling problem, incorporating machine preferences and preemption conditions. Notably, the uncertainty in vessel launching dates caused by delays in the erection process at graving docks is considered in the scheduling problems. To address the dynamic quay wall allocation problems, this study develops a dynamic quay wall allocation algorithm based on deep reinforcement learning, which adaptively allocates vessels to quay walls based on the working status of quay walls and the progress of outfitting operations. For this purpose, a novel Markov decision process is proposed, where a compound state representation composed of heterogeneous graphs and auxiliary matrices is devised to capture the complex relationships between quay walls and outfitting operations. In addition, an extended scheduling action space incorporating operation interruptions is defined, which can effectively utilize preemption conditions to enhance the scheduling performance. The performance of the proposed algorithm is evaluated through extensive numerical experiments based on test instances generated from real-world shipyard data under various environmental conditions. Experimental results demonstrate that the proposed algorithm consistently outperforms traditional rule-based heuristics and exhibits superior scalability compared to genetic programming, making it a promising solution for large-scale quay wall allocation problems.}, archive = {J_EAAI}, author = {Young-in Cho and Seung-heon Oh and Jae-ho Choi and Jong Hun Woo}, doi = {10.1016/j.engappai.2025.110598}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110598}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Solving quay wall allocation problems based on deep reinforcement learning}, volume = {150}, year = {2025}, }
• (2025). Optimizing machine learning algorithms for fault classification in rolling bearings: A bayesian optimization approach. EAAI, 150, 110597. (www)

  @article{ , abstract = {Modern power machinery is inherently complex and operates under dynamic operating conditions, so they demand advanced solutions based on deep learning to diagnose bearing faults inside rotating equipment that cause unplanned downtime and safety issues, leading to operational challenges. However, most deep learning approaches aim to improve performance by incorporating hybrid neural networks that rely on multiple convolutional and temporal units, often overlooking optimizing the large number of hyperparameters that define the structure and performance of hybrid models along with the associated computational constraints. To address this gap, this study presents an innovative approach for the detection and classification of bearing faults by integrating an optimized sparse deep autoencoder (DAE) with a Bidirectional Long Short-Term Memory model (Bi-LSTM). The optimal network structure and hyperparameters are determined through Bayesian optimization (BO) with parallel settings, which automatically searches for network configurations that improve the feature extraction ability of the DAE and the generalization ability of the Bi-LSTM for more efficient fault classification in rolling bearings. Parallel optimization accelerates network structure and hyperparameter tuning by evaluating multiple configurations at once. It leverages the full potential of available multi-core Central Processing Units (CPUs)/Graphics Processing Units (GPUs) in conjunction with a lightweight BO surrogate model. This autonomous and user-friendly framework generates inputs from principal component analysis for linear and BO-DAE for non-linear feature extraction and selection, which are then used to train a BO-enhanced Bi-LSTM. This three-stage optimized method effectively captures spatial and temporal dependencies in vibrational signals, achieving superior efficiency, accuracy, and reliability compared to shallow and deep learning models. Evaluation metrics, including macro precision (99.50 %), recall (99.60 %), F1-Score (99.57 %), and Cohen's Kappa metric (Cκ = 99.53 %), demonstrate the efficacy of our approach for bearing fault classification in industrial applications.}, archive = {J_EAAI}, author = {Muhammad Zain Yousaf and Josep M. Guerrero and Muhammad Tariq Sadiq}, doi = {10.1016/j.engappai.2025.110597}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110597}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing machine learning algorithms for fault classification in rolling bearings: A bayesian optimization approach}, volume = {150}, year = {2025}, }
• (2025). Dynamic scheduling in flexible and hybrid disassembly systems with manual and automated workstations using reward-shaping enhanced reinforcement learning. EAAI, 150, 110588. (www)

  @article{ , abstract = {As e-waste grows at an alarming rate, efficient disassembly systems have become crucial for sustainable production practices. Existing disassembly systems rely heavily on fixed automation and limited manual intervention, making it challenging to adapt to dynamic issues such as workstation failures and system bottlenecks, leading to inefficiencies and suboptimal resource allocation. To address these issues, a hybrid disassembly system is developed that integrates manual and automated workstations, allowing for the flexible variation of manual resources as needed to optimize the disassembly process, with a focus on reducing time and maximizing profit. Through the proposal of a Proximal Policy Optimization (PPO) algorithm enhanced with Reward-Shaping, the research effectively tackles key challenges of uncertainty and dynamic conditions in disassembly systems, including workstation failures and system bottlenecks. These issues are explored through a refrigerator disassembly simulation model. The results demonstrate that the PPO algorithm significantly outperforms traditional rule-based methods and two other reinforcement learning techniques in managing complex dynamic scheduling and resource allocation tasks, offering greater efficiency and flexibility. These findings contribute to the advancement of automated disassembly processes and their integration into modern industrial systems.}, archive = {J_EAAI}, author = {Jinlong Wang and Qihuiyang Liang and Min Li and Zelin Qu and Yuanyuan Zhang}, doi = {10.1016/j.engappai.2025.110588}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110588}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic scheduling in flexible and hybrid disassembly systems with manual and automated workstations using reward-shaping enhanced reinforcement learning}, volume = {150}, year = {2025}, }
• (2025). Soft actor-critic enhanced nonsingular terminal synergetic control for serial manipulators with quantized input and state. EAAI, 150, 110587. (www)

  @article{ , abstract = {Synergetic control is vulnerable to model uncertainty, which affects control performances and stability. In addition, quantized sensor measurements and control signals compound these challenges. Equally important, a unique problem in serial manipulator robots is the varying control response across joints, which affects the end effector’s pose when moving to the desired position. The mentioned challenges motivate this study to propose soft actor-critic reinforcement learning as a novel adaptive approach for nonsingular terminal synergetic control on a 4-degree-of-freedom serial manipulator robot. The control law is designed based on the nonsingular terminal synergetic control evolution constraint, manifold, and macrovariable. Subsequently, the soft actor-critic reinforcement learning dynamically adjusts the evolution constraint parameters, adapting to the changing state of the environment. The Lyapunov stability theorem proves the control law stability. This study also introduces a novel reinforcement learning reward function that encourages state convergence using the Cauchy probability density function and macrovariable. The agent training environment accounts for state and input quantization, enabling a seamless sim-to-real transition. The simulations and physical experiments validate the effectiveness of the proposed controller in improving transient response, tracking response, and uniform performance across joints. Additionally, online training validates the safe exploration property of the proposed adaptive control.}, archive = {J_EAAI}, author = {Muhammad Auzan and Yong-Lin Kuo}, doi = {10.1016/j.engappai.2025.110587}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110587}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Soft actor-critic enhanced nonsingular terminal synergetic control for serial manipulators with quantized input and state}, volume = {150}, year = {2025}, }
• (2025). Artificial neural networks for finger vein recognition: A survey. EAAI, 150, 110586. (www)

  @article{ , abstract = {Finger vein recognition is an emerging biometric recognition technology. Different from the other biometric features on the body surface, the venous vascular tissue of the fingers is buried deep inside the skin. Due to this advantage, finger vein recognition is highly stable and private. Finger veins are virtually impossible to steal and difficult to interfere with by external conditions. Unlike the finger vein recognition methods based on traditional machine learning, the artificial neural network technique, especially deep learning, does not rely on feature engineering and has superior performance. To summarize the development of finger vein recognition based on artificial neural networks,this paper collects 174 related papers. First, we introduce the background of finger vein recognition and the motivation for this survey. Then, the development history of artificial neural networks and the representative networks on finger vein recognition tasks are introduced. The public datasets widely used in finger vein recognition are then described. After that, we summarize the related finger vein recognition tasks based on classical neural networks and deep neural networks, respectively. Finally, the challenges and potential development directions in finger vein recognition are discussed. This paper provides a comprehensive and novel summary of the application of artificial neural networks in the finger vein recognition field.}, archive = {J_EAAI}, author = {Yimin Yin and Renye Zhang and Pengfei Liu and Wanxia Deng and Dayu Hu and Siliang He and Chen Li and Jinghua Zhang}, doi = {10.1016/j.engappai.2025.110586}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110586}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial neural networks for finger vein recognition: A survey}, volume = {150}, year = {2025}, }
• (2025). A dual-module cooperative control method for on-ramp area in heterogeneous traffic flow using reinforcement learning. EAAI, 150, 110584. (www)

  @article{ , abstract = {In the on-ramp area, vehicle conflicts significantly reduce traffic efficiency and increase collision risks. This study introduces a novel dual-module cooperative control approach designed for on-ramps that accommodate heterogeneous traffic flows, including connected and automated vehicles (CAVs) and human driving vehicles (HDVs). By utilizing reinforcement learning techniques, the approach aims to enhance both traffic efficiency and safety. The approach comprises two key modules: the merging control module and the lane-changing control module. The merging control module facilitates cooperation between mainline and ramp vehicles, while the lane-changing control module assists mainline CAVs in making informed lane-change decisions. Agents within these modules are trained using the proximal policy optimization algorithm, known for its strong convergence properties. After 100 to 200 training episodes, the agents achieve stable peak average rewards. Simulation results demonstrate significant improvements in traffic efficiency and safety with the dual-module control method in on-ramp areas, especially in scenarios involving CAV-HDV heterogeneous traffic flows. With a CAV penetration rate of just 0.2, average vehicle delay is reduced by 26 %. Furthermore, from a safety perspective, when the CAV penetration rate reaches or exceeds 0.3, the time-exposed time-to-collision decreases by approximately 45 %. Transferability analysis indicates that integrating reinforcement learning agents into the control strategy produces positive results across varying maximum speeds and flow rates. In heterogeneous traffic environments, it is advisable to train agents at high CAV penetration rates. Comparative studies further show that the proposed method significantly enhances traffic efficiency and safety, maintaining robust performance even at lower CAV penetration rates.}, archive = {J_EAAI}, author = {Wenzhang Yang and Changyin Dong and Ziqian Zhang and Xu Chen and Hao Wang}, doi = {10.1016/j.engappai.2025.110584}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110584}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual-module cooperative control method for on-ramp area in heterogeneous traffic flow using reinforcement learning}, volume = {150}, year = {2025}, }
• (2025). Class imbalance-aware domain specific transfer learning approach for medical image classification: Application on COVID-19 detection. EAAI, 150, 110583. (www)

  @article{ , abstract = {Convolutional Neural Networks (CNNs) show promise in medical image classification; however, their effectiveness is often constrained by the availability of large, annotated datasets, which are not always accessible. Additionally, the lack of medically relevant transfer models limits the potential of Transfer Learning (TL) in addressing this challenge. Existing TL methods typically employ conventional approaches that result in suboptimal performance and exhibit issues related to network bias, especially in imbalanced datasets. To overcome these limitations, we introduce a novel class imbalance-aware, domain-specific transfer learning framework (CIDSTL-Net) designed specifically for medical imaging tasks. CIDSTL-Net adopts a two-stage training approach: initially developing domain-specific models followed by fine-tuning on targeted medical datasets. This method incorporates an innovative class weighting strategy in its loss calculation to address dataset bias and enhances the transfer head network with a novel combination of fully connected, batch normalization, and dropout layers. Additionally, CIDSTL-Net employs cyclically scheduled learning rates to optimize parameter exploration and exploitation during training. We have rigorously evaluated CIDSTL-Net on four publicly available COVID-related datasets, covering chest X-ray and Computed Tomography (CT) images for the classification of COVID, Non-COVID, Normal, Pneumonia, and Lung Opacity conditions. The results demonstrate state-of-the-art performance with 5-fold cross-validation mean accuracies of 96.87 %, 96.50 %, 99.70 %, and 99.55 % for the respective datasets, marking significant improvements over existing methods. Among various CNN architectures tested, DenseNet-121 proved to be the most effective, offering superior accuracy with fewer parameters. Given the pressing global challenge posed by the COVID-19 pandemic, CIDSTL-Net holds significant potential to aid medical practitioners in the rapid and accurate classification of COVID-19 cases.}, archive = {J_EAAI}, author = {Marut Jindal and Birmohan Singh}, doi = {10.1016/j.engappai.2025.110583}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110583}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Class imbalance-aware domain specific transfer learning approach for medical image classification: Application on COVID-19 detection}, volume = {150}, year = {2025}, }
• (2025). Intelligent evaluation of pavement friction at high speeds with artificial intelligence powered three-dimensional laser imaging technology. EAAI, 150, 110580. (www)

  @article{ , abstract = {Accurate non-contact pavement friction evaluation at high speeds faces many challenges when using low-resolution (LR) three-dimensional (3D) texture images or other deficient texture data. With artificial intelligence (AI) powered 0.1-mm (0.1-mm) 3D laser imaging technology, this paper proposed a two-step deep learning (DL) network, named Friction-8KNet, for accurate and intelligent non-contact pavement friction evaluation at high speeds. Particularly, a 3D laser imagining device was employed to collect LR texture images at a speed of 30 mph, which were processed via a DL-based super-resolution (SR) algorithm to obtain 0.1 mm high-resolution (HR) images with a size of 8192 × 4096 (8K) pixels. The Friction-8KNet comprises a network backbone for texture feature extraction in Step 1 and a triple attention network for friction evaluation in Step 2. The network backbone is developed to precisely extract features of an HR image without requiring large graphics processing unit (GPU) memory. The triple attention net is designed with three function-specific attention modules to utterly mine the extracted features for accurate friction prediction. Experimental results show that Friction-8KNet can achieve 99.19 % prediction accuracy and transcends models using other texture data, including small HR 3D images, LR 3D images, and two-dimensional (2D) texture profiles. This research promotes an accurate and efficient measurement of pavement friction for production level in a non-contact manner in the future.}, archive = {J_EAAI}, author = {Guolong Wang and Kelvin C.P. Wang and Guangwei Yang}, doi = {10.1016/j.engappai.2025.110580}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110580}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent evaluation of pavement friction at high speeds with artificial intelligence powered three-dimensional laser imaging technology}, volume = {150}, year = {2025}, }
• (2025). Corrigendum to “Understanding the disparities in mathematics performance: An interpretability-based examination” [Eng. appl. artif. intell. 133 part b (2024) 108109]. EAAI, 150, 110579. (www)

  @article{ , archive = {J_EAAI}, author = {Ismael Gómez-Talal and Luis Bote-Curiel and José Luis Rojo-Álvarez}, doi = {10.1016/j.engappai.2025.110579}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110579}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Corrigendum to “Understanding the disparities in mathematics performance: An interpretability-based examination” [Eng. appl. artif. intell. 133 part b (2024) 108109]}, volume = {150}, year = {2025}, }
• (2025). Standalone and hybrid machine learning approaches to predict sediment load in an alluvial channel. EAAI, 150, 110578. (www)

  @article{ , abstract = {A significant amount of sediment transported in an alluvial river can alter the morphology and shape of the river. Accurate prediction of sediment load is essential in studying the change in geomorphology and dynamics of rivers and also to evaluate its impact on aquatic ecosystems, infrastructure, and human activities dependent on water resources. The present study demonstrates framework for predicting sediment load in alluvial channels using both standalone and hybrid machine learning (ML) models. Multiple datasets collected from various river surveys and flume studies were used to evaluate the significance of key variables such as friction slope ( Sf ), channel discharge ( Q ), and bed shear stress ( τ b ) affecting the sediment transport employing ML models (Bagging (BA), Random Committee (RC)) and the standalone ML models (Multi-Layer Perceptron Regression (MLPR) and Reduced Error Pruning Tree (REPT). The hybrid Bagging-REPT (BA-REPT) model outperformed other models with a Nash-Sutcliffe Efficiency (NSE) of 0.915, followed by RC-REPT (NSE = 0.906). Among the various variables, friction slope ( S f ) was identified as the most influential variable affecting sediment transport behavior. It was also observed that Hybrid models can predict sediment transport behavior more accurately as compared to standalone models and empirical equations. The findings of the study thus demonstrate the importance of hybrid learning in addressing the nonlinear complexity of sediment transport processes.}, archive = {J_EAAI}, author = {Sanjit Kumar and Vishal Deshpande and Mayank Agarwal}, doi = {10.1016/j.engappai.2025.110578}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110578}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Standalone and hybrid machine learning approaches to predict sediment load in an alluvial channel}, volume = {150}, year = {2025}, }
• (2025). A lightweight vision transformer with embedded hybrid attention for quick response code defect classification. EAAI, 150, 110575. (www)

  @article{ , abstract = {Quick Response (QR) code label printing quality is crucial to product control. Due to the limited number of defect samples, unclear features, and the need to detect a large number of labels in real time, automated visual inspection faces challenges. For efficient and accurate automated visual defect recognition of printed QR code production, we propose a lightweight Vision Transformer network, Vision Transformer with Embedded Hybrid Attention (ViT-EHA). First, the Mixed Depthwise Convolution Block (MDConvBlock) is introduced to capture QR code defect details and feature information. This method additionally reduces the number of model parameters and computational costs. Furthermore, the LeAttention-Local Convolution-Multilayer Perceptron (LeALCM) module is proposed to enhance the ability to capture global information of the model and improve the effect of minor defect recognition. Ultimately, a hybrid attention (HA) module has been integrated to enhance the processing of low-level image features and to strengthen the interplay between shallow and deep features. To verify the validity and generalization of the model, the experimental results show that the proposed ViT-EHA method achieved an accuracy of 99.00% and a parameter count of 4.198 million (M) on the self-constructed dataset Code-10 (QR Code Dataset with 10 Classes), and the accuracy reached 98.33% and 97.73% on the public datasets NEU-CLS (Northeastern University Classification Dataset) and NEU-CLS-64 (Northeastern University Classification Dataset with 64 × 64 images), respectively.}, archive = {J_EAAI}, author = {Dianlu Hu and Lun Zhao and Yu Ren and Sen Wang and Xuanlin Ye and Haohan Zhang and Changqing Peng}, doi = {10.1016/j.engappai.2025.110575}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110575}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight vision transformer with embedded hybrid attention for quick response code defect classification}, volume = {150}, year = {2025}, }
• (2025). Evaluating pre-trained convolutional neural networks and foundation models as feature extractors for content-based medical image retrieval. EAAI, 150, 110571. (www)

  @article{ , abstract = {Medical image retrieval refers to the task of finding similar images for given query images in a database, with applications such as diagnosis support. While traditional medical image retrieval relied on clinical metadata, content-based medical image retrieval (CBMIR) depends on image features, which can be extracted automatically or semi-automatically. Many approaches have been proposed for CBMIR, and among them, using pre-trained convolutional neural networks (CNNs) is a widely utilized approach. However, considering the recent advances in the development of foundation models for various computer vision tasks, their application for CBMIR can also be investigated. In this study, we used several pre-trained feature extractors from well-known pre-trained CNNs and pre-trained foundation models and investigated the CBMIR performance on eight types of two-dimensional (2D) and three-dimensional (3D) medical images. Furthermore, we investigated the effect of image size on the CBMIR performance. Our results show that, overall, for the 2D datasets, foundation models deliver superior performance by a large margin compared to CNNs, with the general-purpose self-supervised model for computational pathology (UNI) providing the best overall performance across all datasets and image sizes. For 3D datasets, CNNs and foundation models deliver more competitive performance, with contrastive learning from captions for histopathology model (CONCH) achieving the best overall performance. Moreover, our findings confirm that while using larger image sizes (especially for 2D datasets) yields slightly better performance, competitive CBMIR performance can still be achieved even with smaller image sizes. Our codes to reproduce the results are available at: https://github.com/masih4/MedImageRetrieval .}, archive = {J_EAAI}, author = {Amirreza Mahbod and Nematollah Saeidi and Sepideh Hatamikia and Ramona Woitek}, doi = {10.1016/j.engappai.2025.110571}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110571}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evaluating pre-trained convolutional neural networks and foundation models as feature extractors for content-based medical image retrieval}, volume = {150}, year = {2025}, }
• (2025). Thermal image-guided complementary masking with multiscale fusion for multi-spectral image semantic segmentation. EAAI, 150, 110569. (www)

  @article{ , abstract = {Semantic segmentation of visible and thermal image is a kind of significant method for harsh environments. Most existing works focus on designing a multi-modal feature fusion module. However, these works may result in over-dependence on a specific modality and a lack of consideration for local and global context-aware information. Motivated by these issues, (1) a thermal image-guided complementary masking strategy is proposed to encourage the network to focus on regions with abundant semantic information; (2) a multi-modal fusion module is developed to integrate both local and global information and ensure consistency for semantic segmentation; (3) a self-distillation loss between unmasked and masked input modalities is introduced to enhance the robustness and consistency of the network. Particularly, the proposed masking strategy can force the network to concentrate on the meaningful area in all modalities, and thus the network can enhance the ability to connect context information. Experimental results on three public datasets demonstrate the superiority of our model.}, archive = {J_EAAI}, author = {Zeyang Chen and Mingnan Hu and Bo Chen}, doi = {10.1016/j.engappai.2025.110569}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110569}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Thermal image-guided complementary masking with multiscale fusion for multi-spectral image semantic segmentation}, volume = {150}, year = {2025}, }
• (2025). An electric vehicle sales hybrid forecasting method based on improved sentiment analysis model and secondary decomposition. EAAI, 150, 110561. (www)

  @article{ , abstract = {The swift proliferation of electric vehicles has triggered profound shifts in consumer behavior, emphasizing the critical role of precise sales forecasts as the cornerstone for data-driven policy and production planning by both governments and electric vehicle manufacturers. However, extant forecasting models face challenges in accurately capturing consumer sentiment conveyed by online reviews and effectively extracting multiscale features of high-frequency sequences. Therefore, an electric vehicle sales hybrid forecasting method based on BERT-Bi-LSTM (Bidirectional Encoder Representations from Transformers-Bidirectional long short-term memory) sentiment analysis and secondary decomposition is proposed. First, the BERT-Bi-LSTM model is developed to perform sentiment analysis on online reviews. The model can better capture the relationship between each word and its surrounding words in text information. Second, a secondary decomposition model is constructed to decompose multisource data series, it can extract the seasonal components of the series and high-frequency complex data features, also solve potential issues of incomplete decomposition that may arise from a single decomposition. Finally, machine learning methods are utilized for hybrid forecasting. To verify the effectiveness of the proposed model, multiple sets of comparative experiments are conducted. The empirical results indicate the proposed model has higher prediction accuracy and robustness.}, archive = {J_EAAI}, author = {Jinpei Liu and Hui Pan and Rui Luo and Huayou Chen and Zhifu Tao and Zhijing Wu}, doi = {10.1016/j.engappai.2025.110561}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110561}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An electric vehicle sales hybrid forecasting method based on improved sentiment analysis model and secondary decomposition}, volume = {150}, year = {2025}, }
• (2025). Causal explanation of nitrogen oxide emission predictions for fluid catalytic cracking unit based on convergent cross mapping: Predict the future and explain how. EAAI, 150, 110560. (www)

  @article{ , abstract = {Industrial chemical processes are inherently intricate, characterized by prolonged operational sequences and correlations among features. Solely utilizing temporal information limits the prediction precision of deep learning methods. Moreover, causative features identified based on data may not align with the principles of chemical process. To solve these problems, this study proposes a spatial–temporal information based deep learning method, attention-based temporal graph convolutional network and convergent cross mapping (ATGCN-CCM). The devices are abstracted as nodes in a computational graph (CG) to represent the process, enabling the incorporation of spatial information into the predictive model. Attention mechanism is conducted within each node to dynamically weight the features. The CG is also used to select input features for causal analysis, ensuring that the identified causative features are not only consistent with the characteristics of the data, but also with the prior knowledge of the process. ATGCN-CCM is applied to datasets from industrial fluid catalytic cracking (FCC) units for nitrogen oxides (NOx) concentration prediction and causative feature identification. The prediction results demonstrate superior precision of ATGCN-CCM compared to some state-of-the-art spatial feature based, temporal feature based and hybrid methods. The identified features exhibit strong alignment with the principles of the chemical processes and the field experiences, thereby significantly enhancing model interpretability. The proposed ATGCN-CCM method illustrate its advanced capabilities in both precision and robustness, compared with attention-based methods and other causal analysis methods.}, archive = {J_EAAI}, author = {Han Jiang and Shucai Zhang and Jingru Liu and Xin Peng and Weimin Zhong}, doi = {10.1016/j.engappai.2025.110560}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110560}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Causal explanation of nitrogen oxide emission predictions for fluid catalytic cracking unit based on convergent cross mapping: Predict the future and explain how}, volume = {150}, year = {2025}, }
• (2025). Keypoint-guided feature enhancement and alignment for cross-resolution vehicle re-identification. EAAI, 150, 110557. (www)

  @article{ , abstract = {Resolution mismatch between low-resolution query images and high-resolution gallery images in vehicle re-identification is rarely studied but ubiquitous in real-world applications. An intuitive approach to solving cross-resolution vehicle re-identification is to utilize super-resolution algorithms to recover detailed information from low-resolution query images. However, vehicle super-resolution algorithms not only recover the detailed information of the vehicle but also enhance the background noise, which would degrade the re-identification performance. In addition, the view mismatch problem also significantly limits the performance of vehicle re-identification. To handle these problems, we propose a novel Keypoint Guiding Network, which simultaneously addresses the problems of resolution mismatch and view mismatch from the perspective of keypoints in an end-to-end learning framework, for cross-resolution vehicle re-identification. In particular, we first generate a set of vehicle keypoints via an effective Gaussian localization method, and then adaptively construct two keypoint-based guidances using attention models. We integrate these two guidances into vehicle super-resolution and view alignment to handle the problems of resolution mismatch and view mismatch respectively. Moreover, to alleviate the heterogeneity between super-resolution query images and high-resolution gallery ones, we design a dual-path teacher–student distillation scheme to narrow their feature distributions. Comprehensive experiments on two down-sampled benchmark datasets demonstrate the effectiveness of our Keypoint Guiding Network against the state-of-the-art methods.}, archive = {J_EAAI}, author = {Aihua Zheng and Longfei Zhang and Weijun Zhang and Zi Wang and Chenglong Li and Xiaofei Sheng}, doi = {10.1016/j.engappai.2025.110557}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110557}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Keypoint-guided feature enhancement and alignment for cross-resolution vehicle re-identification}, volume = {150}, year = {2025}, }
• (2025). Runner system geometry prediction using variational autoencoder deep learning model. EAAI, 150, 110555. (www)

  @article{ , abstract = {The novelty of this article is the choice of the architecture of neural networks for fluid channel topology optimization and its application in the design of multi-cavity injection molds. A generative deep learning model was developed to extract characteristics representing the inverse field of the permeability of a fluid in a porous medium, which in turn represents the runner system geometry of a thermoplastic injection mold. The model comprises a variational autoencoder network and multilayer perceptron. The characteristics extracted from the variational autoencoder are used as a set of multilayer perceptron output vectors, whereas the input data are determined by the positions of the input and outputs of the injection channels. The field of the inverse permeability in each case was obtained by topology optimization using a heuristic algorithm. The trained model displayed statistical metrics indicating good performance and task generalizability. The mean absolute error was 0.0106 for the entire dataset. Speed up compared to traditional computational fluid dynamics software was 625 times faster in one case. For 150 cases, it was 76907 times faster⁠. The purpose of generating a deep learning model in this area is to reduce the design time of injection molds and the computational requirements. The developed neural network reduces the runner system volume by 16% or its hydraulic resistance by 25%.}, archive = {J_EAAI}, author = {Evgenii Kurkin and Jose Gabriel Quijada Pioquinto and Vladislava Chertykovtseva and Evgenii Minaev}, doi = {10.1016/j.engappai.2025.110555}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110555}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Runner system geometry prediction using variational autoencoder deep learning model}, volume = {150}, year = {2025}, }
• (2025). Feature selection considering synergy between features based on soft neighborhood rough sets. EAAI, 150, 110553. (www)

  @article{ , abstract = {Neighborhood entropy-based measures provide a powerful framework for feature selection to select features that are more useful for classification. However, most of these feature selection methods do not pay attention to the complementarities and synergies between features, as well as the interactions between them. In addition, most existing neighborhood rough sets are subjective in the determination of neighborhood radius when dealing with classification problems, which may lead to the omission of useful information. To solve these problems, a soft neighborhood rough set model-based feature selection method (SNCMI) is proposed. Firstly, the method dynamically adjusts the neighborhood radius, significantly minimizing its influence on the uncertainty measurement. Secondly, it comprehensively considers the correlation, redundancy, complementarity, and synergy between features through soft neighborhood uncertainty measures. Thirdly, an innovative objective evaluation function is introduced to evaluate the interactions between features. Finally, we compare the proposed SNCMI algorithm with several well-known feature selection algorithms on twenty public datasets and demonstrate the effectiveness of SNCMI.}, archive = {J_EAAI}, author = {Lubin Chen and Jinkun Chen and Yaojin Lin}, doi = {10.1016/j.engappai.2025.110553}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110553}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Feature selection considering synergy between features based on soft neighborhood rough sets}, volume = {150}, year = {2025}, }
• (2025). Forecasting train travel times of China–Europe railway express through a hybrid deep learning model optimized with a bandit-based approach. EAAI, 150, 110552. (www)

  @article{ , abstract = {With the globalization of economic trade, the China–Europe Railway Express (CRE) has emerged as a crucial means of international freight transportation. However, since the travel process of CRE trains is subject to various factors (e.g., customs clearance efficiency, weather changes, etc.), existing models struggle to handle the complex nonlinear characteristics of the travel time data, failing to achieve accurate train travel time predictions. This significantly affects the scheduling and utilization of capacity resources along the CRE routes. To address this issue, this study proposes a novel hybrid deep learning model, i.e., Discrete Wavelet Transform (DWT)-Convolutional Neural Networks (CNN)-Bidirectional Gated Recurrent Unit (BiGRU) (DWT-CNN-BiGRU). Specifically, the DWT technique is first used to preprocess historical train travel time data to reduce noise interference and improve data quality. Then, the CNN module focuses on extracting local spatial features from the data, whereas the BiGRU module emphasizes its long-term temporal dependencies. Furthermore, a bandit-based approach is applied to hyperparameter optimization to further exploit model potentials. By testing on a real-life CRE dataset, the DWT-CNN-BiGRU model demonstrates superior prediction accuracy with root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) values respectively equal to 10.7347 h, 7.5482 h, and 2.2034%, and it outperforms the other ten popular baseline models. In conclusion, the proposed DWT-CNN-BiGRU model features a lightweight structure and strong robustness, offering reliable technical support to alleviate capacity resource shortages and improve the service quality of CRE.}, archive = {J_EAAI}, author = {Yongxiang Zhang and Liting Gu and Jingwei Guo and Xu Yan and Xin Hu and Zhen-Song Chen}, doi = {10.1016/j.engappai.2025.110552}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110552}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forecasting train travel times of China–Europe railway express through a hybrid deep learning model optimized with a bandit-based approach}, volume = {150}, year = {2025}, }
• (2025). A hybrid architecture of sparse convolutional neural network-transformer for enhanced spatial-geometric feature learning in surface reconstruction. EAAI, 150, 110550. (www)

  @article{ , abstract = {Learning-based methods have garnered significant attention in indoor scene reconstruction tasks. However, researchers have often overlooked the crucial role of the surface prediction stage. Our study specifically focuses on this phase. According to our experiments and analysis, this phase primarily addresses spatial voxel occupancy and geometric structure maintenance. Simple structural designs are insufficient to effectively solve these problems. To address these challenges, we propose a hybrid model, which combines the strengths of Convolution Neural Networks and Transformer architectures for fine reconstruction. Additionally, we introduce several new techniques, including the Sparse Positional Attention mechanism, Sparse Channel Decoding Block, and Mixed Feature Fusion mechanism. These techniques, leveraging the characteristics of sparse computation, enhance feature utilization in both spatial and channel dimensions. With limited training and testing resources, our network achieves optimal results on the ScanNet dataset, improving precision and F-score by 2.1% and 1.6%, respectively, and reducing the Chamfer distance to 0.055 m. To our knowledge, our model is the first use of hybrid structures in the surface prediction phase of an indoor scene reconstruction task. Moreover, we hope that our design and analysis can provide a new paradigm for task network design in this phase.}, archive = {J_EAAI}, author = {Mingyang Li and Wei Zhang and Yanyan Liu and Xiang Feng and Changsong Liu and Yimeng Fan and Lixue Xu}, doi = {10.1016/j.engappai.2025.110550}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110550}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid architecture of sparse convolutional neural network-transformer for enhanced spatial-geometric feature learning in surface reconstruction}, volume = {150}, year = {2025}, }
• (2025). Enhancing pose estimation for mobile robots: A comparative analysis of deep reinforcement learning algorithms for adaptive extended kalman filter-based estimation. EAAI, 150, 110548. (www)

  @article{ , abstract = {The Extended Kalman Filter (EKF) is a widely used algorithm for state estimation in control systems. However, its lack of adaptability limits its performance in dynamic and uncertain environments. To address this limitation, we used an approach that leverages Deep Reinforcement Learning (DRL) to achieve adaptive state estimation in the EKF. By integrating DRL techniques, we enable the state estimator to autonomously learn and update the values of the system dynamics and measurement noise covariance matrices, Q and R, based on observed data, which encode environmental changes or system failures. In this research, we compare the performance of four DRL algorithms, namely Deep Deterministic Policy Gradient (DDPG), Twin Delayed Deep Deterministic Policy Gradient (TD3), Soft Actor-Critic (SAC), and Proximal Policy Optimization (PPO), in optimizing the EKF’s adaptability. The experiments are conducted in both simulated and real-world settings using the Gazebo simulation environment and the Robot Operating System (ROS). The results demonstrate that the DRL-based adaptive state estimator outperforms traditional methods in terms of estimation accuracy and robustness. The comparative analysis provides insights into the strengths and limitations of different DRL agents, showing that the TD3 and the DDPG are the most effective algorithms, with TD3 achieving superior performance, resulting in a 91% improvement over the classic EKF, due to its delayed update mechanism that reduces training noise. This research highlights the potential of DRL to advance state estimation algorithms, offering valuable insights for future work in adaptive estimation techniques.}, archive = {J_EAAI}, author = {Islem Kobbi and Abdelhak Benamirouche and Mohamed Tadjine}, doi = {10.1016/j.engappai.2025.110548}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110548}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing pose estimation for mobile robots: A comparative analysis of deep reinforcement learning algorithms for adaptive extended kalman filter-based estimation}, volume = {150}, year = {2025}, }
• (2025). Graph neural networks with scattering transform for network anomaly detection. EAAI, 150, 110546. (www)

  @article{ , abstract = {As cyber-attacks become increasingly sophisticated and frequent, the demand for advanced and proactive Network Intrusion Detection Systems (NIDS) has become more urgent than ever. To address critical shortcomings in existing NIDS approaches, such as high false-positive rates that trigger unnecessary alerts, inability to capture complex relationships between network nodes, and oversimplified node representation initialization that fails to reflect real-world network behaviors, we introduce a novel solution called Scattering Transform Edge Graph (STEG). STEG harnesses the wavelet scattering transform to extract edge feature information and employs a graph-based representation to effectively capture the topological relationships between network nodes. Additionally, we enhance STEG by incorporating node embedding techniques like DeepWalk for initializing node representations, moving beyond conventional uniform initialization methods. Comprehensive evaluations on benchmark NIDS datasets reveal that STEG outperforms current state-of-the-art methods. Moreover, the integration of Node2Vec-based initialization further boosts performance, marking a significant advancement in the effectiveness of network intrusion detection systems.}, archive = {J_EAAI}, author = {Abdeljalil Zoubir and Badr Missaoui}, doi = {10.1016/j.engappai.2025.110546}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110546}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph neural networks with scattering transform for network anomaly detection}, volume = {150}, year = {2025}, }
• (2025). Comparative analysis and evaluation of ageing forecasting methods for semiconductor devices in online health monitoring. EAAI, 150, 110545. (www)

  @article{ , abstract = {Semiconductor devices, especially MOSFETs (Metal–oxide–semiconductor field-effect transistor), are crucial in power electronics, but their reliability is affected by ageing processes influenced by cycling and temperature. The primary ageing mechanism in discrete semiconductors and power modules is the bond wire lift-off, caused by crack growth due to thermal fatigue. The process is empirically characterized by exponential growth and an abrupt end of life, making long-term ageing forecasts challenging. This research presents a comprehensive comparative assessment of different forecasting methods for MOSFET failure forecasting applications. Classical tracking, statistical forecasting and Neural Network (NN) based forecasting models are implemented along with novel Temporal Fusion Transformers (TFTs). A comprehensive comparison is performed assessing their MOSFET ageing forecasting ability for different forecasting horizons. For short-term predictions, all algorithms result in acceptable results, with the best results produced by classical NN forecasting models at the expense of higher computations. For long-term forecasting, only the TFT is able to produce valid outcomes owing to the ability to integrate covariates from the expected future conditions. Additionally, TFT attention points identify key ageing turning points, which indicate new failure modes or accelerated ageing phases.}, archive = {J_EAAI}, author = {Adrian Villalobos and Iban Barrutia and Rafael Peña-Alzola and Tomislav Dragicevic and Jose I. Aizpurua}, doi = {10.1016/j.engappai.2025.110545}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110545}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comparative analysis and evaluation of ageing forecasting methods for semiconductor devices in online health monitoring}, volume = {150}, year = {2025}, }
• (2025). Prediction of the flexural strength and elastic modulus of cementitious materials reinforced with carbon nanotubes: An approach with artificial intelligence. EAAI, 150, 110544. (www)

  @article{ , abstract = {Over the past decade, researchers investigated incorporating carbon nanotubes (CNTs) to improve the mechanical properties of cementitious materials. Recently, few studies developed Machine Learning (ML)-based predictive models to maximize insights from limited experimental data. However, these models often fail to identify key parameters and their complex correlations with mechanical properties. This study aims to improve the prediction of the mechanical properties of CNT-reinforced cementitious materials, specifically, elastic modulus and flexural strength, by leveraging multiple predictive Artificial Intelligence (AI)-based models. Deep Neural Networks (DNN), ensemble-bagging, and Support Vector Regression (SVR) were proposed and rigorously tested to predict the flexural strength and elastic modulus of the composite material. The feature selection was performed based on the domain knowledge and the informative metrics including the permutation importance analyses and Pearson's correlation analyses. The research identified several parameters that have traditionally been overlooked but proved to be critical. With a total of nineteen input parameters analyzed, the findings indicate that the mechanical properties of the composite material are primarily influenced by surfactant-to-CNT mass ratio, CNT content and physical properties, as well as ultrasonication process. Conversely, sand type and CNT purity are found to have minimal importance to the change in mechanical properties. In addition, the proposed DNN models outperform other ML models in predicting both flexural strength and elastic modulus, achieving R-squared values of 0.93 and 0.86 with mean absolute percentage errors of 8.16% and 7.22%, respectively.}, archive = {J_EAAI}, author = {Mahyar Ramezani and Do-Eun Choe and Abdur Rasheed}, doi = {10.1016/j.engappai.2025.110544}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110544}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction of the flexural strength and elastic modulus of cementitious materials reinforced with carbon nanotubes: An approach with artificial intelligence}, volume = {150}, year = {2025}, }
• (2025). Reparameterization convolutional neural networks for handling imbalanced datasets in solar panel fault classification. EAAI, 150, 110541. (www)

  @article{ , abstract = {Solar photovoltaic technology has grown significantly as a renewable energy, with unmanned aerial vehicles equipped with thermal infrared cameras effectively inspecting solar panels. However, long-distance capture and low-resolution infrared cameras make the targets small, complicating feature extraction. Additionally, the large number of normal photovoltaic modules results in a significant imbalance in the dataset. Furthermore, limited computing resources on unmanned aerial vehicles further challenge real-time fault classification. These factors limit the performance of current fault classification systems for solar panels. The multi-scale and multi-branch Reparameterization of convolutional neural networks can improve model performance while reducing computational demands at the deployment stage, making them suitable for practical applications. This study proposes an efficient framework based on reparameterization for infrared solar panel fault classification. We propose a Proportional Balanced Weight asymmetric loss function to address the class imbalance and employ multi-branch, multi-scale convolutional kernels for extracting tiny features from low-resolution images. The designed models were trained with Exponential Moving Average for better performance and reparameterized for efficient deployment. We evaluated the designed models using the Infrared Solar Module dataset. The proposed framework achieved an accuracy of 83.8% for the 12-Class classification task and 74.0% for the 11-Class task, both without data augmentation to enhance generalization. The accuracy improvements of up to 16.4% and F1-Score gains of up to 18.7%. Additionally, we achieved an inference speed that is 3.4 times faster than the training speed, while maintaining high fault classification performance.}, archive = {J_EAAI}, author = {Jielong Guo and Chak Fong Chong and Pedro Henriques Abreu and Chao Mao and Jiaxuan Li and Chan-Tong Lam and Benjamin K. Ng}, doi = {10.1016/j.engappai.2025.110541}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110541}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reparameterization convolutional neural networks for handling imbalanced datasets in solar panel fault classification}, volume = {150}, year = {2025}, }
• (2025). Blockchain data with ransomware detection based on deep feed forward maxout network. EAAI, 150, 110538. (www)

  @article{ , abstract = {The widespread use of the internet and technology has become more common and essential in daily life. The main risk to the network is malware, and ransomware is considered a destructive kind of malware. The ransomware resulted in massive data losses and produced huge financial losses. In order to overcome this issue, the Deep Feed Forward Maxout Network (DFFMN) is developed to detect ransomware using blockchain data in this study. To accomplish this, initially, the input data from the blockchain is given to feature extraction to extract features. Then, data normalization is performed and the extracted features are fused together using a Deep Belief Network (DBN) with Lorentzian similarity. Lastly, ransomware detection is executed by utilizing the DFFMN technique, which is created by the integration of Deep Maxout Network and Deep Feedforward Neural Network. The experimental results exemplify that the proposed DFFMN acquired accuracy value of 91.57 %, sensitivity value of 91.04 %, precision value of 89.35 %, False Negative Rate (FNR) of 0.086, False Positive Rate (FPR) of 0.090 and F-Measure of 89.44 %.}, archive = {J_EAAI}, author = {Vemireddi Srinadh and Buddi Padmaja and Dhanunjaya Rao Chigurukota and Mallikharjuna Rao Karreddula and Balajee Maram and Smritilekha Das}, doi = {10.1016/j.engappai.2025.110538}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110538}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Blockchain data with ransomware detection based on deep feed forward maxout network}, volume = {150}, year = {2025}, }
• (2025). Dual-branch crack segmentation network with multi-shape kernel based on convolutional neural network and mamba. EAAI, 150, 110536. (www)

  @article{ , abstract = {Cracks are one of the most common pavement diseases. If not promptly repaired, they will hasten the deterioration of the road. Semantic segmentation is the most convenient pavement crack detection method to assess the damage level. Convolutional neural networks (CNN) excel at extracting local spatial information, but they have limitations in capturing global contextual information. Therefore, a dual-branch crack segmentation network (DBCNet) with Mamba and multi-shape convolutional kernels is proposed. First, a dual-branch encoder is employed to extract both spatial and contextual information, consisting of the spatial branch and the context branch. The cross-like block (CrossBlock) that excels in extracting spatial information horizontally and vertically from cracks is proposed. Multiple CrossBlocks are stacked to construct a lightweight network as a spatial branch. The improved Visual State Space Model (VMamba) serves as a context branch for modeling long-range dependencies for more accurate pixel-by-pixel segmentation. Second, the Feature Fusion Module (FFM), based on squeeze-and-excitation attention, is constructed to dynamically fuse the features from the two branches layer by layer. Third, a Cross-aware Mamba Module (CMM) with the hybrid CNN-Mamba architecture is proposed to compose the decoder. Fourth, comprehensive evaluations were conducted on three public datasets. Performs on multiple metrics achieved considerable progress, outperforming the seven state-of-the-art models. The mean intersection over union (mIoU) on Deepcrack, CrackTree 260, and CFD reached 87.87%, 85.34%, and 81.35%, respectively. Code and data will be available at https://github.com/name191/DBCNet .}, archive = {J_EAAI}, author = {Jianming Zhang and Dianwen Li and Zhigao Zeng and Rui Zhang and Jin Wang}, doi = {10.1016/j.engappai.2025.110536}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110536}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-branch crack segmentation network with multi-shape kernel based on convolutional neural network and mamba}, volume = {150}, year = {2025}, }
• (2025). FA-SconvAE-LSTM: Feature-aligned stacked convolutional autoencoder with long short-term memory network for soft sensor modeling. EAAI, 150, 110535. (www)

  @article{ , abstract = {The advancement of soft sensor technology has enabled the real-time estimation of critical parameters in complex industrial processes, where direct measurement through hardware sensors is often infeasible. Industrial process data typically exhibit both spatial correlations and temporal dependencies, necessitating sophisticated modeling approaches to capture these characteristics effectively. In this study, a spatio-temporal model, termed the feature-aligned stacked convolutional autoencoder with long short-term memory, is proposed to develop soft sensors for nonlinear dynamic industrial processes. The proposed model begins with the systematic training of a stacked convolutional autoencoder using a layer-by-layer pre-training technique. This approach facilitates the extraction of high-level spatial feature representations from the process variables. To address the issue of feature misalignment in the spatial features extracted by the stacked convolutional autoencoder, a feature alignment strategy is implemented, ensuring that the extracted spatial features are properly aligned. Subsequently, the aligned spatial features are fed into a long short-term memory network to capture temporal dependencies, with quality variables serving as the output for soft sensor development. The effectiveness and superiority of the proposed method are demonstrated through experiments conducted on two industrial processes: the sulfur recovery unit and the multiphase flow process. Comparative analyses with other state-of-the-art methods reveal that the proposed model achieves the highest performance, with R 2 values of 0.86222 for the sulfur recovery unit and 0.94307 for the multiphase flow process, outperforming all compared methods.}, archive = {J_EAAI}, author = {Ping Wu and Zengdi Miao and Ke Wang and Jinfeng Gao and Xujie Zhang and Siwei Lou and Chunjie Yang}, doi = {10.1016/j.engappai.2025.110535}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110535}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {FA-SconvAE-LSTM: Feature-aligned stacked convolutional autoencoder with long short-term memory network for soft sensor modeling}, volume = {150}, year = {2025}, }
• (2025). Surrogate-assisted motion planning and layout design of robotic cellular manufacturing systems. EAAI, 150, 110530. (www)

  @article{ , abstract = {A surrogate-assisted multi-objective evolutionary algorithm is proposed for simultaneous optimization of robot motion planning and layout design in robotic cellular manufacturing systems. A sequence-pair is used to represent the layout of components in a robotic cell to avoid overlapping in the evolutionary computation. The robot motion planning with Rapidly exploring Random Trees Star (RRT*) is applied to compute the total operation time of a robot arm for each layout. Non-dominated Sorting Genetic Algorithm II (NSGA-II) is used to minimize the total required layout area and the operation time for a robot arm. The proposed surrogate model can estimate the robot’s operation time with 98% of accuracy without explicit computations of the motion planning algorithm. The experimental results with a physical 6 Degree of Freedom (DOF) manipulator show that the total computation time is approximately 1/400, significantly shorter than the conventional methods.}, archive = {J_EAAI}, author = {Tomoya Kawabe and Tatsushi Nishi and Ziang Liu and Tomofumi Fujiwara}, doi = {10.1016/j.engappai.2025.110530}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110530}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Surrogate-assisted motion planning and layout design of robotic cellular manufacturing systems}, volume = {150}, year = {2025}, }
• (2025). Self-labeled framework with semi-supervised ball K-means clustering-based synthetic example generation for semi-supervised classification in industrial applications. EAAI, 150, 110528. (www)

  @article{ , abstract = {While self-labeled methods can exploit labeled and unlabeled instances to train classifiers, they are severely restricted by the labeled instance number and distribution. One category of the existing solutions tends to combine oversampling techniques, with the goal of creating labeled synthetic instances to improve the labeled instance number and distribution. The synthetic example generation-based framework for semi-supervised classification (SEG-SSC) is the only state-of-the-art instance of the above category. Nevertheless, it still suffers from the following issues: a) relying on 7 hyper-parameters, b) ineffectively improving the labeled instance number and distribution in sparser regions with fewer labeled instances, and c) having a relatively high time complexity of O ( nlogn + G×n ). To this end, a self-labeled framework with semi-supervised ball k -means clustering-based synthetic example generation (SEGBallKmeans-SSC), having only one hyper-parameter and the time complexity of O ( n ), is proposed for semi-supervised classification. The main uniqueness is that: a) firstly, a semi-supervised ball k -means clustering (SSBallKmeans) with a compact-cluster assumption is proposed to divide semi-supervised data into compact ball clusters, intending to reveal regions with different labeled instance numbers; b) secondly, an SSBallKmeans-based oversampling method (OMSSBallKmean) is proposed to create more labeled synthetic instances on compact ball clusters with fewer labeled instances, intending to improve the labeled instance number and distribution, especially on sparser regions with fewer labeled instances. After that, any self-labeled method are executed on improved labeled instances and unlabeled instances to train more accurate classifiers. Experiments have proven that SEGBallKmeans-SSC outperforms 7 state-of-the-art self-labeled frameworks on extensive benchmark datasets from various industrial applications.}, archive = {J_EAAI}, author = {Junnan Li and Lufeng Wang and Shun Fu ( Revision ) and Wei Fu and Xin Pan}, doi = {10.1016/j.engappai.2025.110528}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110528}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-labeled framework with semi-supervised ball K-means clustering-based synthetic example generation for semi-supervised classification in industrial applications}, volume = {150}, year = {2025}, }
• (2025). A deep learning-based sentiment flow analysis model for predicting financial risk of listed companies. EAAI, 150, 110522. (www)

  @article{ , abstract = {With the advancement of natural language processing technology, more studies are applying deep learning models to extract features from unstructured data and predict corporate financial risk through horizontal comparisons. This paper proposes the Sentiment Flow Analysis (SFA) model designed to capture the nuanced emotional dynamics present in corporate annual reports and analyst reports from a longitudinal perspective. By leveraging advanced contextual embeddings from a Transformer architecture and employing a sophisticated recurrent neural network, the model effectively processes sequential data, allowing for a comprehensive understanding of sentiment trends over a five-year period. The effectiveness of our model was validated through experiments predicting the removal of special treatment (ST) status for 344 listed companies under special treatment in 2022 and 2023, achieving a prediction accuracy of up to 82.27%. Compared to state-of-the-art models in the same field, our model demonstrates improvements in prediction accuracy and recall, showcasing the application of Artificial Intelligence (AI) in financial risk assessment.}, archive = {J_EAAI}, author = {Feifei Tao and Wenya Wang and Rongke Lu}, doi = {10.1016/j.engappai.2025.110522}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110522}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep learning-based sentiment flow analysis model for predicting financial risk of listed companies}, volume = {150}, year = {2025}, }
• (2025). Efficient self-learning disturbance-resistant control for high-speed flight vehicle based on dual heuristic dynamic programming. EAAI, 150, 110521. (www)

  @article{ , abstract = {Recent advancements in high-speed flight vehicles (HSFVs) have sparked significant interest due to their strategic importance and emerging civilian applications. These vehicles exhibit strong nonlinearities and multi-axis interactions and are usually influenced by uncertainties such as modeling errors, parameter perturbations, and external disturbances. Neglecting these challenges in attitude controller design can lead to trajectory tracking deviations and potential mission failure due to instability. Motivated by this issue, an efficient disturbance-resistant control method with online self-learning capability is proposed. Firstly, a feedback linearization baseline controller combined with finite-time extended state observers (FESOs) is designed to ensure stability. Next, a dual heuristic dynamic programming (DHP) controller with critic-only structure is developed for online performance optimization. Update laws of the critic neural network (NN) are derived based on policy iteration, and zero-sum game (ZSG) theory is incorporated to enhance the system’s adaptive capacity to uncertainties. Lyapunov theory is subsequently employed to validate the convergence of network weights and the system stability. The proposed method, compared to common adaptive dynamic programming (ADP) approaches for attitude control, demonstrates superior learning efficiency and guarantees the convergence of online learning without the necessity for pre-training. Simulation results indicate that the method equips the HSFV with robust dynamic performance throughout a broad flight envelope, with attitude tracking errors constrained to less than 0.5°. Future research will focus on developing fault-tolerant and prescribed performance control frameworks with online learning ability, representing an advancement in the current technique.}, archive = {J_EAAI}, author = {Xu Huang and Jiarun Liu and Yue Peng and Yuan Zhang and Zhaolei Wang and Weimin Bao}, doi = {10.1016/j.engappai.2025.110521}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110521}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient self-learning disturbance-resistant control for high-speed flight vehicle based on dual heuristic dynamic programming}, volume = {150}, year = {2025}, }
• (2025). Remote sensing scene classification with relation-aware dynamic graph neural networks. EAAI, 150, 110513. (www)

  @article{ , abstract = {Remote sensing scene classification (RSSC) is challenging due to the complexity and diversity of scenes. Existing methods need help to capture long-range and structural relationships among image regions, limiting their performance. This paper proposes a novel graph-based model that learns relation-aware dynamic graph representations for remote sensing scene classification tasks. The proposed model consists of three main components: Multi-Scale Feature Extraction (MSFE), Relation-Aware Graph Processing (RAGP), and Scene Classification with Weighted Pooling (SCWP). MSFE uses a multi-scale feature extraction strategy to generate low-level feature nodes from remote sensing images. RAGP applies several cascaded graph processing blocks to dynamically learn the relations between nodes in high-level semantic spaces using relation-aware graph convolutional and node feature update operators. SCWP performs weighted pooling on the learned node features from RAGP to obtain global representations of remote sensing images and makes scene decisions using a fully feed-forward network-based classifier. We evaluate our model on three benchmark datasets and compare it with state-of-the-art RSSC methods. Our experimental results show that our model outperforms existing methods on all three datasets, demonstrating the effectiveness of a graph-based model with the proposed techniques for RSSC tasks.}, archive = {J_EAAI}, author = {Qionghao Huang and Fan Jiang and Changqin Huang}, doi = {10.1016/j.engappai.2025.110513}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110513}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Remote sensing scene classification with relation-aware dynamic graph neural networks}, volume = {150}, year = {2025}, }
• (2025). Learning adaptive distractor-aware-suppression appearance model for visual tracking. EAAI, 150, 110511. (www)

  @article{ , abstract = {Some algorithms based on Siamese networks aim to improve the representation of target by combining background and target information, but they seldom consider adjusting the influence of background distractors on appearance modeling. In this paper, we propose an adaptive distractor suppression appearance model for robust visual tracking. Firstly, to fully utilize the valuable clues provided by the background, a new distractor model is specially designed to determine the weight of each distractor based on the similarity between the distractor and the target. This model adaptively fuses the distractors according to their weights, thereby focusing on distractors that are highly similar to the target. Secondly, a distractor model transformation strategy is constructed to rank the influence of the distractor model on appearance modeling, which mines the similarity relationship between the background distractor and target using regularized linear regression, effectively controlling the influence of the distractor model. Finally, we unify them into a learning adaptive distractor-aware-suppression appearance model for improving the discriminant ability of the appearance model, which selectively introduces the distractor model to suppress distractors according to the intensity of the distractor, achieving robust tracking in the presence of background interference. Experimental results on six benchmarks demonstrate that the proposed tracker achieves excellent performance in various challenging tracking tasks, particularly when facing background interference, where the tracking precision and success rate of our algorithm reach state-of-the-art levels.}, archive = {J_EAAI}, author = {Huanlong Zhang and Linwei Zhu and Yanchun Zhao and Fusheng Li and Deshuang Huang}, doi = {10.1016/j.engappai.2025.110511}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110511}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Learning adaptive distractor-aware-suppression appearance model for visual tracking}, volume = {150}, year = {2025}, }
• (2025). A multiagent architecture for industrial internet of things safety applications. EAAI, 150, 110495. (www)

  @article{ , abstract = {Industrial Internet of Things (IIoT) is a key technological pillar of the Fourth Industrial Revolution, also known as Industry 4.0. In this context, an area of considerable interest is human safety technology. Solutions rely on multiple sensors connected to a central monitoring system and supported with software to autonomously or semi-autonomously identify safety hazards. To this end, Computer vision systems are leveraged. However, streaming continuous video from numerous sensors can strain network resources, risking timely hazard response in large industrial setups. This work proposes a reference IIoT architecture based on Multi-Agent Systems to manage safety risks. It allows for scalable sensor integration and dynamically assesses sensor input based on risk levels. To prevent network overload, the architecture uses sensor-level intelligence at the edge layer to assess situational risks and decide whether to forward video signals to a centralized local cloud agent. The central cloud agent, using strategies like ensemble learning, selectively requests additional data from distributed edge agents based on the diagnosed risk. This approach was tested in monitoring safety during aircraft assembly, showing that edge processing reduces network load by limiting unnecessary data transmission without compromising accuracy. This architecture effectively distributes processing to the edge, maintaining detection accuracy while minimizing network traffic compared to continuous centralized video transmission.}, archive = {J_EAAI}, author = {Gibson Barbosa and Djamel F.H. Sadok and Judith Kelner and Luis Ribeiro}, doi = {10.1016/j.engappai.2025.110495}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110495}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multiagent architecture for industrial internet of things safety applications}, volume = {150}, year = {2025}, }
• (2025). A systematic review of end-to-end framework for contactless fingerprint recognition: Techniques, challenges, and future directions. EAAI, 150, 110493. (www)

  @article{ , abstract = {Contactless fingerprint biometrics have seen rapid advancements in the recent years due to its intrinsic advantages such as resilience against latent fingerprints, and enhanced hygiene due to the absence of physical contact between a finger and the sensor. These advantages boosted the development of novel techniques for contactless fingerprint recognition. An exponentially increasing number of publications related to these developments are becoming part of the literature. However, no systematic review that consolidates these developments has been presented to date, thereby leaving a significant void. Hence, there is a need to fill this void by presenting a comprehensive review of contactless fingerprint biometric technology. A review of this kind will be highly beneficial for individuals keen on pursuing research in this domain. This study presents a systematic review of the methods used in an end-to-end framework for contactless fingerprint recognition, including acquisition, segmentation, enhancement, feature extraction, and matching, using both traditional and deep learning techniques. As per the review protocol and inclusion-exclusion criteria, 112 papers have been finally included in this review. The primary focus of the review is to present the underlying methods, their reported performance outcomes, and their strengths and weaknesses. The review evaluates the recent research findings, highlights the research issues that have been effectively addressed, presents the biases in the studies, identifies ongoing challenges that remain in the field, and provides the future research directions.}, archive = {J_EAAI}, author = {Pooja Kaplesh and Aastha Gupta and Divya Bansal and Sanjeev Sofat and Ajay Mittal}, doi = {10.1016/j.engappai.2025.110493}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110493}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A systematic review of end-to-end framework for contactless fingerprint recognition: Techniques, challenges, and future directions}, volume = {150}, year = {2025}, }
• (2025). Thoughtful and cautious reasoning: A fine-tuned knowledge graph-based multi-hop question answering framework. EAAI, 150, 110479. (www)

  @article{ , abstract = {The aim of Knowledge Graph Question Answering (KGQA) is to find the answer entity by utilizing the Knowledge Graph (KG). Despite remarkable successes in recent years, the existing multi-hop KGQA research still faces numerous challenges. First, a multi-hop question often contains multiple entities and their relationships, and the semantic information is complex. The current methods extract the semantics of the question through an encoder that cannot completely extract the complex and rich semantic information in the multi-hop questions. Second, current question answering models use the coarse information filtering mechanism in the process of reasoning, which lead to the loss of effective information and introduce additional noise. To address these issues, we propose a Thoughtful and Cautious Reasoning framework for Knowledge Graph Question Answering (TCR-KGQA). We design a new question encoder that can extract and fully fuse the local semantic information of the question at different levels, focusing on the unique local features of the multi-hop question text. Based on the advantages of Gated Recurrent Unit (GRU) for information filtering, we propose a loop instruction update framework based on residual-GRU to effectively capture key information in the reasoning process. Extensive experiments on three broad benchmark datasets demonstrate the effectiveness of our model on KGQA tasks, and it also yields excellent results in the case of incomplete knowledge graphs with missing question–answer pairs.}, archive = {J_EAAI}, author = {Yinghao Zheng and Ling Lu and Yang Hu and Yinong Chen and Aijuan Wang}, doi = {10.1016/j.engappai.2025.110479}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110479}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Thoughtful and cautious reasoning: A fine-tuned knowledge graph-based multi-hop question answering framework}, volume = {150}, year = {2025}, }
• (2025). A surrogate-assist quasi-affine transformation evolutionary for multi-objective optimization of empty train deployment on heavy-haul railways. EAAI, 150, 110475. (www)

  @article{ , abstract = {Surrogate-assisted evolutionary algorithms have become increasingly popular for solving expensive and time-consuming single-objective and multi-objective problems. However, in multi-objective optimization, the Pareto non-dominated solution set space can quickly become computationally intractable, making it challenging to select which samples to evaluate using the expensive fitness function. In this paper, we propose a Surrogate-Assist Quasi-Affine Transformation Evolutionary algorithm (SA-QUATRE/MO) for solving multi-objective optimization problems. The SA-QUATRE/MO algorithm uses a radial basis function network as a surrogate model to improve the speed of the algorithm operation by replacing the expensive fitness evaluation with the surrogate model. To ensure the excellence and diversity of the selected samples while keeping the archived sample space fixed, we propose a technique called Vector Space Sampling, which samples objective points in the current set of non-dominated solutions by dividing several sub-vector spaces. Additionally, we propose an uncertain sample infilling strategy to select samples for real fitness evaluation using a designed uncertainty function. We compare the SA-QUATRE/MO algorithm with three state-of-the-art algorithms for multi-objective problems in three test function suites. Finally, we applied the SA-QUATRE/MO algorithm to optimize empty train deployment in a heavy-haul railway at the loading end and build a model based on the S12 section of a specific railway. The final experimental results demonstrate the practicality and effectiveness of our proposed method.}, archive = {J_EAAI}, author = {Zhi-Gang Du and Jeng-Shyang Pan and He-Ying Xu and Shu-Chuan Chu and Shao-Quan Ni}, doi = {10.1016/j.engappai.2025.110475}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110475}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A surrogate-assist quasi-affine transformation evolutionary for multi-objective optimization of empty train deployment on heavy-haul railways}, volume = {150}, year = {2025}, }
• (2025). Navigating beyond the training set: A deep learning framework for inverse design of architected composite materials. EAAI, 150, 110473. (www)

  @article{ , abstract = {This study introduces a deep learning (DL)-based inverse design framework for two-phase composite materials. The artificial intelligence (AI) contribution lies in the integration of Deep Convolutional Generative Adversarial Networks (DCGAN) and Convolutional Neural Networks (CNN) into a framework that enhances material discovery and design, particularly for out-of-distribution (OOD) targets. The major contribution is the development of a strategy that balances latent space exploration and optimization, achieving low design errors – below 10% – for targeted properties located in near- and extreme-OOD regions of the material property space (MPS). The engineering application focuses on designing composites with tailored linear elastic properties, accelerating inverse design and reducing reliance on traditional simulation-based approaches. An image dataset of 12,000 Representative Unit Cells (RUCs) was assembled using a parametric Voronoi diagram generator, with elastic responses computed through finite element (FE) simulations. The DCGAN generated synthetic samples with novel features not present in the original dataset, demonstrating interpolation and extrapolation capabilities. A single round of Active Learning (AL) and Transfer Learning (TL) enhanced the CNN’s predictive accuracy on synthetic data. The framework offers significant computational efficiency, with optimization complexity O ( m ⋅ n 2 ) , where m is the number of iterations and n the latent vector dimensionality. This complexity is considerably lower than that of direct FE-based topology optimization, which ranges from O ( m ⋅ N 4 ) to O ( m ⋅ N 6 ) , where N × N represents the RUC grid size. These findings demonstrate the scalability and adaptability of the framework for advanced material design and engineering applications.}, archive = {J_EAAI}, author = {José Pablo Quesada-Molina and Hossein Mofatteh and Abdolhamid Akbarzadeh and Stefano Mariani}, doi = {10.1016/j.engappai.2025.110473}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110473}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Navigating beyond the training set: A deep learning framework for inverse design of architected composite materials}, volume = {150}, year = {2025}, }
• (2025). Thermal buckling of graphene platelets reinforced microplates with piezoelectric layers using artificial neural network. EAAI, 150, 110469. (www)

  @article{ , abstract = {Functionally graded microplates are extensively employed in advanced engineering applications, including aerospace, micro-electromechanical systems, and smart structures, due to their exceptional mechanical performance, thermal resistance, and adaptability to multifunctional environments. This study examines the thermal buckling behavior of functionally graded graphene platelet-reinforced composite microplates integrated with piezoelectric layers under externally applied voltage. The modified couple stress theory is utilized to account for micro-scale effects, and the material properties of the composite layers are determined using the Halpin-Tsai micromechanical model. The governing equations based on first shear deformation theory are solved using the Ritz method to generate training data for an artificial neural network (ANN). To address computational challenges inherent in conventional methods, an ANN-based framework, leveraging the Levenberg-Marquardt algorithm, is developed to predict the critical buckling temperature with high precision and significantly reduced computational effort. The nanofiller dimensions, weight fraction, and piezoelectric layer thickness serve as inputs, while the thermal buckling load is the output. The obtained results show that the ANN offers a significant reduction in computational time, achieving speed improvements of over 85% across all cases. Also, the ANN reliably predicts the critical buckling temperatures, achieving a maximum discrepancy of only 0.26% when compared with the Ritz method. The novelty of this work lies in combining modified couple stress theory with ANN-assisted prediction models for efficient thermal buckling analysis. This methodology offers a practical solution for the rapid and reliable design of graphene platelet-reinforced composite microplates integrated with piezoelectric layers in applications demanding high precision and performance under thermal loading.}, archive = {J_EAAI}, author = {Hongxia Liu and Qiyu Wang and Zilin Zhang}, doi = {10.1016/j.engappai.2025.110469}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110469}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Thermal buckling of graphene platelets reinforced microplates with piezoelectric layers using artificial neural network}, volume = {150}, year = {2025}, }
• (2025). Optimization strategy for batch-stochastic configuration network models and their application in component content prediction. EAAI, 150, 110461. (www)

  @article{ , abstract = {This study addresses the challenges of inefficiency and uncertain quality in incrementally adding hidden layer nodes to a stochastic configuration network (SCN). An optimized approach for batch-wise stochastic configuration network (BSCN) using an enhanced differential evolution (DE)algorithm is introduced. Initially, the inequality constraints of SCN is studied to analyze and establish the correlation between objective parameters and network residuals. Subsequently, to speed the network’s training velocity, a BSCN is utilized for developing the regression model. Combining the DE algorithm with regional contraction and greedy selection strategies. Specifically it leverages the robust global search prowess of the standard DE while mitigating its limitations in local search and ensuring global convergence. The formulation of this enhanced DE, termed regional contraction and greedy selection differential evolution (SGDE), is elaborated in detail, and an analysis and validation of its global convergence are conducted. Comparative experiments with the conventional DE underscore the superior optimization efficacy of SGDE. The applicability of SGDE enhanced BSCN (SGDE-BSCN) is corroborated through six real-world regression tasks. These tasks demonstrate that SGDE-BSCN not only excels in configuring hidden layer nodes but also exhibits enhanced error minimization and superior generalization capabilities with an equivalent number of hidden layer nodes. Additionally, a practical case study focused on predicting component content in rare earth extraction processes validates the effectiveness of the proposed method. The empirical results show that the application of SGDE-BSCN to artificial intelligence (AI) and artificial intelligence in the field of rare earth extraction shows high prediction accuracy and fast processing speed.}, archive = {J_EAAI}, author = {RongXiu Lu and XingRong Hu and Cong Pei and Hui Yang and WenHao Dai and JianYong Zhu}, doi = {10.1016/j.engappai.2025.110461}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110461}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimization strategy for batch-stochastic configuration network models and their application in component content prediction}, volume = {150}, year = {2025}, }
• (2025). Continual learning for behavior-based driver identification. EAAI, 150, 110459. (www)

  @article{ , abstract = {Behavior-based Driver Identification is an emerging technology that recognizes drivers based on their unique driving behaviors, offering important applications such as vehicle theft prevention and personalized driving experiences. However, most studies fail to account for the real-world challenges of deploying Deep Learning models within vehicles. These challenges include operating under limited computational resources, adapting to new drivers, and changes in driving behavior over time. The objective of this study is to evaluate if Continual Learning (CL) is well-suited to address these challenges, as it enables models to retain previously learned knowledge while continually adapting with minimal computational overhead and resource requirements. We tested several CL techniques across three scenarios of increasing complexity based on a well-known dataset for the Driver Identification problem. This work provides an important step forward in scalable driver identification solutions, demonstrating that CL approaches, such as Dark Experience Replay (DER), can obtain strong performance with only an 11% reduction in accuracy compared to the static scenario. Furthermore, to enhance the performance, we propose two new methods, Smooth Experience Replay (SmooER) and Smooth Dark Experience Replay (SmooDER), that leverage the temporal continuity of driver identity over time to enhance classification accuracy. Our novel method, SmooDER, achieves optimal results with only a 2% accuracy reduction compared to the 11% of the DER approach. In conclusion, this study proves the feasibility of CL approaches to address the challenges of Driver Identification in dynamic environments, making them suitable for deployment on cloud infrastructure or directly within vehicles.}, archive = {J_EAAI}, author = {Mattia Fanan and Davide Dalle Pezze and Emad Efatinasab and Ruggero Carli and Mirco Rampazzo and Gian Antonio Susto}, doi = {10.1016/j.engappai.2025.110459}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110459}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Continual learning for behavior-based driver identification}, volume = {150}, year = {2025}, }
• (2025). Self-supervised contrastive representation learning for classifying internet of things malware. EAAI, 150, 110299. (www)

  @article{ , abstract = {The rapid increase in malware prevalence poses a substantial security threat to Internet of Things (IoT) devices. Classifying IoT malware has emerged as a crucial area, essential for identifying attack patterns and developing effective defense strategies. Many methods for classifying malware utilize supervised learning. However, supervised learning in malware classification requires a considerable amount of labeled samples, which poses challenges and costs in acquiring and labeling malware samples. Furthermore, Some malware classification models struggle to fully extract features. This article proposes a self-supervised contrastive learning framework. Initially, the malware is converted to greyscale. The encoder is then pre-trained by self-supervised contrastive learning. The encoder with the new structure is able to extract features more comprehensively, while the projection header with attention is enabled to project features into the low-dimensional space more efficiently. Finally, the pre-trained encoder and classifier are fine-tuned to form a classification model using labeled samples. Experiments have shown that the proposed method has better accuracy regardless of the number of labeled samples. Experiments conducted using the publicly benchmarked datasets, Malware Image (Malimg) and the Microsoft Malware Classification Challenge (BIG2015), demonstrate that our framework outperforms state-of-the-art deep learning models and traditional methods in terms of accuracy, with achieved rates of 99.46% and 99.22%, respectively. Using only 5% of the labels from BIG2015, the proposed framework produces an impressive accuracy of 94.76%. Furthermore, it also outperforms baseline methods in identifying evolving malware, as indicated by its accuracy of 79% in a benchmarked dataset for trustworthy malware family classification (BenchMFC-G1P1P2).}, archive = {J_EAAI}, author = {Fangwei Wang and Yinhe Chen and Hongfeng Gao and Qingru Li and Changguang Wang}, doi = {10.1016/j.engappai.2025.110299}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110299}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-supervised contrastive representation learning for classifying internet of things malware}, volume = {150}, year = {2025}, }
• (2025). A hybrid rough aggregation approach for the selection of artificial intelligence-based industrial cleaning robots used in public spaces from the perspective of urban waste management. EAAI, 150, 109566. (www)

  @article{ , abstract = {Waste management is becoming increasingly complex and challenging, especially in megacities with large populations. Unlike the past, when urban waste was simply collected and disposed of, modern waste management requires careful planning and execution of collection, separation, recycling, and reuse processes. Effective management of this complex system now needs more than just human effort. Integrating artificial intelligence (AI)-based systems into waste management can enhance waste reduction, reuse, and recycling effectiveness and efficiency. Selecting suitable AI-based cleaning robots (AI-ICR) for crowded public spaces, such as stations, train stations, and airports, poses complex decision-making challenges. The primary challenge is the novelty of the technology, which leads to uncertainties in selecting AI-ICRs. To address this challenge, we have developed a decision-making approach based on rough Archimedean-Dombi partitioned aggregation. This approach, termed “rough Archimedean-Dombi partitioned aggregation,” combines the flexibility of Archimedean operators, the smoothness of Dombi operators, and the structured decomposition of Partitioned operators. This model is mainly chosen for its ability to handle the uncertainty and complexity inherent in multiple criteria decision-making (MCDM) processes. Leveraging rough numbers provides a robust framework for evaluating AI-ICRs under uncertain conditions. The main advantage of this model is its robustness, consistency, stability, and ability to handle complex uncertainties. We applied the proposed model to assess four AI-ICR alternatives identified through extensive research. We evaluated these alternatives using eighteen criteria established through comprehensive field studies. Based on the results, “Recycling cost (B12)” emerged as the most crucial criterion for selecting AI-ICRs. Additionally, the research identifies the SD45 manufactured by Peppermint Robotics Co. as the optimal AI-ICR candidate. Finally, the sensitivity and benchmark analyses to validate the proposed model confirm its robustness, consistency, and reliability.}, archive = {J_EAAI}, author = {Ömer Faruk Görçün and Abhijit Saha and Pydimarri Venkata Ravi Kumar and Bijoy Krishna Debnath}, doi = {10.1016/j.engappai.2024.109566}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {109566}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid rough aggregation approach for the selection of artificial intelligence-based industrial cleaning robots used in public spaces from the perspective of urban waste management}, volume = {150}, year = {2025}, }
• (2025). A study on sustainable system for managing municipal solid waste through a multi-criteria group decision-making technique. EAAI, 150, 109393. (www)

  @article{ , abstract = {Municipal solid waste management, a unique aspect of sustainable development, is a crucial social-ecological system that intersects with the economy, society, and environment. The introduction of volume-based waste fees in some developed countries has been a step towards promoting recycling and waste reduction. However, the sustainability of high recycling targets and the impact of public satisfaction on waste management efficiency are areas that demand further exploration. A review of the literature on municipal solid waste management and technology selection from various countries reveals that many studies need more precise justification and a resolution to the ambiguity in decision-making. Meanwhile, some researchers have developed the fuzzy multi-criteria decision-making technique in the context of waste management. However, significant performance criteria for ’5R’s (refuse, reduce, reuse, repurpose, recycle)’ waste management technology selection and cause-and-effect group criteria still need to be identified. This study strongly emphasizes the potential of the ’5R’s’ waste management system to revolutionize waste management practices. The ’5R’s’ waste management system uses a multi-criteria group decision-making technique using fuzzy-based artificial intelligence methods, employing the novel fuzzy technique for order of preference by similarity to the ideal solution. This study also proposes a new way to rank generalized interval type-2 trapezoidal fuzzy numbers and defuzzifies them to address the uncertainties that arise when using fuzzy linguistic terms to make decisions. Finally, a numerical example of the ’5R’s’ waste management problem is discussed with new ranking methods and compared with existing methods, underscoring the significant potential of the ’5R’s’ waste management system.}, archive = {J_EAAI}, author = {Marimuthu Dharmalingam and Daekook Kang}, doi = {10.1016/j.engappai.2024.109393}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {109393}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A study on sustainable system for managing municipal solid waste through a multi-criteria group decision-making technique}, volume = {150}, year = {2025}, }
• (2025). A graph convolutional network and gated recurrent unit-based surrogate for agent-based diffusion models. EAAI, 149, 110610. (www)

  @article{ , abstract = {This study addresses the challenge of high computational costs in agent-based diffusion models (ABMs), which are widely used for simulating complex diffusion processes but become prohibitively expensive in large-scale applications. To mitigate this issue, we introduce a Graph Convolutional Network (GCN) and Gated Recurrent Unit (GRU)-based Surrogate Network (G2SN) for ABMs. The GCN module captures the social network structure and seed set, while the GRU module models the diffusion time series. Computational complexity analysis demonstrates that G2SN significantly outperforms ABM simulations in efficiency. Experimental results confirm that G2SN accurately predicts ABM dynamics, reducing the mean absolute deviation (MAD) by 71.7 % on training sets and 77.7 % on test sets compared to traditional machine learning surrogate models. Case studies on new product diffusion further illustrate the effectiveness of the G2SN-based calibration approach, improving parameter search efficiency by 50.8 % and 37.2 % over alternative surrogate model-based methods. Additionally, these studies underscore the critical importance of social network and seed set in enhancing ABM prediction accuracy. This approach provides a more efficient and scalable tool for ABM calibration and new product diffusion forecasting, aiding managers in production, inventory, and marketing decisions.}, archive = {J_EAAI}, author = {Yu Xiao and Yuanyuan Zhou and Ziyi Wang}, doi = {10.1016/j.engappai.2025.110610}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110610}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A graph convolutional network and gated recurrent unit-based surrogate for agent-based diffusion models}, volume = {149}, year = {2025}, }
• (2025). Uncertainty-aware focal loss for object segmentation. EAAI, 149, 110599. (www)

  @article{ , abstract = {In the loss function of object segmentation models, misclassified pixels whose prediction are opposite to the ground truth and uncertain pixels whose predicted probability is close to 0.5 are more important for model training. Focusing on misclassified pixels can improve the segmentation accuracy of the model, and focusing on uncertain pixels can help the model to form better decision surfaces. However, existing methods fail to take both types of pixels into account simultaneously. To enhance the learning on these two types of important pixels, the Uncertainty-aware Focal Loss (UFL) is proposed based on the analysis of Uncertainty-aware Loss (UAL). Then, by leveraging the S-shaped property of the sigmoid function, a loss function is constructed that can simultaneously increase the loss and loss derivatives of misclassified and uncertain pixels. In order to solve the gradient vanishing problem of the sigmoid function on well-classified pixels, a regularization constraint term is defined, whose value is the square of predicted probability. Finally, the pixel loss value is dynamically adjusted at different stages of training according to the changes in the contributions of misclassified and uncertain pixels to the model training, which improves the targeted learning for misclassified and uncertain pixels. Experimental results on two different types of network structures and six datasets demonstrate that the proposed method can better segment the uncertain and misclassified pixels. Especially, on the DUT-O dataset, UFL improves mean Intersection over Union (mIoU ) by almost 2.7 % compared to UAL.}, archive = {J_EAAI}, author = {Lei Chen and Yang Wang and Jibin Yang and Yunfei Zheng and Tong Han and Bo Zhang and Tieyong Cao}, doi = {10.1016/j.engappai.2025.110599}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110599}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Uncertainty-aware focal loss for object segmentation}, volume = {149}, year = {2025}, }
• (2025). Self-supervised combustion state diagnosis using a noise-augmented generative adversarial network and flame image sequences. EAAI, 149, 110574. (www)

  @article{ , abstract = {Reliable diagnosis of combustion states, particularly distinguishing between stable and unstable flame conditions, is crucial for maintaining power generation efficiency and stability. However, accurate detection of unseen unstable combustion states remains challenging due to the complex dynamics of flames and the limited availability of unstable flame data. To address this challenge, this study proposes a self-supervised combustion state diagnosing method based on a noise-augmented generative adversarial network (NAGAN) and flame image sequences. The proposed method employs a convolutional autoencoder (CAE) and principal component analysis (PCA) to extract abstract flame features from image sequences. A novel multi-generator NAGAN architecture, comprising a long short-term memory (LSTM)-based generator and two Gaussian noise-augmented generators, is designed to synthesize diverse unstable flame feature sequences with temporal dynamics and identify the combustion state. A Gaussian abnormal flame feature generator (GAFG) leveraging Gaussian noise and binary masking is introduced to simulate a wide range of anomalies, enabling the discriminator to learn diverse representations of unstable combustion states. Experimental results on methane-air flames show that the proposed NAGAN achieves an accuracy of 0.978 and an F1 score of 0.986 on the flame stability diagnosis, with a recall rate of 0.975 for unseen unstable flames, outperforming most existing unsupervised machine learning and deep-learning based diagnostic methods. These results demonstrate the potential of the proposed method to improve combustion state monitoring, enhancing the reliability and efficiency of power generation systems.}, archive = {J_EAAI}, author = {Xiaojing Bai and Liwen Fei and Weiqi Liu and Hua Wu and Yong Yan and Weicheng Xu}, doi = {10.1016/j.engappai.2025.110574}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110574}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-supervised combustion state diagnosis using a noise-augmented generative adversarial network and flame image sequences}, volume = {149}, year = {2025}, }
• (2025). An extended multi-criteria group decision-making method based on preference ranking under Z-number environments. EAAI, 149, 110573. (www)

  @article{ , abstract = {Compared with traditional fuzzy numbers, using Z-numbers to illustrate fuzzy events offers two key advantages. It makes fuzzy events more intuitive for decision-makers, and the second component of Z-numbers acts as a measure of the reliability of the first component. While significant progress has been made on Z-numbers, from the theoretical and practical perspectives, some gaps remain. For example, scholars have seldom focused on the likelihood of Z-numbers, most existing decision-making methods with Z-numbers rarely consider the consensus-reaching processes, and little has been reported on the superior ordering methods in the Z-number environment. To overcome these limitations, first, the likelihood of Z-numbers is defined in combination with the preference ranking organization method for enrichment evaluations (PROMETHEE) type V preference function. Second, the ordering rules for PROMETHEE are discussed. Then, a procedure of the feedback-adjustment method is introduced to help the consensus level of group-alternative ranking reach the threshold. On these bases, an extended PROMETHEE multi-criteria group decision-making method with Z-numbers is proposed. Finally, to verify the feasibility and effectiveness of the proposed method, we examined an intelligent medical-diagnostic-system selection problem and conducted a comparison analysis. We applied the Z-number PROMETHEE approach to a challenging case study requiring a dual-data-driven application. Furthermore, the study suggests future directions for improving the proposed framework in other related contexts.}, archive = {J_EAAI}, author = {Jian Li and Yuanyuan Xiang and Honggang Peng and Jianqiang Wang}, doi = {10.1016/j.engappai.2025.110573}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110573}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An extended multi-criteria group decision-making method based on preference ranking under Z-number environments}, volume = {149}, year = {2025}, }
• (2025). Investigating the effectiveness of hybrid gradient boosting models and optimization algorithms for concrete strength prediction. EAAI, 149, 110568. (www)

  @article{ , abstract = {This study aims to evaluate and predict the compressive strength of concrete using 8 different machine learning (ML) models, including Extreme Gradient Boosting (XGBoost), Light Gradient Boosting machine (LightGBM), Gradient Boosting with Categorical features support (CatBoost), Gradient Boosting Regressor (GBR), Adaptive Boosting (AdaBoost), Decision Tree (DT), Random Forest (RF), and Support Vector Machine Regression (SVR). The study employed Bayesian optimisation process with two surrogate models (Gaussian Processes and Random Forest) and Random Search optimisation process to optimise the hyperparameters of these ML models. 1030 data samples were used to train the models and analyse the feature importance of each input variable using SHapley Additive exPlanations (SHAP). The results indicated that all 8 hybrid ML models performed well with R2 values larger than 0.80 and four models (XGBoost, CatBoost, GBR, and LightGBM) being the standout models, achieving R2 values of 0.94, 0.94, 0.92, and 0.92 on testing dataset, respectively. The four leading models (XGBoost, CatBoost, GBR, LightGBM) were applied to six sub-databases of concrete types, significantly enhancing accuracy with all models achieving R2 values over 0.98 on the testing dataset. The study also found that curing age, cement content, and amount of water were the most important variables affecting compressive strength while fly ash was the least important. By deploying the three best models to the cloud, it is now possible to make predictions using any web browser on any device.}, archive = {J_EAAI}, author = {Khuong Le Nguyen and Mahmoud Shakouri and Lanh Si Ho}, doi = {10.1016/j.engappai.2025.110568}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110568}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Investigating the effectiveness of hybrid gradient boosting models and optimization algorithms for concrete strength prediction}, volume = {149}, year = {2025}, }
• (2025). An agent-based emotional persuasion model driven by integrated trust assessment. EAAI, 149, 110567. (www)

  @article{ , abstract = {Recent research on automated negotiation has primarily focused on improving the artificial intelligence of agents and equipping them with more flexible internal mechanisms to facilitate high-quality negotiations. However, the study on the systematic modeling of human-like psychological and behavioral activities and their role in the negotiation process is still in its early stages. In light of this, this paper proposes an emotional persuasion model that takes into account the effect of negotiators' integrated trust assessments on negotiation. Firstly, the paper presents a negotiation agent with both cognitive and emotional functions, detailing its internal system and operating mechanism. Secondly, the integrated trust of a negotiator is obtained by evaluating multiple single trusts, and the mapping of the integrated trust to the negotiation round parameter is modeled. Integrated trust is also parameterized into the agent's cognitive processes. Finally, the paper introduces a new framework for the generation of emotional persuasive behavior to assist agents in making new proposals. A series of experiments were conducted, yielding the following results: Compared with the non-emotional model, the performance of negotiation rounds and utility differences improved by 7.97 % and 4.81 %, respectively. Furthermore, the trust-driven emotional persuasion model outperformed the several existing competing models by at least 31.1 % in utility difference and 81.0 % in negotiation rounds. Additionally, a case study of human-computer negotiation demonstrated that the agent designed using the proposed method has negotiating capabilities comparable to those of a real human, which further showcases the application effect of the artificial intelligence agent in practice.}, archive = {J_EAAI}, author = {Jinghua Wu and Ya Zhang and Ruiyang Cao and Yan Li}, doi = {10.1016/j.engappai.2025.110567}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110567}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An agent-based emotional persuasion model driven by integrated trust assessment}, volume = {149}, year = {2025}, }
• (2025). Strength prediction of recycled concrete using hybrid artificial intelligence models with gaussian noise addition. EAAI, 149, 110566. (www)

  @article{ , abstract = {This research uses machine learning and computational techniques to investigate methods for the compressive strength of recycled concrete prediction. Waste management and environmental sustainability are addressed using construction and mineral waste as aggregate. Previous attempts with traditional linear and polynomial models were insufficient for capturing complex data relationships, achieving R-squared (R 2 ) values of only 0.22. Therefore, advanced models were employed, including support vector regression (SVR), one-dimensional convolutional neural networks (1D-CNN), and a multi-model hybrid approach. Combining Elastic Networks, Random Forest Algorithm, and Light Gradient Boosting Decision (LGBM), the hybrid model demonstrated exceptional performance, achieving an R 2 value of 0.9072 - a 312 % improvement over traditional methods. Gaussian noise augmentation during training enhanced the model's generalization capabilities. Experimental data validation confirmed the hybrid model's predictive accuracy. The potential of computational methods is highlighted to optimize the use of recycled materials in concrete, promoting sustainable construction practices and laying the foundation for selecting recycled materials for future practical civil engineering applications. Future research should focus on expanding datasets and exploring additional data augmentation techniques to improve model accuracy further.}, archive = {J_EAAI}, author = {Yuzheng Geng and Yongcheng Ji and Dayang Wang and Hecheng Zhang and Zhizhu Lu and Aotian Xing and Mingjie Gao and Maoyang Chen}, doi = {10.1016/j.engappai.2025.110566}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110566}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Strength prediction of recycled concrete using hybrid artificial intelligence models with gaussian noise addition}, volume = {149}, year = {2025}, }
• (2025). Fast stereo conformer: Real-time stereo matching with enhanced feature fusion for autonomous driving. EAAI, 149, 110565. (www)

  @article{ , abstract = {Accurate stereo matching is critical for depth estimation in autonomous driving, yet remains challenging in weakly-textured regions where conventional convolutional neural networks (CNNs) lack global context and Transformers overlook local details. We propose Fast Stereo Conformer Net (FSCN), a real-time parallel network integrating CNNs and Transformers, enhanced by a Convolutional Coupling Transformer (CCT) for synergistic local-global feature fusion. To address data scarcity, we introduce random translation transformations (RTTs) during pretraining and fine-tuning. On the Karlsruhe Institute of Technology and Toyota Technological Institute at Chicago 2015 Dataset (KITTI 2015) benchmark, FSCN achieves a 3-pixel error rate of 1.79 %, representing a 7.25 % improvement over the baseline and a 5.29 % superior performance compared to state-of-the-art real-time methods, while processing frames efficiently in 57 milliseconds (ms). It also demonstrates outstanding generalization on other datasets, achieving improvements of 6.98 % on KITTI 2012, 9.72 % on Middlebury, and 9.30 % on Swiss Federal Institute of Technology in Zurich 3D Dataset (ETH3D) compared to the baseline. These results highlight the synergistic effectiveness of CNNs and Transformers in feature extraction and the significant potential of our approach for advancing stereo matching technology in autonomous driving applications.}, archive = {J_EAAI}, author = {Yunhua Lu and Xianzhong He and Qingwei Zhang and Ding Zhang}, doi = {10.1016/j.engappai.2025.110565}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110565}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast stereo conformer: Real-time stereo matching with enhanced feature fusion for autonomous driving}, volume = {149}, year = {2025}, }
• (2025). Speech signal-based accurate neurological disorders detection using convolutional neural network and recurrent neural network based deep network. EAAI, 149, 110558. (www)

  @article{ , abstract = {Neurological diseases often manifest in subtle alterations to the human voice due to damage in the sound-related regions of the brain. Leveraging advancements in artificial intelligence (AI) technologies, computers can discern minute variations in sound imperceptible to the human ear, enabling rapid and precise diagnostic support. This paper presents a novel approach to neurological disease classification utilizing voice recordings of individuals diagnosed with various neurological conditions alongside healthy controls. By employing AI techniques, particularly a hybrid deep network framework integrating Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), we aimed to classify one-sentence audio inputs of Multiple Sclerosis (MS) patients, healthy individuals, and other neurological diseases. In our dataset, we have compiled audio recordings from 95 healthy individuals, 99 individuals diagnosed with multiple sclerosis (MS), and 96 individuals with other neurological disorders. Of these, 20 % of the data was reserved for testing. Our proposed architecture achieved remarkable performance metrics in experimental evaluations, exhibiting 96.55 % accuracy, 98.25 % specificity, 96.49 % sensitivity, 96.97 % precision, and 96.56 % F1-Score. The results obtained are more successful compared to the methods of AlexNet from scratch, fine-tuned AlexNet, Long Short-Term Memory (LSTM) based CNN, and Gated Recurrent Unit (GRU) based CNN. The results of our study highlight the potential of this framework to be integrated into clinical diagnostic workflows, providing clinicians with an effective tool for early and precise detection of neurological diseases, ultimately improving patient outcomes through timely intervention and personalized treatment strategies.}, archive = {J_EAAI}, author = {Emel Soylu and Sema Gül and Kübra Aslan Koca and Muammer Türkoğlu and Murat Terzi and Abdulkadir Şengür}, doi = {10.1016/j.engappai.2025.110558}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110558}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Speech signal-based accurate neurological disorders detection using convolutional neural network and recurrent neural network based deep network}, volume = {149}, year = {2025}, }
• (2025). A novel multimodal deep learning framework for predicting residual strength of corroded rectangular hollow-section columns. EAAI, 149, 110554. (www)

  @article{ , abstract = {Corrosion, recognized as a thermodynamically spontaneous process, is one of the key issues affecting the health of rectangular hollow steel section columns under working conditions, and has attracted much attention in recent years. Traditional approaches, such as multi-layer perceptron, often rely solely on the degree of volume loss to predict residual strength, overlooking the spatial complexity of actual corrosion patterns. To address these limitations, this study presents a novel multimodal deep learning network for accurately predicting the residual strength of corroded hollow steel section columns with random, nonuniform corrosion distributions. Our approach integrates (i) image‐based corrosion distributions on four steel walls, and (ii) tabular geometric parameters, through five distinct data-fusion methods proposed in this work, three employing Late Fusion (via a novel multi‐head attention module) and two using Early Fusion (via pixel−level merging). The image information extraction core is built upon a lightweight convolutional neural network and a channel−spatial attention block, while the tabular extraction module leverages a revised multi-layer perceptron architecture. After Bayesian hyperparameter optimization, the best‐performing model achieves a coefficient of determination of 0.971 on the test set, surpassing conventional machine learning and other multimodal fusion techniques by 0.01–0.161. Further analysis shows that the reverse visualization technique highlights corrosion−critical regions that closely coincide with the experimentally validated failure zones. Consequently, the proposed framework not only predicts residual strength with high accuracy but also localizes vulnerable areas for targeted reinforcement. This methodology holds promise for large‐scale corrosion monitoring and structural health assessment of steel infrastructure.}, archive = {J_EAAI}, author = {Yu-Jia Zhang and Lei Zhang and Yu Zhou and Tian-Xiang Li and Reece Lincoln and Jing-Zhong Tong and Jia-Jia Shen}, doi = {10.1016/j.engappai.2025.110554}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110554}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel multimodal deep learning framework for predicting residual strength of corroded rectangular hollow-section columns}, volume = {149}, year = {2025}, }
• (2025). Comprehensive evaluation of deep reinforcement learning for permanent magnet synchronous motor current tracking and speed control applications. EAAI, 149, 110551. (www)

  @article{ , abstract = {Permanent Magnet Synchronous Motors (PMSMs) are indispensable in industrial applications, requiring precise control to ensure optimal performance. Traditional model-based methods, such as Proportional-Integral (PI) control and Model Predictive Control (MPC), face inherent limitations in robustness and adaptability under complex conditions. Deep Reinforcement Learning (DRL), as a model-free, data-driven approach, offers a transformative solution for PMSM control. This study proposes a DRL-based current control strategy and systematically evaluates the performance of three representative DRL algorithms: Deep Q-Network (DQN), Proximal Policy Optimization (PPO), and Advantage Actor-Critic (A2C) in PMSM control tasks. Key contributions include hyperparameter sensitivity analysis, transfer learning for improved training efficiency, and the application of DRL to multi-objective speed control under varying operational scenarios. Experimental results reveal the hyperparameter sensitivities of different DRL algorithms and provide theoretical insights. The findings demonstrate that transfer learning significantly improves DRL training efficiency and control performance. DRL outperforms traditional controllers in current and speed control, achieving superior dynamic response, tracking accuracy, and adaptability to complex conditions. This study offers new insights into the application of DRL in industrial PMSM control and serves as a reference for its further optimization and practical deployment.}, archive = {J_EAAI}, author = {Yiming Zhang and Jingxiang Li and Hao Zhou and Chin-Boon Chng and Chee-Kong Chui and Shengdun Zhao}, doi = {10.1016/j.engappai.2025.110551}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110551}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comprehensive evaluation of deep reinforcement learning for permanent magnet synchronous motor current tracking and speed control applications}, volume = {149}, year = {2025}, }
• (2025). An efficient boundary prediction method based on multi-fidelity gaussian classification process for class-imbalance. EAAI, 149, 110549. (www)

  @article{ , abstract = {Failure boundary prediction aims to explore the boundary between failure and safety limit states in a given problem. In recent years, the use of surrogate models instead of physical experiments to explore the failure boundary has become one of the promising methods to reduce costs. Among various surrogate models, multi-fidelity Gaussian process classification (MFGPC) can produce accurate predictions while reducing simulation expenses by using a small set of high-fidelity samples and a larger set of low-fidelity samples. However, the current MFGPC model primarily utilizes Markov chain Monte Carlo sampling to estimate the model hyperparameters. The modeling process involves a significant amount of calculation and slow convergence speed. Additionally, the practical application of MFGPC in the engineering field is limited, and the presence of a class imbalance in real data can lead to poor model training performance. Therefore, to improve the modeling efficiency of MFGPC, this paper proposes an improved MFGPC method for the imbalanced problem. The improved MFGPC model utilizes the Stein variational gradient descent algorithm to reduce the computational cost of the model. Additionally, it incorporates a synthetic sampling algorithm to effectively handle class-imbalance issues in real-world engineering cases. Two engineering examples are solved, including the failure boundary prediction of the zero Poisson ratio structure and the operating boundary of an axial flow compressor. It is demonstrated that the proposed method can accelerate the convergence speed of the model by tens of magnitudes while maintaining good prediction accuracy.}, archive = {J_EAAI}, author = {Jinlang Luo and Lingzhi Liu and Youwei He and Kuan Tan}, doi = {10.1016/j.engappai.2025.110549}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110549}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient boundary prediction method based on multi-fidelity gaussian classification process for class-imbalance}, volume = {149}, year = {2025}, }
• (2025). Flow field reconstruction and prediction of the two-dimensional cylinder flow using data-driven physics-informed neural network combined with long short-term memory. EAAI, 149, 110547. (www)

  @article{ , abstract = {Physics-informed neural network (PINN) has attracted significant attention for various fluid related problems in recent years. Although the basic principle allows the PINN to require less data to train a reliable model, data-driven PINN often suffers from the quantity and quality of the training data, without which will result in diminishing fitting capability in the edge regions and extrapolation beyond the dataset. To address this issue, a new coupled model called LSTM-PINN (LP) is proposed, which combines the advantages of Long Short-Term Memory (LSTM) in handling long-term dependencies for time-sequenced data with the existing PINN framework. By incorporating the time-sequenced predictions at different spatial points generated by the LSTM into the training set of the PINN, the edge errors that refer to errors at boundary of space or time are reduced and the time-sequenced prediction capability of flow field is enhanced. The comparative study is conducted on the velocities along two directions predicted by the coupled model with those obtained from the benchmark PINN model, while the numerical solutions in the case of 2D (two-dimensional) flow around cylinder described by the Navier-Stokes (NS) equations are selected as training dataset. The results demonstrate that the proposed LP model improves the accuracy of prediction compared to the conventional PINN model, showing great potential for flow field reconstruction and time series prediction.}, archive = {J_EAAI}, author = {Yehao Dou and Xun Han and Pengzhi Lin}, doi = {10.1016/j.engappai.2025.110547}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110547}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Flow field reconstruction and prediction of the two-dimensional cylinder flow using data-driven physics-informed neural network combined with long short-term memory}, volume = {149}, year = {2025}, }
• (2025). A large-scale mobile application knowledge graph for the research of cybersecurity: Construction and application. EAAI, 149, 110543. (www)

  @article{ , abstract = {Large-scale datasets for mobile applications (terms as “apps”) have been developed and become important assets for malware identification and other tasks of cybersecurity. However, existing datasets focus on extending the scale of apps, while ignoring the relevance among apps. On the other hand, several works try to integrate different metadata of apps to discover the relevance among apps, but most of them pay little attention to the roles such as normal users, developers, cybersecurity analysts, and they do not take full advantage of these metadata so that the fine-grained correlations among apps are difficult to be captured. To fill these gaps, we present a mobile application knowledge graph, which collects millions of apps’ information from various resources, including application markets, encyclopedias and news. Precisely, a lightweight ontology is designed for our knowledge graph. It defines a unified semantic schema of collected apps so that more linkages of these apps can be shared with each other. Moreover, we employ several promising algorithms of information extraction and knowledge alignment, and evaluate their performances during the process of construction. To detect more relevance with respect to sensitive apps, we propose a hybrid embedding-based method, in which the vector representations of apps are iteratively encoded with knowledge graph embedding methods and network embedding models. Experimental results show that our hybrid method can obtain better performances than several existing models for the relevance detection of sensitive apps. Finally, we list three use cases of mobile application knowledge graph for cybersecurity and discuss their limitations that would be improved in future works.}, archive = {J_EAAI}, author = {Weizhuo Li and Heng Zhou and Yiming Tan and Weiqi Luo and Qiu Ji and Yuyang Bian}, doi = {10.1016/j.engappai.2025.110543}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110543}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A large-scale mobile application knowledge graph for the research of cybersecurity: Construction and application}, volume = {149}, year = {2025}, }
• (2025). Two-layer knowledge graph transformer network-based question and answer explainable recommendation. EAAI, 149, 110542. (www)

  @article{ , abstract = {The question and answer (Q&A) recommendation in community question answering (CQA) helps users quickly and accurately find the desired Q&A. However, existing studies face the problems of sparse interaction data, cold starts, and a lack of explanations. This paper proposes a novel Q&A explainable recommendation approach based on a two-layer knowledge graph transformer network. It alleviates the sparse data and cold start problem by the novel two-layer knowledge graph. First, a two-layer knowledge graph in CQA is constructed. The interaction layer helps to enrich the associations between users and questions and answers (Q&As). The semantic layer provides semantic associations and reflects contextual domain knowledge. Second, a critical meta-path recognition module is constructed to learn the critical meta-paths between users and documents from the interaction layer. Then, a user and Q&A embedding method based on a two-layer knowledge graph is proposed to enhance the user and Q&A representations. Finally, a recommendation and explanation layer is established to obtain personalized Q&A recommendation results and corresponding explanations. Compared with the baselines, the proposed method shows superior performance. It achieves average improvements of 21.28%, 28.41% and 27.18% in precision, recall and F1-measure, respectively, in the top- K Q&A recommendation separately. It improves the area under the curve and F1-measure of the click-through rate prediction recommendation by 11.32% and 23.06%, respectively.}, archive = {J_EAAI}, author = {Ying Li and Ming Li and Jin Ding and Yixue Bai}, doi = {10.1016/j.engappai.2025.110542}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110542}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Two-layer knowledge graph transformer network-based question and answer explainable recommendation}, volume = {149}, year = {2025}, }
• (2025). Metal oxide semiconductor based electronic nose data pre-processing, review. EAAI, 149, 110540. (www)

  @article{ , abstract = {Metal Oxide Semiconductor (MOS) based Electronic Nose (E-nose) mimics Human Olfaction mechanisms in Odor Detection by creating a Unique Odor-Print that is processed by Pattern Recognition (PARC) algorithms. Odor Metal Oxide Semiconductor sensors do not behave as independent sensors. Each gas sensor in E-nose responds more selectively to a certain group of Volatile Organic Compounds (VOCs) but also shows a broad, overlapping response and sensitivity to the other Gas Compounds. Like in Natural Olfaction, a key role is played by Data Analysis. In this comprehensive literature review, the current state of different design stages in Metal Oxide Semiconductor (MOS)-based Electronic Nose Raw Data Pre-processing to condition input data prior to array processing and pattern recognition were thoroughly discussed. In depth technical application of various data pre-processing techniques to both static and dynamic sensor responses, emphasizing their importance in improving the sensors' performance and pattern recognition accuracy were provided. A use case has been attempted, discussing data samples, feature extraction processes, and dimensionality reduction. A sample MATLAB code was provided to show cluster formations. Transient feature extraction improved classification accuracy by up to 30 %, while piecemeal signal feature extraction proved highly effective. Multivariate analysis and chemometric methods reduced data dimensionality by approximately 50–80 %, for creating Parsimonious Odor Classification Models.}, archive = {J_EAAI}, author = {Meisam Vajdi}, doi = {10.1016/j.engappai.2025.110540}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110540}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Metal oxide semiconductor based electronic nose data pre-processing, review}, volume = {149}, year = {2025}, }
• (2025). Tube-LaneNet: Predict each three-dimensional lane as a completed structure via geometric priors. EAAI, 149, 110539. (www)

  @article{ , abstract = {Monocular three-dimensional lane detection is a critical task for intelligent vehicles. However, most current methods, which mainly extend the two-dimensional paradigms, regard lanes as separated points set and constrain loss through the orthogonal projection on the two-dimensional plane. In this work, a novel deep learning framework is proposed to detect each lane as a continuous completed three-dimensional spatial structure. Concretely, three-dimensional lane anchors are implemented to extract proposal features through geometric priors to guarantee the continuous linear spatial structure. To enhance the feature of proposals, a relation-aware mechanism is further introduced to extract the spatial relationship between three-dimensional lanes. In particular, a novel tube-like intersection over union (TubeIOU) is proposed, which extends each three-dimensional lane to the tube-like structure as a completed unified entity in the three-dimensional space. Experiments on different datasets demonstrate the state-of-art performance of the proposed framework, especially achieves the fastest efficiency with 69 frames per second. The code will be made publicly available.}, archive = {J_EAAI}, author = {Genghua Kou and Shihao Wang and Ying Li}, doi = {10.1016/j.engappai.2025.110539}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110539}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Tube-LaneNet: Predict each three-dimensional lane as a completed structure via geometric priors}, volume = {149}, year = {2025}, }
• (2025). Q-learning based estimation of distribution algorithm for scheduling distributed heterogeneous flexible flow-shop with mixed buffering limitation. EAAI, 149, 110537. (www)

  @article{ , abstract = {There are growing interests in the distributed shop scheduling research owing to the diversification of market demand. However, most prevailing studies disregard the synergistic influence of mixed buffering and due window on production efficiency. To reduce cost loss caused by delay in due window, this paper studies a distributed heterogeneous flexible flow-shop scheduling with mixed buffering limitation, i.e., finite buffers and no-wait requirements. The motivation of this work is to fill in void and offer practical insights for exploring how to intelligently implement, optimize and deploy a distributed production system. A mathematical model is established, aiming to minimize total weighted earliness and tardiness cost. An innovative Q-learning based estimation of distribution algorithm (QLEDA) is well-designed to address this problem. The QLEDA proposes well-tailored three-stage dynamic decoding and opposition-based learning to decode and promote the job sequence group. To balance global and local searchability of QLEDA, we introduce problem-specific Q-learning and Chebyshev chaotic mapping. To build a probability model of self-adaptation and self-selection, the job sequence group implements discrete actions by interacting with distributed environment and state space through Q-learning. Numerous experiments demonstrate that the QLEDA can generate more satisfactory results over other three well-performing rivals. The finding corroborates the applicability and effectiveness of presented QLEDA in solving the considered problem.}, archive = {J_EAAI}, author = {Hua Xuan and Qian-Qian Zheng and Lin Lv and Bing Li}, doi = {10.1016/j.engappai.2025.110537}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110537}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Q-learning based estimation of distribution algorithm for scheduling distributed heterogeneous flexible flow-shop with mixed buffering limitation}, volume = {149}, year = {2025}, }
• (2025). State of the art: Application of machine learning in ground motion modeling. EAAI, 149, 110534. (www)

  @article{ , abstract = {This paper presents a detailed review of the application of machine learning (ML) techniques in Ground Motion Models (GMMs), which are vital tools for predicting seismic responses during earthquakes. Traditional GMMs, or Ground Motion Prediction Equations (GMPEs), rely on statistical regression methods that use predefined functional forms to estimate ground motion intensity measures (IMs), such as peak ground acceleration (PGA) and pseudo-spectral acceleration (PSA). While effective, these models often struggle with limitations, including handling nonlinear relationships and managing large datasets as earthquake catalogs grow. In response to these challenges, researchers have turned to ML methods, which offer the flexibility to model complex patterns without requiring predefined forms. This article explores various ML techniques, such as artificial neural networks (ANNs), support vector regression (SVR), random forest (RF), and gradient boosting (GB), and their application in GMMs. The paper also discusses the advantages of ML, such as improved accuracy and the ability to process extensive data, alongside challenges like interpretability issues and the risk of overfitting. Several case studies are provided to illustrate the practical benefits of ML in enhancing GMMs, particularly in reducing residuals and variability, ultimately contributing to more accurate seismic hazard assessments.}, archive = {J_EAAI}, author = {Najme Alidadi and Shahram Pezeshk}, doi = {10.1016/j.engappai.2025.110534}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110534}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {State of the art: Application of machine learning in ground motion modeling}, volume = {149}, year = {2025}, }
• (2025). Enhancing accuracy of one-dimensional characteristic predictions for axial compressors using deep learning. EAAI, 149, 110533. (www)

  @article{ , abstract = {To enhance the accuracy of one-dimensional (1D) characteristic predictions for multistage axial flow compressors and improve design efficiency, this paper presents a deviation angle prediction model based on deep learning, utilizing data generated from Computational Fluid Dynamics (CFD) simulations. This model was integrated into a compressor characteristic calculation program. A CFD simulation dataset was created with National Advisory Committee for Aeronautics 65-series (NACA65) blade profiles, parameterized through a Latin hypercube design. After solving blade cascade flow fields, deviation angles under various conditions were obtained, establishing a mapping between design variables and deviation angles using deep learning. Test results showed a correlation coefficient of 0.9978 and a mean absolute error of 0.0785°. The surrogate model was embedded into the axial flow compressor 1D calculation program (HARIKA), replacing the original deviation angle model. Performance calculations on transonic two-stage and high-subsonic eight-stage compressors at different speeds demonstrated that the updated HARIKA program provided predictions closer to experimental values at high speeds, though slightly overestimated at lower speeds. These results confirm the model’s accuracy and practicality, suggesting that the improved HARIKA algorithm has potential for engineering applications in predicting the aerodynamic performance of multistage axial flow compressors.}, archive = {J_EAAI}, author = {Yulin Ma and Zhou Du and Quanyong Xu}, doi = {10.1016/j.engappai.2025.110533}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110533}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing accuracy of one-dimensional characteristic predictions for axial compressors using deep learning}, volume = {149}, year = {2025}, }
• (2025). A novel adaptive gating neurons model with physical features weighted for bearing fault diagnosis under strong noise. EAAI, 149, 110532. (www)

  @article{ , abstract = {Deep Learning (DL) has made significant progress in the field of intelligent diagnosis. However, existing intelligent diagnosis models encounter challenges, including inadequate fault feature extraction capability, poor generalization performance, and unclear diagnostic mechanisms, especially under strong noise interference. To address these issues, we first propose a novel adaptive gating neuron with physical features weighted, which can adaptively adjust the combination of quadratic convolutions based on the physical features of the signals. On this basis, we establish a mathematical connection between the physical features-weighted adaptive gating neuron and learnable weighted autocorrelation, and theoretically demonstrate its superior feature extraction capability, which assigns higher weights to periodic features in bearing time-domain signals. Secondly, an adaptive gating neurons convolutional network (AGNCN) with physical features weighted was proposed, incorporating a residual structure. By decomposing the physical features-weighted adaptive gating neuron, we reveal an embedded attention mechanism, which enhances the intrinsic interpretability of the AGNCN model for diagnosing bearing faults in the presence of noise interference. Experiments conducted on both public datasets and our proprietary dataset demonstrate that our proposed method can efficiently and reliably diagnose bearing faults under strong noise interference.}, archive = {J_EAAI}, author = {Panpan Guo and Weiguo Huang and Ning Jia and Chuancang Ding and Yifan Huangfu and Xingxing Jiang and Juanjuan Shi}, doi = {10.1016/j.engappai.2025.110532}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110532}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel adaptive gating neurons model with physical features weighted for bearing fault diagnosis under strong noise}, volume = {149}, year = {2025}, }
• (2025). Loss of coolant accident monitoring and pipe break diagnosis in pressurized water reactors using bayesian-optimized long short-term memory models. EAAI, 149, 110531. (www)

  @article{ , abstract = {While safety systems have been established for nuclear power plant accidents, the risk of human error remains ever present, particularly in high-pressure situations. This work utilizes the long short-term memory model (LSTM) as surrogate for monitoring loss of coolant accidents (LOCA) and diagnosing pipe breaks, aiding operators in assessing LOCAs more accurately. The models were trained using simulated accident scenarios covering the entire range of hot leg and cold leg pipe break extents. To ensure generalizability, Bayesian optimization was implemented as an automatic hyperparameter tuning method to optimize each model. In total, 15 LSTM forecasters with varying lookbacks and epochs and 20 LSTM predictors with varying trimmed sizes were developed. The forecasters showed excellent results in the water level and hot leg temperature forecasting tasks with average root mean squared error (RMSE) values of 4% and 4 °C, respectively. The overall best forecasting model was determined to have a lookback of 7 and 100 epochs, and trends relating accuracy to the epochs and lookbacks were also identified. The study also showed that LSTM predictors can provide early pipe break diagnosis accurately, using only at least 4% of the accident time series data from the onset. These predictors were able to predict the extent of pipe break with RMSE ranging from 0.57% to 1.77% and the location of pipe break with classification accuracy from 94% to 99%. Lastly, the robustness of the pipe break extent predictions were verified within expected noise levels. This study has further demonstrated the potential of artificial intelligence methods in reinforcing nuclear safety and effective accident diagnostics.}, archive = {J_EAAI}, author = {Johndel Obra and Shuichiro Miwa}, doi = {10.1016/j.engappai.2025.110531}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110531}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Loss of coolant accident monitoring and pipe break diagnosis in pressurized water reactors using bayesian-optimized long short-term memory models}, volume = {149}, year = {2025}, }
• (2025). Skyline recency–frequency–monetary pattern mining based on different constraint degrees. EAAI, 149, 110529. (www)

  @article{ , abstract = {Knowledge discovery plays a pivotal role in the field of Artificial Intelligence as it involves extracting valuable and previously unknown patterns (information, insights) from vast datasets. Skyline pattern mining emerges to deal with multiple-objective knowledge discovery in a threshold-free manner. Existing Skyline pattern mining methods mainly have two limitations: they only settle the problem with the restriction in dimension two, which is not applicable in dimension three or more; they focus on frequency and utility (monetary) without considering the temporal information. In this paper, we aim to overcome those limitations by proposing the problem model for finding skyline patterns considering three dimensions: recency, frequency, and monetary. We also define degrees of constraints to formulate strict/weak skyline patterns and present algorithms to mine them. Several pruning strategies are designed to reduce the search space. The max monetary score matrix structure and corresponding sparse storage are designed to facilitate the implementation of the pruning strategies. The skyline pattern storage matrix helps to add and delete patterns in the mining process efficiently. Experimental results show that the proposed algorithms discover more patterns without losing to the baseline. The runtime difference is usually within 10s without a clear winner. The proposed method wins tens to hundreds of megabytes of memory in most cases.}, archive = {J_EAAI}, author = {Xiaojie Zhang and Guoting Chen and Linqi Song and Wensheng Gan}, doi = {10.1016/j.engappai.2025.110529}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110529}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Skyline recency–frequency–monetary pattern mining based on different constraint degrees}, volume = {149}, year = {2025}, }
• (2025). Identification of martian minerals based on multiscale spatial-spectral fusion network. EAAI, 149, 110527. (www)

  @article{ , abstract = {Accurate identification of Martian surface minerals is crucial for analyzing the planet's geological environment and resource potential. Deep learning offers a powerful solution for automating Martian mineral identification. However, existing methods suffer from recognition inaccuracies due to the insufficient representation of the multiscale spectral and spatial features. In this article, a multiscale spatial-spectral fusion network (MSSFNet) is proposed to identify Martian minerals from hyperspectral images. In MSSFNet, the spectral multiscale feature extraction (Spe-MFE) module is proposed to extract multiscale spectral features from adjacent mineral bands in the spectral dimension. The spatial multiscale feature extraction (Spa-MFE) module extracts the spatial correlations of minerals from low-frequency and high-frequency features and retains more spatial information. At the end of the model, the spatial-spectral feature adaptive fusion (SSFAF) module utilizes attention-based feature weighting to fuse spectral and spatial features, improving its representation ability. Experimental results demonstrate that the proposed method achieved 99.51% accuracy on a constructed Martian mineral identification dataset, improving the precision of deep learning models in Martian surface mineral recognition. On the benchmark hyperspectral datasets of Indian Pines and Pavia University, MSSFNet achieved overall accuracies of 98.87% and 99.42%, respectively, outperforming state-of-the-art methods and validating its effectiveness and superiority.}, archive = {J_EAAI}, author = {Kai Wang and Xubing Zhang and Xianmin Wang and Zhouyuan Qian}, doi = {10.1016/j.engappai.2025.110527}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110527}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identification of martian minerals based on multiscale spatial-spectral fusion network}, volume = {149}, year = {2025}, }
• (2025). Attention based spatial-temporal multi-graph ordinary differential equation network for traffic flow prediction. EAAI, 149, 110526. (www)

  @article{ , abstract = {Traffic flow prediction plays an important role in the intelligent transportation systems. Most of the current state-of-the-art traffic flow prediction models extract the spatial and temporal features of traffic flow data through graph neural network (GNN) and temporal extraction module. However, the semantic relevance of road network nodes and historical data relevance are ignored in those models. Moreover, there exists an over-smoothing problem as the number of GNN layers increases, and it is impossible to capture spatial relevance of long-range nodes through modeling. To address this challenge, we propose an attention based spatial-temporal multi-graph ordinary differential equation network (ASTMGODE). Specifically, ASTMGODE mainly consists of three independent components, to jointly model the spatial-temporal correlations and semantic correlations with various features in the traffic flow. The three spatial attributes are distance matrix, semantical matrix and historical data correlation matrix, respectively, based on the distance between road nodes, functional similarities and historical data. Each component can effectively capture the spatial-temporal correlations in the traffic data via the spatial-temporal attention mechanism. Subsequently, spatial-temporal features are characterized by temporal and spatial extraction modules, in which the temporal extraction module is composed of temporal attention mechanism and temporal convolution (TCN), while the spatial extraction module comprises tensor-based ordinary differential equation (ODE) and graph convolution network (GCN). During experiments on six real-world traffic datasets, the ASTMGODE model reduced root mean square error (RMSE) by approximately 7.65% compared to the state-of-the-art known models.}, archive = {J_EAAI}, author = {Ying-Ting Chen and Cheng Li and Shuang Li}, doi = {10.1016/j.engappai.2025.110526}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110526}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Attention based spatial-temporal multi-graph ordinary differential equation network for traffic flow prediction}, volume = {149}, year = {2025}, }
• (2025). Django-based framework database for leakage detection using machine learning for water distribution networks. EAAI, 149, 110525. (www)

  @article{ , abstract = {Leakage in water supply pipe networks is a critical issue, with traditional detection methods being inefficient and error-prone. Acoustic-based leak detection often lacks standardized databases, limiting its effectiveness. This study proposes an integrated system using MySQL, Python, and Django for managing and analyzing acoustic leakage data. The system incorporates Variable Modal Decomposition (VMD), Wavelet Threshold Noise Reduction, Feature Extraction, and Support Vector Machine (SVM) for accurate leak detection. Experimentation on 413 labeled acoustic samples achieved classification accuracies of 96.1% (training set) and 97.4% (test set). This approach enhances detection precision and offers a scalable solution for real-time monitoring, with significant practical implications for improving water distribution system management and decision-making.}, archive = {J_EAAI}, author = {Yiwei Xie and Mengze Gao and Fan Luo and Ao Zhou and Yunfeng Yang and Jian Hu and Wei Jiang and Yuanyao Ye}, doi = {10.1016/j.engappai.2025.110525}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110525}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Django-based framework database for leakage detection using machine learning for water distribution networks}, volume = {149}, year = {2025}, }
• (2025). Maneuvering target interception via deep reinforcement learning guidance using only line-of-sight rate measurement. EAAI, 149, 110523. (www)

  @article{ , abstract = {Intercepting a maneuvering target based solely on line-of-sight (LOS) rate measurement constitutes a classic partially observable Markov decision process (POMDP), posing significant challenges in formulating the interceptor's guidance law. To improve the interception probability, reduce energy consumption, and enhance the robustness of the guidance law for intercepting maneuvering targets, a deep reinforcement learning (DRL) compensated guidance law based on the gated recurrent unit (GRU) neural network is proposed. Firstly, based on the collision triangle, a three-dimensional uncertain confrontation model of maneuvering target and interceptor is established, and then the compensation of proportional guidance law is designed as the DRL action, which effectively improves the learning efficiency of the agent. Finally, a POMDP model for interception is established, and a process reward model with energy consumption and a soft terminal reward model with a “transition section" is proposed. GRU is used to fully mine the characteristics of the LOS rate time series, and a high-performance interception policy is explored based on the proximal policy optimization algorithm. Monte Carlo simulation results reveal that the novel guidance law achieves an average miss distance value of 0.091 m (m) and energy consumption of 192.26 m per second ( m ⋅ s − 1 ), outperforming traditional guidance laws with improved interception probability and lower energy consumption.}, archive = {J_EAAI}, author = {Leliang Ren and Yong Xian and Zhenyu Liu and Shaopeng Li and Daqiao Zhang and Weilin Guo}, doi = {10.1016/j.engappai.2025.110523}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110523}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Maneuvering target interception via deep reinforcement learning guidance using only line-of-sight rate measurement}, volume = {149}, year = {2025}, }
• (2025). What is behind the curtain? increasing transparency in reinforcement learning with human preferences and explanations. EAAI, 149, 110520. (www)

  @article{ , abstract = {In this work, we investigate whether the transparency of a robot’s behaviour is improved when human preferences on the actions the robot performs are taken into account during the learning process. For this purpose, a shielding mechanism called Preference Shielding is proposed and included in a reinforcement learning algorithm to account for human preferences. We also use the shielding to decide when to provide explanations of the robot’s actions. We carried out a within-subjects study involving 26 participants to evaluate the robot’s transparency. Results indicate that considering human preferences during learning improves legibility compared with providing only explanations. In addition, combining human preferences and explanations further amplifies transparency. Results also confirm that increased transparency leads to an increase in people’s perception of the robot’s safety, comfort, and reliability. These findings show the importance of transparency during learning and suggest a paradigm for robotic applications when a robot has to learn a task in the presence of or in collaboration with a human.}, archive = {J_EAAI}, author = {Georgios Angelopoulos and Luigi Mangiacapra and Alessandra Rossi and Claudia Di Napoli and Silvia Rossi}, doi = {10.1016/j.engappai.2025.110520}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110520}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {What is behind the curtain? increasing transparency in reinforcement learning with human preferences and explanations}, volume = {149}, year = {2025}, }
• (2025). Enhancing histopathological image analysis: An explainable vision transformer approach with comprehensive interpretation methods and evaluation of explanation quality. EAAI, 149, 110519. (www)

  @article{ , abstract = {Deep learning models are increasingly reshaping medical imaging, with growing attention on ensuring transparency and trust in their decision-making processes. This study presents the Explainable Vision Transformer (XViT), a model specifically designed for histopathological image analysis. By incorporating advanced interpretability techniques, the XViT model addresses three core aspects: feature learning and classification, generating explainable outputs, and qualitatively evaluating these explanations. Three novel interpretability methods are introduced: attention-based, model-agnostic, and gradient-based, offering diverse perspectives on model behavior. The model's performance and generalizability were rigorously evaluated on two histopathological datasets: lung colon 25000 (LCS25000) with 96.2% accuracy across three classes and Kangbuk Samsung Hospital (KBSMC) with 88.6% accuracy across four classes. XViT provides actionable insights by highlighting diagnostically relevant regions in input images, significantly enhancing clinical trust and decision-making. The evaluation of its explainability methods through metrics like sensitivity, faithfulness, and complexity demonstrated that layer-wise relevance propagation for transformers outperforms standard techniques like local interpretable model-agnostic explanations (LIME) and attention visualization. This robust performance underscores the XViT model's potential to bridge the gap between AI accuracy and interpretability in medical imaging. Our findings emphasize the need for well-defined evaluation criteria when comparing interpretability methods and highlight the model's potential for integration into clinical workflows. This work represents a step forward in creating reliable and interpretable AI solutions, ensuring that the benefits of advanced deep learning models extend seamlessly into practical healthcare settings.}, archive = {J_EAAI}, author = {Aqib Nazir Mir and Danish Raza Rizvi and Md Rizwan Ahmad}, doi = {10.1016/j.engappai.2025.110519}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110519}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing histopathological image analysis: An explainable vision transformer approach with comprehensive interpretation methods and evaluation of explanation quality}, volume = {149}, year = {2025}, }
• (2025). Automated processing of eXplainable artificial intelligence outputs in deep learning models for fault diagnostics of large infrastructures. EAAI, 149, 110518. (www)

  @article{ , abstract = {Deep Learning (DL) models processing images to recognize the health state of large infrastructure components can exhibit biases and rely on non-causal shortcuts. eXplainable Artificial Intelligence (XAI) can address these issues but manually analyzing explanations generated by XAI techniques is time-consuming and prone to errors. This work proposes a novel framework that combines post-hoc explanations with semi-supervised learning to automatically identify anomalous explanations that deviate from those of correctly classified images and may therefore indicate model abnormal behaviors. This significantly reduces the workload for maintenance decision-makers, who only need to manually reclassify images flagged as having anomalous explanations. The proposed framework is applied to drone-collected images of insulator shells for power grid infrastructure monitoring, considering two different Convolutional Neural Networks (CNNs), GradCAM explanations and Deep Semi-Supervised Anomaly Detection. The average classification accuracy on two faulty classes is improved by 8 % and maintenance operators are required to manually reclassify only 15 % of the images. We compare the proposed framework with a state-of-the-art approach based on the faithfulness metric: the experimental results obtained demonstrate that the proposed framework consistently achieves F 1 scores larger than those of the faithfulness-based approach. Additionally, the proposed framework successfully identifies correct classifications that result from non-causal shortcuts, such as the presence of ID tags printed on insulator shells.}, archive = {J_EAAI}, author = {G. Floreale and P. Baraldi and E. Zio and O. Fink}, doi = {10.1016/j.engappai.2025.110518}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110518}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated processing of eXplainable artificial intelligence outputs in deep learning models for fault diagnostics of large infrastructures}, volume = {149}, year = {2025}, }
• (2025). Assessment of hybrid kernel function in extreme support vector regression model for streamflow time series forecasting based on a bayesian estimator decomposition algorithm. EAAI, 149, 110514. (www)

  @article{ , abstract = {Diverse decomposition algorithms have been widely employed to streamflow time series forecasting. Their applications, however, are hindered by the plausible high accuracy in the overall decomposition-based framework. This paper firstly introduces a novel decomposition algorithm named Bayesian estimator of abrupt change, seasonality and trend (BEAST) into streamflow forecasting to alleviate the boundary effect. Practical samples are generated under the modified two-stage decomposition prediction (TSDP) framework. A hybrid kernel function, which benefits from two different standalone ones, is designed for kernel extreme support vector regression and the HKESVR model is trained on the samples using 10-fold cross-validation strategy. Comparative experiments are conducted on three monthly streamflow series from basins with diverse hydroclimatic conditions. The results in different lead times (1-, 3-, and 5-month-ahead) show that the BEAST algorithm imposes an average improvement of 5.14% and 12.25% for the root-mean-square error and Nash-Sutcliffe efficiency coefficient respectively on the standalone models and shares a comprehensive similar performance on the mean absolute percentage error. And the nonparametric test results reveal that the BEAST method shows a significant improvement on the comprehensive performance compared with a conventional decomposition method. By contrast, the differences between machine learning models are much smaller. The hybrid kernel function works well in some specific cases in which the standalone kernel function fails. The hybrid BEAST-HKESVR is reliable enough to rank the second place among the fifteen tested models. Finally, the effects of hyperparameters in the BEAST algorithm are discussed and relevant suggestions on them are provided.}, archive = {J_EAAI}, author = {Peng Shi and Lei Xu and Simin Qu and Hongshi Wu and Qiongfang Li and Yiqun Sun and Xiaoqiang Yang and Wei Gao}, doi = {10.1016/j.engappai.2025.110514}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110514}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Assessment of hybrid kernel function in extreme support vector regression model for streamflow time series forecasting based on a bayesian estimator decomposition algorithm}, volume = {149}, year = {2025}, }
• (2025). Dual-stage manifold preserving mixed supervised learning for bogie fault diagnosis under variable conditions. EAAI, 149, 110512. (www)

  @article{ , abstract = {Bogie fault diagnosis for bogie is crucial to the safety of rail systems. However, since bogies work under normal states most of the time, the sporadic faulty samples are often submerged in massive normal samples, which are difficult to be distinguished and labeled. Therefore, the labeled training data are always insufficient or even lack of some certain fault states (novel faults), which brings great challenges to fault diagnosis, especially under variable working conditions. Therefore, this paper proposes a new framework named dual-stage manifold preserving mixed supervised learning (d-MMSL) to simultaneously absorb from labeled and unlabeled data effectively. Firstly, manifold similarity (MSLP) is presented to cluster unlabeled samples according to one-off calculation of the manifold similarity. In MSLP, the Best-versus-Second-Best differences and uncertain values are utilized to measure manifold distance and maintain the inherent structure of data. Secondly, Local manifold regularization - broad learning system (LMR-BLS) is presented to o deal with the problem of linear and nonlinear function transformation using simple incremental structure, which could further separate fuzzy sets from MSLP and distinguish the operation conditions of known states accurately. The proposed framework has been verified by a classical dataset and actual vibration data collected from bogies, which achieves a F1-score of 0.99. It is proven that this framework outperforms traditional methods in accuracy and efficiency.}, archive = {J_EAAI}, author = {Ning Wang and Limin Jia and Yong Qin and Dechen Yao and Jianwei Yang and Zhipeng Wang}, doi = {10.1016/j.engappai.2025.110512}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110512}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-stage manifold preserving mixed supervised learning for bogie fault diagnosis under variable conditions}, volume = {149}, year = {2025}, }
• (2025). Weighted class-aware matching adaptation network for aero-engine imbalanced multi-source cross-domain fault diagnosis under class shift. EAAI, 149, 110510. (www)

  @article{ , abstract = {Domain adaptation methods have been effectively applied in aero-engine fault diagnosis to address the issue of data distribution shifts. In practice, fault data for aero-engines are often collected from multiple source domains, each containing only a subset of health states, leading to class shift. Furthermore, the collection of fault data is costly, resulting in a long-tail distribution. However, most existing research focuses on single-source domain settings and lacks a unified framework to address both class shift and class imbalance. To address this gap, a weighted class-aware matching adaptation network is proposed. By leveraging the maximum mean discrepancy metric, we design a class-aware domain discrepancy metric that models class discrepancies between the source and target domains, enhancing alignment at class boundaries. This metric also integrates class balance weights to emphasize minority classes in a cost-sensitive manner. Moreover, a class-aware health state recognition loss encourages the model to focus on source domains resembling the target domain, enhancing knowledge transfer. Finally, the proposed method is validated through cross-domain fault diagnosis experiments using the turbofan engine fault dataset and the Case Western Reserve University bearing dataset, demonstrating its effectiveness compared to several advanced methods.}, archive = {J_EAAI}, author = {Yu-Qiang Wang and Yong-Ping Zhao and Bo-Yu Liang and Tian-Ding Zhang and Kuan-Xin Hou}, doi = {10.1016/j.engappai.2025.110510}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110510}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weighted class-aware matching adaptation network for aero-engine imbalanced multi-source cross-domain fault diagnosis under class shift}, volume = {149}, year = {2025}, }
• (2025). Applications of artificial intelligence in industry 4.0 and smart manufacturing. EAAI, 149, 110509. (www)

  @article{ , abstract = {This special issue was initiated at the 10th IFAC triennial conference MIM 2022 concerning the topics dealing with applications of artificial intelligence in Industry 4.0, for especially process planning, reconfigurable manufacturing, robotics, production planning, fault detection and diagnostics.}, archive = {J_EAAI}, author = {Audrey Cerqueus and Alexandre Dolgui and Dmitry Ivanov and Alexandr Klimtchik and David Lemoine and Anatol Pashkevich}, doi = {10.1016/j.engappai.2025.110509}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110509}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Applications of artificial intelligence in industry 4.0 and smart manufacturing}, volume = {149}, year = {2025}, }
• (2025). Hyperspectral image mixed noised removal via jointly spatial and spectral difference constraint with low-rank tensor factorization. EAAI, 149, 110508. (www)

  @article{ , abstract = {The denoising of hyperspectral image (HSI) plays a crucial role in the subsequent interpretation and application. The rise of artificial intelligence technology has brought new opportunities for hyperspectral image denoising, and its potential and advantages in this field are gradually changing the traditional denoising pattern. This paper proposes a jointly spatial and spectral difference constraints with low-rank tensor factorization. Firstly, the spatial and spectral difference is combined in the framework of low-rank tensor factorization, to fully mine global spatial–spectral information and improve the removal ability of complex distribution noise. Secondly, based on the premise of effectively preserving HSI intrinsic three-dimensional structure, the spatial horizontal and vertical difference constraints are used to mine the local smoothness and similarity of spatial. Thirdly, the full-band spectral difference constraint could not only characterize the continuity and sparsity of the whole spectral domain, but also effectively characterize the noise distribution with linear structure. Finally, experiments on simulated and real HSIs show that the proposed method outperforms state-of-the-art methods in removing mixed noise performance.}, archive = {J_EAAI}, author = {Qiang Zhang and Yaming Zheng and Yushuai Dong and Chunyan Yu and Qiangqiang Yuan}, doi = {10.1016/j.engappai.2025.110508}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110508}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hyperspectral image mixed noised removal via jointly spatial and spectral difference constraint with low-rank tensor factorization}, volume = {149}, year = {2025}, }
• (2025). An image segmentation method for solid-liquid separation on shale shaker based on an improved U2Net. EAAI, 149, 110507. (www)

  @article{ , abstract = {In the actual production process of shale shakers, detecting the solid-liquid separation state of the screen surface faces numerous challenges, such as difficulty in recognizing the mud boundary, insufficient anti-interference ability, and misjudgment caused by background interference. To address these issues, this paper proposes a screen surface mud image segmentation method based on U 2 Net, namely CBAM-U 2 Net. By introducing the Convolutional Block Attention Module (CBAM) and combining it with Multi-layer Recursive Residual Blocks (RSU), a network structure is designed that can efficiently fuse global and local features, significantly improving segmentation accuracy and robustness. The network includes encoder and decoder parts, employing convolution, batch normalization, ReLU activation, and multi-scale feature fusion strategies. Experimental results show that the CBAM-U 2 Net method demonstrates excellent segmentation performance under various working conditions, achieving outstanding results with mIoU, F1-score, Precision, and Recall at 83.38%, 89.75%, 89.38%, and 92.64%, respectively, with significantly enhanced anti-interference capability. The CBAM-U 2 Net method provides an efficient and reliable solution for the intelligent monitoring of the solid-liquid separation state in shale shakers, offering significant practical application value.}, archive = {J_EAAI}, author = {Wenbin Wang and Yongjun Hou and Rui Jiang and Pan Fang and Hong Peng and Qing Li and Huachuan Li}, doi = {10.1016/j.engappai.2025.110507}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110507}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An image segmentation method for solid-liquid separation on shale shaker based on an improved U2Net}, volume = {149}, year = {2025}, }
• (2025). Efficient kriging-based wall deflection prediction in braced excavation considering model and measurement errors. EAAI, 149, 110506. (www)

  @article{ , abstract = {The deflection prediction of diaphragm walls stands as a critical aspect of safety management for excavation construction. Bayesian updating holds a prominent position among several existing prediction methods, owing to its probabilistic predictions and the ability to simultaneously consider prior knowledge and observational data. This paper develops a comprehensive Bayesian updating framework for wall deflection prediction in braced excavation, taking into account both model error associated with the prediction model chosen and measurement error. Further, both of these errors typically exhibit depth dependency and spatial correlation and thus are modeled as stochastic processes. An innovative efficient Kriging (e-Kriging) surrogate model and an Expectation-Maximum algorithm based adaptive importance sampling (AIS) method are proposed to improve computational efficiency. The prediction performance of this proposed framework is verified via an exhaustively reported Taipei National Enterprise Center (TNEC) excavation project. Moreover, this paper compares the influences of different selections of prior distributions and measurement data on the predictions. Results suggest that using only a subset of the middle portion of measurement data for updating is often a more appropriate choice.}, archive = {J_EAAI}, author = {Xiong Xiao and Quanwang Li and Hao Zhang}, doi = {10.1016/j.engappai.2025.110506}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110506}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient kriging-based wall deflection prediction in braced excavation considering model and measurement errors}, volume = {149}, year = {2025}, }
• (2025). Consistency-based decision-making method with linguistic Q-rung orthopair fuzzy preference relation for power battery selection of new energy vehicles. EAAI, 149, 110505. (www)

  @article{ , abstract = {In the era of global petrochemical depletion and increasingly serious environmental pollution, new energy vehicles, as a key industry to build a sustainable low-carbon society, have been paid more and more attention by countries all over the world. As the “heart” of new energy vehicles, power battery plays an important role in the core competitiveness of enterprises. Aiming at the fuzziness and uncertainty of complex power battery selection, a two-stage consistency optimization model based on preference relations and an interactive consistency improvement process are established in this paper. Firstly, by considering the interaction between membership and non-membership, this paper proposes an improved linguistic q-rung orthopair fuzzy weighted averaging operator. Then, the concept of linguistic q-rung orthopair fuzzy preference relation (Lq-ROFPR) is proposed, and its additive consistency index is given based on linguistic scaling function. Whereafter, for the Lq-ROFPR with unacceptable consistency, an interactive mechanism is proposed to improve the consistency level, which considers the minimum adjustment size of preference modification and the minimum number of adjustment elements in turn. Moreover, the method for solving the multi-attribute decision-making problems is formed and applied to the selection of power batteries in XP automobile company. Finally, the simulation experiment and comparative analysis with other methods show the effectiveness and rationality of this method in consistency optimization.}, archive = {J_EAAI}, author = {Xin Dong and Peide Liu and Peng Wang and Xiaoming Wu}, doi = {10.1016/j.engappai.2025.110505}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110505}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Consistency-based decision-making method with linguistic Q-rung orthopair fuzzy preference relation for power battery selection of new energy vehicles}, volume = {149}, year = {2025}, }
• (2025). Syn-rPPG: Improving unsupervised remote photoplethysmography extraction with synthesized videos using generative models. EAAI, 149, 110504. (www)

  @article{ , abstract = {Remote photoplethysmography (rPPG) is a non-contact technology used to capture cardiac activity from the face, providing measurements of physiological parameters. Current unsupervised methods for rPPG tasks often focus on contrastive learning, which highlights relationships between samples but struggles with a lack of diverse training data, particularly in terms of varying skin colors and motion types. This limits model effectiveness in complex real-world scenarios. Generative models offer a potential solution by creating synthetic samples to enrich the training data. In this study, we explore the impact of using synthetic videos generated by style transfer and motion transfer techniques to enhance unsupervised rPPG tasks. We generate two types of synthetic videos: skin color synthetic videos and motion synthetic videos. These address the key challenges in rPPG, namely skin color variations and motion artifacts. Our analysis shows that these synthetic videos provide valuable physiological information, improving the performance and robustness of unsupervised models. Additionally, we propose a novel lightweight rPPG network, Style-Aware rPPG Fusion Net (SAFNet), based on an encoder–decoder structure, which is optimized for joint training with synthetic videos. By incorporating a feature fusion approach, SAFNet captures richer spatiotemporal information, resulting in superior performance and robustness. Extensive experiments on four public benchmark datasets demonstrate that our method achieves excellent results, particularly in challenging conditions, proving the effectiveness of using synthetic data to enhance remote physiological measurements.}, archive = {J_EAAI}, author = {Tianqi Liu and Hanguang Xiao and Yisha Sun and Kun Zuo and Qihang Zhang and Zhipeng Li and Feizhong Zhou}, doi = {10.1016/j.engappai.2025.110504}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110504}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Syn-rPPG: Improving unsupervised remote photoplethysmography extraction with synthesized videos using generative models}, volume = {149}, year = {2025}, }
• (2025). Democratizing eye-tracking? appearance-based gaze estimation with improved attention branch. EAAI, 149, 110494. (www)

  @article{ , abstract = {Appearance-based gaze estimation in 2-dimensional screen coordinates–the prediction of the users’ gaze from webcam footage–cannot yet compete in accuracy with infrared (IR) eye trackers. Yet by circumventing the constraints of requiring dedicated hardware, it shows great potential in many technological industries, as evidenced by some readily available commercial solutions, bringing democratization of eye tracking closer to the people. We present Residual Appearance-based Gaze Estimation network (RAGE-net), a novel convolutional neural network for gaze estimation without need of calibration, utilizing a fraction of computational resources required by similar networks, while also achieving competitive accuracy. The angular error is measured as 4.08°in the MPIIFaceGaze dataset (Max Planck Institute for Informatics Faze Gaze) and 3.96°in the MPIIGaze dataset. The architecture’s principles, covered by a comprehensive ablation study, include an attention branch, residual learning, weight sharing between eye channels, batch normalization and an eye image input normalization pipeline that removes dependence on full face input. With RAGE-net, we conduct an applicability study for gaze estimation approaches of similar accuracy for interpreting on-screen gaze in praxis. Findings demonstrate low heatmap validity, with coarse heatmaps as potential adaptation to approximate IR eye tracking. The effects of environmental factors such as camera position, illumination, distance and glasses are analyzed in-depth.}, archive = {J_EAAI}, author = {Eduard Kuric and Peter Demcak and Jozef Majzel and Giang Nguyen}, doi = {10.1016/j.engappai.2025.110494}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110494}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Democratizing eye-tracking? appearance-based gaze estimation with improved attention branch}, volume = {149}, year = {2025}, }
• (2025). A three-dimensional dynamic spatial-temporal graph neural network for ocean temperature field prediction. EAAI, 149, 110492. (www)

  @article{ , abstract = {Accurate prediction of the ocean temperature field is vital for the protection of marine ecosystems under climate change. However, most existing methods only consider temporal changes, ignoring the rich three-dimensional (3D) dynamic spatial characteristics of the ocean temperature field. To improve prediction accuracy, we propose a fine-grained modeling method that uses the spatial correlations in the ocean temperature field, called the 3D Dynamic Spatial-Temporal Graph Neural Network (3D-DSTGN). Specifically, according to the 3D spatial features and dynamic spatial dependencies of the ocean temperature field, we first decompose the spatial correlation of the ocean temperature field into long-term static and short-term dynamic parts through statistical analysis of real ocean temperature datasets. We then build a 3D dynamic graph structure learning module to create static and dynamic graph structures with 3D spatial features to model and capture the corresponding spatial correlations. Next, based on the two graph structures, we apply a dual-mode graph convolution block to fully capture the dynamic spatial dependencies of the ocean temperature field. Furthermore, we use a multi-scale temporal convolution block to capture complex temporal dependencies from historical ocean temperature data. Finally, the dual-mode graph convolution block and the multi-scale temporal convolution block construct the spatio-temporal recurrent module, which extracts complex dynamic spatio-temporal contextual dependencies. On a large-scale real ocean temperature dataset from the sea surface to a subsurface depth of approximately 2,000 m, 3D-DSTGN outperforms other baselines in experiments across different temporal and spatial scales, effectively modeling the dynamic three-dimensional spatial relationships of the ocean temperature field.}, archive = {J_EAAI}, author = {Shuai Zhang and ZhuoLin Li and XiaoYu He and Jie Yu and LingYu Xu}, doi = {10.1016/j.engappai.2025.110492}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110492}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A three-dimensional dynamic spatial-temporal graph neural network for ocean temperature field prediction}, volume = {149}, year = {2025}, }
• (2025). Ergonomic conscious scheduling of maintenance activities in marine vehicles using an optimized non-dominated sorting genetic algorithm-II – An application of job-shop scheduling. EAAI, 149, 110491. (www)

  @article{ , abstract = {Incorporating ergonomics in marine activities is critical due to the extreme working conditions and limited crew in marine vehicles, aiming to enhance productivity and job performance by reducing the risks of fatigue, stress, and work-related musculoskeletal disorders. This paper introduces an innovative analogy of the flexible job-shop scheduling problem with ergonomic considerations (AFJSP-ER) to schedule maintenance activities in marine systems, addressing the dual objectives of optimizing productivity and promoting ergonomic relief. A novel metric, ‘ergonomic impact load’ is introduced to assess the actual workload of the crew by combining the processing time and the rapid entire body assessment (REBA) score of an operation. To solve the AFJSP-ER, an optimized non-dominated sorting genetic algorithm-II (ONSGA) is proposed, incorporating an optimized random crossover (ORX) operator. The ORX operator is fine-tuned using the Taguchi method to determine the optimal number of elements for crossover, while non-dominated sorting ensures the selection of superior individuals after crossover and mutation. The effectiveness of the proposed ONSGA has been validated through extensive experiments on newly developed test instances and using an industrial case study from the ship engine compartment. The results also indicate that the AFJSP-ER approach effectively optimizes productivity and promotes ergonomic relief, offering a practical solution for scheduling in ergonomically challenging marine environments.}, archive = {J_EAAI}, author = {Shaban Usman and Cong Lu}, doi = {10.1016/j.engappai.2025.110491}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110491}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ergonomic conscious scheduling of maintenance activities in marine vehicles using an optimized non-dominated sorting genetic algorithm-II – An application of job-shop scheduling}, volume = {149}, year = {2025}, }
• (2025). Multi-stage feature aggregation transformer for image rain and haze joint removal. EAAI, 149, 110490. (www)

  @article{ , abstract = {This study investigates the complex meteorological phenomenon of rain-haze mixtures, resulting from water vapor condensation in real-world rainy environments. Current methodologies predominantly focus on rain removal, often neglecting the image degradation caused by haze. We propose a multi-stage feature aggregation transformer (MFAFormer). The model consists of multiple stages, each comprising an information calibration module and a feature aggregation module. Specifically, the information calibration module extracts local high-frequency rain features, whereas the feature aggregation module extracts global low-frequency haze features and integrates features across stages. Benefiting from multi-stage feature extraction and aggregation, MFAFormer can efficiently and robustly address the complex degradations caused by the simultaneous presence of rain and haze. Additionally, we propose a novel rain and haze mixed dataset RainHaze Synscapes. The dataset contains large variations in rain streaks, haze density, and scene contents. Experimental results indicate that the model surpasses other methods on the dataset. MFAFormer demonstrates a remarkable improvement, achieving a 37.46% increase in peak signal-to-noise ratio (PSNR) and a 10.48% increase in structural similarity (SSIM) compared to the baseline model on the RainHaze Synscapes dataset. This accomplishment offers valuable insights for the research of rain and haze joint removal domain.}, archive = {J_EAAI}, author = {Zhengran Xia and Lei Dai and Zhihua Chen and Kai Chen and Ran Li}, doi = {10.1016/j.engappai.2025.110490}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110490}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-stage feature aggregation transformer for image rain and haze joint removal}, volume = {149}, year = {2025}, }
• (2025). An end-to-end semantic-guided infrared and visible registration-fusion network for advanced visual tasks. EAAI, 149, 110489. (www)

  @article{ , abstract = {The integration of infrared and visible images is key to complex visual tasks, with image registration as a fundamental initial step. We explore these tasks as a unified registration-fusion process. Traditional global registration may use a multitude of feature points for alignment, achieving global optimality. However, for advanced tasks like pose recognition, object tracking, and person re-identification, the relevant semantic information may not align accurately, impacting fusion effectiveness and task performance. To address this issue, we propose a Semantic-Aware on-Demand registration-fusion network. 1) We design the Semantic-Aware Module, employing Grounding DINO and Segment Anything to capture semantic regions of interest needed for advanced tasks. 2) Combining with the Semantic-Aware Module, we design a Hierarchical Orientation Line operator and a Deep Hybrid Matching to ensure precise feature matching within semantic regions of interest. 3) To enhance fusion visual effects, a novel image fusion module is designed to facilitate high-quality image fusion within these regions. This method is versatile and applicable to a wide range of advanced visual tasks. We compare this method with six image matching and nine image fusion methods, underscoring its efficacy in advanced visual tasks. The experimental results indicate that both the registration module and the fusion module achieve optimal performance individually. When it comes to joint tasks of registration and fusion in visual tasks, our method similarly exhibits the best performance in the 6 × 10 joint tasks.}, archive = {J_EAAI}, author = {Meng Sang and Housheng Xie and Jingrui Meng and Yukuan Zhang and Junhui Qiu and Shan Zhao and Yang Yang}, doi = {10.1016/j.engappai.2025.110489}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110489}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An end-to-end semantic-guided infrared and visible registration-fusion network for advanced visual tasks}, volume = {149}, year = {2025}, }
• (2025). Multiscale self-attention for unmanned ariel vehicle-based infrared thermal images detection. EAAI, 149, 110488. (www)

  @article{ , abstract = {Object detection and recognition in unmanned aerial vehicle-based images is critical for various applications but is often challenged by complex backgrounds, diverse object scales, densely clustered small objects, and uneven object distributions. This paper introduces a novel deep learning-based artificial intelligence framework that integrates the Multiscale Self-Attention Guidance and Feature Fusion Network with the You Only Look Once model, tailored explicitly for artificial intelligence-driven unmanned aerial vehicle-based infrared thermal image analysis. The proposed methodology offers four key advancements in the You Only Look Once architecture to enhance object detection performance. First, the Multi-Head Self-Attention Transformer module combines global and local information, enabling precise object localization while mitigating the influence of complex backgrounds. Second, the Multiscale Parallel Sampling Feature Fusion module optimizes the fusion of multiscale features. Third, fine-grained shallow feature maps are integrated into the fusion process to detect densely packed small objects accurately. Lastly, the Inverse-Residual Feature Enhancement module, positioned before the detection head, enhances feature extraction for small objects. Experimental evaluations on the High Altitude Infrared Thermal Unmanned Aerial Vehicle dataset demonstrate significant improvements, achieving a Mean Average Precision of 95.1%, Recall of 92.0%, and F1-Score of 91.0%. The framework’s robustness is further validated on the Wildland-fire Infrared Thermal Unmanned Aerial System dataset, achieving a Mean Average Precision of 82.1%, Recall of 88.0%, and F1-Score of 82.0%. Comparative analyses with state-of-the-art methods confirm its superiority and offer a scalable artificial intelligence-driven solution for unmanned aerial vehicle applications, advancing object detection capabilities in critical scenarios.}, archive = {J_EAAI}, author = {Muhammad Shahroze Ali and Afshan Latif and Muhammad Waseem Anwar and Muhammad Hashir Ashraf}, doi = {10.1016/j.engappai.2025.110488}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110488}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multiscale self-attention for unmanned ariel vehicle-based infrared thermal images detection}, volume = {149}, year = {2025}, }
• (2025). Intelligent fault diagnosis of nonlinear uncertain industrial processes based on kernel local–global interval embedding algorithm. EAAI, 149, 110486. (www)

  @article{ , abstract = {With the rapid development of a new generation of Big Data and artificial intelligence, intelligent fault diagnosis of industrial processes including chemical processes, coal mining equipment operations, etc., has become increasingly important. The local–global interval embedding algorithm (LGIEA) has attracted significant attention for its capability to simultaneously extract local and global features from interval data. However, this method can only process linear interval data and performs poorly in terms of extracting strong nonlinear features. To solve the problem, this study proposes a new intelligent fault diagnosis method based on kernel LGIEA (KLGIEA), which extends the linear process monitoring model to nonlinearity. First, the interval inner product estimation (IIPE) is transformed into the kernel IIPE by introducing kernel function, which can not only inherit the advantage that LGIEA can extract both global and local features of data simultaneously, but also has stronger applicability to nonlinear data in industrial processes. Second, the four statistics defined can effectively monitor the fault of industrial equipment under strong interference environment such as noises, and the nonlinear reconstruction contribution (NRC) can effectively identify the fault variables, improve the fault diagnosis ability of KLGIEA. Finally, two cases of the Tennessee Eastman process (TEP) simulation data from Eastman Company and site shearer fault data obtained from Shaqu No. 2 coal mine show that KLGIEA is significantly superior to complete information principal component analysis (PCA), midpoint-radius kernel PCA, and LGIEA in processing nonlinear interval data, improving accuracy, applicability, and reliability of algorithm.}, archive = {J_EAAI}, author = {Ning Li and Hua Ding and Xiaochun Sun and Zeping Liu}, doi = {10.1016/j.engappai.2025.110486}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110486}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent fault diagnosis of nonlinear uncertain industrial processes based on kernel local–global interval embedding algorithm}, volume = {149}, year = {2025}, }
• (2025). Distributed data-driven iterative learning consensus tracking for unknown multi-agent systems using evolutionary neural networks. EAAI, 149, 110485. (www)

  @article{ , abstract = {This paper provides a data-driven distributed parameter adaptive iterative learning consensus tracking strategy for nonlinear nonaffine discrete-time multi-agent systems with unknown dynamics. By transforming the ideal learning controller on the timeline into a direct iterative learning control strategy in the iterative domain, the design of the control protocol is only data-driven. Unlike existing parameter tuning control methods, the parameter tuning approach presented in this paper adjusts the parameters online through topological information, eliminating the need for multiple trials and adjustments based on experience. The gain time variability of multi-agent systems is learned and compensated by the extended generalized regression neural networks evolution control. By introducing a limited incremental evolution mechanism, the optimal control parameters can be adjusted online during the control process to find the system trajectory to achieve optimal output synchronization, so as to improve the control efficiency of iterative learning control. Different from the existing directed fixed topology works of multi-agent systems, the consensus convergence properties of fixed directed communication topologies and iterative time-varying communication topologies along the iterative domain are established by contraction mapping theorem. Two numerical simulation examples are conducted to validate the effectiveness of the proposed control protocol.}, archive = {J_EAAI}, author = {Kechao Xu and Bo Meng and Zhen Wang}, doi = {10.1016/j.engappai.2025.110485}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110485}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Distributed data-driven iterative learning consensus tracking for unknown multi-agent systems using evolutionary neural networks}, volume = {149}, year = {2025}, }
• (2025). A multi-point dam deformation prediction model based on spatiotemporal graph convolutional network. EAAI, 149, 110483. (www)

  @article{ , abstract = {Dams are essential for services such as water supply and flood protection, making their structural health and safety crucial to prevent catastrophic failures. Monitoring dam displacement is a critical method for assessing its current condition and predicting future behavior. Most displacement monitoring models primarily focus on temporal features and the relationship between environmental factors and dam displacement, often overlooking the spatial relationships in the data. Even multi-point models, designed to handle multiple measurement locations, struggle to effectively account for the spatial coordination between these points. To address these challenges, this paper proposes a multi-point displacement monitoring model based on a variational auto-encoder (VAE) and a spatiotemporal graph convolutional network (STGCN). The graph structure is utilized to represent the coordinated deformation relationships among monitoring points, while also capturing temporal features and nonlinear relationships between environmental factors and dam displacement. The VAE model is first used to extract latent feature representations from historical monitoring data and monitoring point coordinates. The K-nearest neighbors (KNN) method is then applied to calculate the connection weights between monitoring points, constructing the adjacency matrix. Graph convolutional network (GCN) is utilized to extract spatial features, while gated recurrent units (GRU) capture temporal dependencies, enabling accurate multi-point displacement prediction. The model's effectiveness and accuracy are validated through comparisons with both single-point and multi-point models, while the impact of K-nearest neighbors and learnable position encoding on model performance is also evaluated. The results demonstrate that the proposed model significantly outperforms others, showing superior predictive accuracy and generalization capabilities.}, archive = {J_EAAI}, author = {Taiqi Lu and Hao Gu and Chongshi Gu and Chenfei Shao and Dongyang Yuan}, doi = {10.1016/j.engappai.2025.110483}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110483}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-point dam deformation prediction model based on spatiotemporal graph convolutional network}, volume = {149}, year = {2025}, }
• (2025). Visual object tracking using learnable target-aware token emphasis. EAAI, 149, 110482. (www)

  @article{ , abstract = {Visual object tracking, which involves tracking the spatial location of a target object either within a single viewpoint or across various camera perspectives, is an important task in computer vision. Deep neural networks, especially vision transformers, typically outperform traditional methods and have thus become the preferred choice for visual object tasks. However, existing visual object tracking frameworks still struggle to adapt to targets with continuously changing appearances within the current frame, as they rely heavily on the static initial target template rather than continuously emphasizing the evolving target features. In this paper, we introduce a visual object tracking network with a learnable target-aware token emphasis, which is composed of vision transformer backbone embedded in the token emphasizer, localization head and target template update decision module. The learnable target-aware token emphasizer and target template update decision modules in the proposed model contribute to stabilizing visual object tracking across various scenarios. This is achieved not only by emphasizing features that have a relationship between the target template and the search region but also by reducing irrelevant features and consistently updating the high-quality target template online during the process. Qualitative and quantitative analyses, including ablation analysis across a diverse set of tracking benchmark datasets, validate the robustness of the proposed tracking framework. The code and trained models are available at https://github.com/qkdkralsgh/TETrack .}, archive = {J_EAAI}, author = {Minho Park and Jinjoo Song and Sang Min Yoon}, doi = {10.1016/j.engappai.2025.110482}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110482}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Visual object tracking using learnable target-aware token emphasis}, volume = {149}, year = {2025}, }
• (2025). Phased noise enhanced multiple feature discrimination network for fabric defect detection. EAAI, 149, 110480. (www)

  @article{ , abstract = {Fabric defect detection is crucial for evaluating the quality of textile products. However, the subtlety and scarcity of fabric defects pose challenges to the task of detecting. Therefore, we propose a P hased N oise Enhanced M ultiple F eature D iscrimination Network, which is based on phased noise enhancement strategy and multiple feature discrimination module to improve the model’s ability to identify complex and subtle flaws. Specifically, we propose the phased noise enhancement strategy in the feature space to simulate feature-level anomalies that are closer to reality. This strategy can improve the input quality of the feature reconstructor, so that helps its perception and reconstruction ability. Then, we propose the multiple feature discrimination module, which has dual feature branches to improve its ability to distinguish more complex detailed texture features. In addition, we propose a subsampling module to reduce feature redundancy and ensure efficient inference speed. Finally, we conduct extensive experiments and ablation studies on two publicly available fabric datasets, AITEX and Kaggle Fabric. The experimental results show that the proposed method achieved 92% and 100% image level metrics and 97.5% and 67.1% pixel level metrics on two datasets, respectively, which is superior to the current state-of-the-art methods. In addition, our method also demonstrated significant performance in generalization experiments.}, archive = {J_EAAI}, author = {Haoran Ma and Zuoyong Li and Haoyi Fan and Xiangpan Zheng and Jiaquan Yan and Rong Hu}, doi = {10.1016/j.engappai.2025.110480}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110480}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Phased noise enhanced multiple feature discrimination network for fabric defect detection}, volume = {149}, year = {2025}, }
• (2025). Information extraction from multi-layout invoice images using FATURA dataset. EAAI, 149, 110478. (www)

  @article{ , abstract = {Document analysis and understanding models often require extensive annotated data to be trained. However, various document-related tasks extend beyond mere text transcription, requiring both textual content and precise bounding-box annotations to identify different document elements. Collecting such data becomes particularly challenging, especially in the context of invoices, where privacy concerns add an additional layer of complexity. Invoices are critical documents in many business processes, but existing datasets for invoice analysis are limited in size and diversity, hindering the development of robust models. Current datasets do not adequately address the need for diverse layouts and comprehensive annotations, which are essential for training models capable of handling real-world variations in invoice documents. In this paper, we introduce FATURA, a pivotal resource for researchers in the field of document analysis and understanding. FATURA is a highly diverse dataset featuring multi-layout, annotated invoice document images. Comprising 10,000 invoices with 50 distinct layouts, it represents the largest openly accessible image dataset of invoice documents known to date. We provide an extensive evaluation using different information extraction methods under diverse training and evaluation scenarios, including precision, recall, and F1-score. The evaluation includes a visual-based approach using object detection for text region classification, a multi-modal strategy integrating visual and textual data for granular content comprehension, and a hybrid approach combining these methods. The dataset is freely accessible at this https://zenodo.org/record/8261508 , empowering researchers to advance the field of document analysis and understanding.}, archive = {J_EAAI}, author = {Mahmoud Limam and Marwa Dhiaf and Yousri Kessentini}, doi = {10.1016/j.engappai.2025.110478}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110478}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Information extraction from multi-layout invoice images using FATURA dataset}, volume = {149}, year = {2025}, }
• (2025). Finite time multilayer neural network command filter backstepping controller design for large scale uncertain nonlinear systems. EAAI, 149, 110474. (www)

  @article{ , abstract = {This study presents a novel dynamical multilayer neural network finite time command filter backstepping control scheme. This method realizes the finite time robust tracking control of uncertain nonlinear system. The uncertainty in system varies in large scale around system states. Its boundary is unknown and unavailable before design. The multilayer neural network (MNN) approximater is redesigned into the backstepping controller instead of the common radial basis function (RBF) neural network (NN) and Fuzzy System (FS) to realize the accuracy approximation of the large scale uncertain structure. The introduction of the MNN approximater overcomes the drawback of local identification constraint of RBF NN and Fuzzy System without the structure knowledge and boundary of uncertainty before design. Otherwise, owing to the MNN structure is more complex than common three layer RBF NN, the approximation costs more time to dynamically tune weight parameters online. In order to make up the time consistent between the MNN approximation and the backstepping process, the finite time (FT) command filter (CF) backstepping control strategy balancing the two distinct procedures guarantees the MNN identification of larger scale uncertainty and backstepping control process consistently convergence into a smaller area in uniform finite time interval. Finally, through a practical example, the effectiveness and advantages of are illustrated by comparison between this mechanism and traditional RBF NN method.}, archive = {J_EAAI}, author = {Qitian Yin and Quanqi Mu and Jianbai Yang}, doi = {10.1016/j.engappai.2025.110474}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110474}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Finite time multilayer neural network command filter backstepping controller design for large scale uncertain nonlinear systems}, volume = {149}, year = {2025}, }
• (2025). HT-AggNet: Hierarchical temporal aggregation network with near-zero-cost layer stacking for human activity recognition. EAAI, 149, 110465. (www)

  @article{ , abstract = {With the steady growth of sensor technology and wearable devices in pervasive computing applications, sensor-based human activity recognition has gained attention in fields such as healthcare monitoring and fitness tracking. This has resulted in an increased need for accurate and real-time systems. Recent studies to satisfy the real-time conditions have attempted to design lightweight neural networks by mainly restricting the number of layers shallowly, which has decreased both inference time and accuracy. To recover the loss of accuracy, we propose an innovative hierarchical temporal aggregation network (HT-AggNet) that allows the network architecture to be deeper, leading to an accuracy gain with only a near-zero increase in computational cost. Furthermore, a temporal glance convolution is presented to model the global context information of the signal patterns. Consequently, the HT-AggNet hierarchically extracts the local and global temporal information and then merges them based on hierarchical temporal aggregation. In our experiments, the HT-AggNet outperformed existing methods on seven publicly available datasets and achieved state-of-the-art performance. The source code for the HT-AggNet is publicly available at https://github.com/jgpark92/HT-AggNet .}, archive = {J_EAAI}, author = {Jaegyun Park and Dae-Won Kim and Jaesung Lee}, doi = {10.1016/j.engappai.2025.110465}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110465}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {HT-AggNet: Hierarchical temporal aggregation network with near-zero-cost layer stacking for human activity recognition}, volume = {149}, year = {2025}, }
• (2025). Reliability evaluation of solar integrated power distribution systems using an evolutionary swarm algorithm. EAAI, 149, 110464. (www)

  @article{ , abstract = {The reliability of solar-integrated power distribution systems is significantly affected by intermittent solar generation and its impact on feeder voltages. While existing adequacy studies account for intermittency, they frequently overlook feeder voltages due to the computational burden of the Alternating Current Optimal Power Flow (AC-OPF) analysis. Addressing this gap, we propose an efficient framework based on an Evolutionary Swarm Algorithm (ESA) to integrate AC-OPF analysis into the reliability evaluation of power distribution systems. The sampling mechanism of ESA reduces the application of time-consuming AC-OPF and allows the fast estimation of reliability indices. The performance of the proposed framework is compared with Sequential Monte Carlo Simulation (SMCS), classical meta-heuristics, and three state-of-the-art meta-heuristics. Results demonstrate that our proposed framework can estimate the reliability indices approximately 34 times faster than SMCS without sacrificing accuracy. Furthermore, the ESA outperforms classical and state-of-the-art methods by over 23% in event sampling efficiency. Friedman and Nemenyi post-hoc tests conclude that ESA’s results significantly differ from others. We utilize the proposed framework in a case study to analyze the influence of solar photovoltaic integration on distribution system reliability. Another case study investigates the impact of dynamic tap changing of power transformers on the reliability improvement of distribution systems.}, archive = {J_EAAI}, author = {P.A.G.M. Amarasinghe and S.K. Abeygunawardane and C. Singh}, doi = {10.1016/j.engappai.2025.110464}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110464}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reliability evaluation of solar integrated power distribution systems using an evolutionary swarm algorithm}, volume = {149}, year = {2025}, }
• (2025). Dynamic feature capturing in a fluid flow reduced-order model using attention-augmented autoencoders. EAAI, 149, 110463. (www)

  @article{ , abstract = {This study looks into how adding adaptive attention to convolutional autoencoders can help reconstruct flow fields in fluid dynamics applications. The study compares the effectiveness of the proposed adaptive attention mechanism with the convolutional block attention module approach using two different sets of datasets. The analysis encompasses the evaluation of reconstruction loss, latent space characteristics, and the application of attention mechanisms to time series forecasting. Combining adaptive attention with involution layers enhances its ability to identify and highlight significant features, surpassing the capabilities of the convolutional block attention module. This result demonstrates an increase of over 20% in the accuracy of reconstruction. Latent space analysis shows the adaptive attention mechanism’s complex and flexible encoding, which makes it easier for the model to represent different types of data. The study also looks at how attention works and how it affects time series forecasting. It shows that a new method that combines multi-head attention and bidirectional long-short-term memory works well for forecasting over 5 s of futures of flow fields. This research provides valuable insights into the role of attention mechanisms in improving model accuracy, generalization, and forecasting capabilities in the field of fluid dynamics.}, archive = {J_EAAI}, author = {Alireza Beiki and Reza Kamali}, doi = {10.1016/j.engappai.2025.110463}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110463}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic feature capturing in a fluid flow reduced-order model using attention-augmented autoencoders}, volume = {149}, year = {2025}, }
• (2025). Efficient generation of power system topology diagrams based on graph neural network. EAAI, 149, 110462. (www)

  @article{ , abstract = {Power system topology diagrams illustrate the physical and spatial relationship of system nodes and are widely used as a basic tool for displaying system structure. Well-presented topology diagrams provide better situational awareness for the operators, but their efficient generation remains a challenge. Existing approaches struggle to find a balance between visual aesthetics and the generation speed of the diagram. With the rapid changes in power system topology, there is a higher demand for the rendering speed of the graph data. To satisfy both the real-time requirement and the aesthetic quality, this paper proposes an integrated framework for efficiently generating power system topology diagrams. It consists of a Graph Neural Network (GNN) model and a graph fine-tuning model. This framework can directly optimize the raw topology diagram while preserving the relative positions of nodes in the initial layout. It achieves a decent trade-off between layout quality and computational expenses, enabling the generation of aesthetically satisfactory diagrams in a short time. Due to the strong generalization ability of GNN, the proposed model can be trained on small system datasets and used for inference on large systems. Case studies verify that the proposed GNN model can optimize the aesthetic metrics of topology diagram layouts within seconds to an average value of 0.55. Finally, it can be used in power system applications as a fundamental tool for topology diagram generation and optimization.}, archive = {J_EAAI}, author = {Chen Yang and Shengyang Wu and Tao Liu and Yixuan He and Jingyu Wang and Dongyuan Shi}, doi = {10.1016/j.engappai.2025.110462}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110462}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient generation of power system topology diagrams based on graph neural network}, volume = {149}, year = {2025}, }
• (2025). Enhancing small satellite image resolution via shrinking rearranged mechanism and multiscale reparameterized attention. EAAI, 149, 110460. (www)

  @article{ , abstract = {Small satellites, weighing under 1,000 kg, are increasingly used in defense, civil, and commercial sectors due to their cost-effectiveness and portability. However, their imaging resolution is limited by the size and cost of their apertures. Existing super-resolution (SR) algorithms struggle to reconstruct large-area surface information and fine local textures, and often require high computing power, making them unsuitable for small satellites. This study proposes a lightweight super-resolution network, MSRN (Multiscale Shrinking Rearranged Attention Network), engineered for deployment on small satellites. Specifically, MSRN employs a shrinking window-partition strategy to extract different ranges of feature priors and capture local high-frequency details. It also uses a channel rearranged mechanism to expand the receptive field and extract global context information. Additionally, a multiscale reparameterized attention group is designed to efficiently extract features and contours of objects at various scales, enhancing channel information representation. The use of reparameterization technology simplifies the model and enables fast response and processing of small satellite image data. Multiple comparisons on several popular public remote sensing datasets demonstrate that MSRN outperforms mainstream methods in terms of resource occupancy and reconstruction performance, and exhibits strong robustness and generalization across different scenarios.}, archive = {J_EAAI}, author = {Zhibo Zhao and Hu Liang and Yuchen Liu and Shengrong Zhao}, doi = {10.1016/j.engappai.2025.110460}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110460}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing small satellite image resolution via shrinking rearranged mechanism and multiscale reparameterized attention}, volume = {149}, year = {2025}, }
• (2025). Semantically complex audio to video generation with audio source separation. EAAI, 149, 110457. (www)

  @article{ , abstract = {Recent advancements in artificial intelligence for audio-to-video generation have shown the ability to generate high-quality videos from audio, particularly by focusing on temporal semantics and magnitude. However, existing works struggle to capture all semantics from audio, as real world audios often consist of mixed sources, making it challenging to generate semantically aligned videos. To solve this problem, we present a novel multi-source audio-to-video generation framework that incorporates decomposed multiple audio sources into video generative models. Specifically, our proposed Attention Mosaic directly maps each decomposed audio feature to the corresponding spatial attention feature. In addition, our condition injection module is helpful for producing more natural contexts with non-audible objects by leveraging the knowledge of existing generative models. Our experiments show that the proposed framework achieves state-of-the-art performance in representing both multi- and single-source audio-to-video generation methods.}, archive = {J_EAAI}, author = {Sieun Kim and Jaehwan Jeong and Sumin In and Seung Hyun Lee and Seungryong Kim and Saerom Kim and Wooyeol Baek and Sang Ho Yoon and Eugenio Culurciello and Sangpil Kim}, doi = {10.1016/j.engappai.2025.110457}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110457}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semantically complex audio to video generation with audio source separation}, volume = {149}, year = {2025}, }
• (2025). A novel multivariate nonlinear time-delayed grey model for forecasting electricity consumption. EAAI, 149, 110452. (www)

  @article{ , abstract = {Accurate and stable annual electricity consumption forecasting play vital role in modern social and economic development, which can provide effective planning and guaranteeing a reliable supply of sustainable electricity. Given that electricity consumption series present nonlinearity, poor information, and time-delayed characteristics, this paper proposes a multivariate nonlinear time-delayed grey model. Three primary efforts have been made as follows. First, we introduce the nonlinear and time-delayed terms into the typical multivariate grey model to identify the relationship between electricity consumption sequence and its driving factor sequence. Second, based on the Monte-Carlo simulation, an intelligent algorithm matching framework is designed to seek for the optimal model parameters of the model, which enhances the model’s applicability and flexibility. Third, we use datasets of China’s and America’s electricity consumption from 2000 to 2021 to validate the effectiveness of the newly-proposed model. Additionally, sensitivity analysis under different time horizons further verifies the model’s robustness. The experiment results indicates the superior prediction accuracy and robustness when comparing with other prevailing benchmarks. Overall, the newly-designed model is an effective technique for forecasting electricity consumption in China and America. Based on this, the forecasts of China’s and America’s electricity consumption in the following years can serve as a valuable reference for formulating related policies.}, archive = {J_EAAI}, author = {Wen-Ze Wu and Naiming Xie}, doi = {10.1016/j.engappai.2025.110452}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110452}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel multivariate nonlinear time-delayed grey model for forecasting electricity consumption}, volume = {149}, year = {2025}, }
• (2025). A deep reinforcement learning assisted adaptive genetic algorithm for flexible job shop scheduling. EAAI, 149, 110447. (www)

  @article{ , abstract = {The flexible job shop scheduling problem (FJSP) is a challenging combinatorial optimization problem in manufacturing systems. Existing intelligent optimization algorithms for FJSP are often hard to tune key parameters and operations efficiently, which losses the optimality of the obtained solution. To address the issues of genetic algorithms (GA) being prone to local optima and slow convergence, this paper proposes a deep reinforcement learning-assisted adaptive genetic algorithm (DRL-A-GA) for solving FJSP. In the proposed algorithm, continuous state vectors are used to represent the population state of the GA, and four mutation operations with respect to FJSP are designed as actions. Deep reinforcement learning is employed to adaptively tune the key parameters of the GA and dynamically select appropriate genetic operations. To validate the performance of DRL-A-GA, three sets of benchmark instances are selected for testing, and the results are compared with those of classical optimization algorithms and hybrid algorithms. The experimental results demonstrate that the proposed DRL-A-GA significantly outperforms both traditional optimization and intelligent hybrid optimization algorithms for solving FJSP, effectively improving solution quality and accelerating convergence.}, archive = {J_EAAI}, author = {Jian Ma and Weinan Gao and Weitian Tong}, doi = {10.1016/j.engappai.2025.110447}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110447}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep reinforcement learning assisted adaptive genetic algorithm for flexible job shop scheduling}, volume = {149}, year = {2025}, }
• (2025). An enhanced artificial bee colony algorithm with self-learning optimization mechanism for multi-objective path planning problem. EAAI, 149, 110444. (www)

  @article{ , abstract = {In recent years, path planning has been one of the most concerned problems in mobile robotics. This study investigates a multi-objective path planning problem focused on minimizing path length and maximizing path safety. Based on the characteristics of this problem, a mathematical model is established, and then an enhanced artificial bee colony algorithm is proposed to solve this problem. In the proposed algorithm, a new hybrid initialization strategy is designed to generate a high-quality initial population. In the employed bee phase, in addition to the crossover and mutation operators, two objective-oriented evolutionary operators are developed. In the onlooker bee phase, two self-learning optimization mechanisms are applied to the non-dominated and dominated individuals, respectively. Specifically, the collaborative-based optimization mechanism is designed to improve the quality of the non-dominated individuals. The dominance-guide optimization mechanism is developed to guide the dominated individuals to learn from the non-dominated ones. In the scout bee phase, a novel individual-restart strategy that considers the useful information of global best solutions is investigated, which increases the proposed algorithm’s exploration ability. Finally, the proposed algorithm is compared with five state-of-the-art algorithms on sixteen instances from four representative environments. Simulation results show that the proposed algorithm achieved average improvements of 2.60% and 90.77% on the hypervolume and inverted generational distance metrics, respectively, compared with the algorithm with the second-best performance. These demonstrate the effectiveness and high performance of the proposed algorithm for solving multi-objective path planning problems in terms of both population diversity and solution quality.}, archive = {J_EAAI}, author = {Fan Ye and Peng Duan and Leilei Meng and Hongyan Sang and Kaizhou Gao}, doi = {10.1016/j.engappai.2025.110444}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110444}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An enhanced artificial bee colony algorithm with self-learning optimization mechanism for multi-objective path planning problem}, volume = {149}, year = {2025}, }
• (2025). An effective exploration method based on N-step updated dirichlet distribution and Dempster–Shafer theory for deep reinforcement learning. EAAI, 149, 110443. (www)

  @article{ , abstract = {Deep reinforcement learning (DRL) has been regarded as a promising approach for solving decision-making problems. However, how to enhance the agent exploration ability is still an extremely challenging issue for existing methods, especially under sparse rewards. Facing with this challenge, we propose a novel efficient exploration method, which can comprehensively consider the uncertainty of the environment and the uncertainty of Q function, so as to improve the agent exploration efficiency. Specifically, we first construct an exploration policy by n-step updated Dirichlet distribution to implement the adaptive exploration of the agent to the environment, which can reduce the uncertainty of the agent about the environment to achieve global efficient exploration. Next, a state–action basic probability assignment (BPA) is constructed based on the Dempster–Shafer theory. On this basis, an interval Q function is designed by combining BPA and belief interval, which can effectively characterize the uncertainty of the Q function to achieve deep exploration. Then, the proposed method is applied to classic DRL algorithms, deep Q-network (DQN) and double DQN (DDQN), two novel algorithms are proposed. Finally, under a series of sparse external reward tasks, experimental results show that our proposed algorithms outperform several state-of-the-art DRL algorithms in term of exploring efficiency.}, archive = {J_EAAI}, author = {Fanghui Huang and Yixin He and Yu Zhang and Bin Chen and Lina Yang}, doi = {10.1016/j.engappai.2025.110443}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110443}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An effective exploration method based on N-step updated dirichlet distribution and Dempster–Shafer theory for deep reinforcement learning}, volume = {149}, year = {2025}, }
• (2025). HMKRec: Optimize multi-user representation by hypergraph motifs for knowledge-aware recommendation. EAAI, 149, 110441. (www)

  @article{ , abstract = {Knowledge graph-based recommender systems can explore users’ potential interests by learning user similarities, thereby further improving recommendation performance. However, existing methods focus only on the similarity between two users without considering the interaction patterns among multiple users, which overlook the influence of other users in user representation modeling. In this paper, we propose a novel framework using Hypergraph Motifs to optimize Multi-users representation for Recommendation (HMKRec). Specifically, HMKRec constructs a user–item hypergraph and maps it into a user–user adjacency graph. Then, it utilizes hypergraph motifs to model the interaction patterns of multiple users and reconstructs an implicit relationship network with weights and directions to explore high-order associations among multiple users. To learn the features of items and user relationships, we design a hierarchical graph convolution that integrates hypergraph convolutional networks and graph convolutional networks to obtain high-order representations of users. Finally, we propagate user preferences in the knowledge graph using the attention mechanism to obtain high-order representations of items for recommendation. Extensive experiments on three real-world datasets indicate that our method achieves at least a 1% performance improvement over the best-performing state-of-the-art baselines.}, archive = {J_EAAI}, author = {Di Wu and Mingjing Tang and Shu Zhang and Wei Gao}, doi = {10.1016/j.engappai.2025.110441}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110441}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {HMKRec: Optimize multi-user representation by hypergraph motifs for knowledge-aware recommendation}, volume = {149}, year = {2025}, }
• (2025). An adaptive traffic signal control scheme with proximal policy optimization based on deep reinforcement learning for a single intersection. EAAI, 149, 110440. (www)

  @article{ , abstract = {Adaptive traffic signal control (ATSC) is an important means to alleviate traffic congestion and improve the quality of road traffic. Although deep reinforcement learning (DRL) technology has shown great potential in solving traffic signal control problems, the state representation and reward design, as well as action interval time, still need to be further studied. The advantages of policy learning have not been fully applied in TSC. To address the aforementioned issues, we propose a DRL-based traffic signal control scheme with Poximal Policy Optimization (PPO-TSC). We use the waiting time of vehicles and the queue length of lanes represented the spatiotemporal characteristics of traffic flow to design the simplified traffic states feature vectors, and define the reward function that is consistent with the state. Additionally, we compare and analyze the performance indexes obtained by various methods using action intervals of 5s, 10s, and 15s. The algorithm is implemented based on the Actor-Critic architecture, using the advantage estimation and the clip mechanism to constrain the range of gradient updates. We validate the proposed scheme at a single intersection in Simulation of Urban MObility (SUMO) under two different traffic demand patterns of flat traffic and peak traffic. The experimental results show that the proposed method is significantly better than other compared methods. Specifically, PPO-TSC demonstrates a reduction of 24% in average travel time (ATT), a decrease of 45% in the average time loss (ATL), and an increase of 16% in average speed (AS) compared with the existing methods under peak traffic condition.}, archive = {J_EAAI}, author = {Lijuan Wang and Guoshan Zhang and Qiaoli Yang and Tianyang Han}, doi = {10.1016/j.engappai.2025.110440}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110440}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive traffic signal control scheme with proximal policy optimization based on deep reinforcement learning for a single intersection}, volume = {149}, year = {2025}, }
• (2025). Generative adversarial network-based inverse design of self-deploying soft kirigami composites for targeted shape transformation. EAAI, 149, 110417. (www)

  @article{ , abstract = {The design and development of morphing structures that transition from compact, transportable forms to stable, deployable configurations is crucial for advances in soft robotics, healthcare applications, and biomimetic systems. These structures often require customized functionalities and must self-deploy into precise target shapes. Therefore, the deformed shapes of such structures are usually prescribed and the parameters for their design are unknown. To obtain the fabrication parameters, the inverse problem needs to be solved, which quickly becomes quite challenging using conventional methods due to the high-dimensional nature of the inverse problem as well as the material and geometric nonlinearities. To overcome these challenges, we combine the best of the two worlds – physics and data – and present a data-driven approach for the inverse design of two-layered soft composites that utilize the principles of kirigami and strain mismatch to self-deploy into different three-dimensional shapes. At the center of our methodology is the generative adversarial network, designed to generate the necessary fabrication parameters. By using a pre-trained simulator network, we condition the generative model to generate feasible and accurate fabrication parameters that are used to make composites that deploy into the target shapes. Our findings demonstrate that the generative model is able to effectively predict kirigami patterns and pre-stretch values required to realize complex three-dimensional shapes from simple and diverse planar designs. By performing simulations and precise desktop experiments, we compare the target with deployed shapes and demonstrate the predictive capacity of the method.}, archive = {J_EAAI}, author = {Tomaž Brzin and M. Khalid Jawed and Miha Brojan}, doi = {10.1016/j.engappai.2025.110417}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110417}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generative adversarial network-based inverse design of self-deploying soft kirigami composites for targeted shape transformation}, volume = {149}, year = {2025}, }
• (2025). Image retrieval using deep saliency edge feature. EAAI, 149, 110416. (www)

  @article{ , abstract = {Exploiting a compact representation has become a hotspot in deep learning-based image retrieval. Fine-tuning the deep networks can provide the high discriminating representations, but it is very difficult to obtain sufficient labeled data. However, using the unsupervised methods to provide efficient representations remains challenging. Therefore, we propose a compact and robust representation, namely deep saliency edge feature (DSEF), to image retrieval. Its main highlights are: (1) Color differences, spatial layout, and edge cues within various object regions are combined into saliency edge feature maps. It can reflect a large amount of discriminative information contain in deep feature maps, thereby improving the discriminative power of deep features. (2) Edge cues are utilized to highlight the rough targets contained in deep feature maps. It can reduce the semantic disconnect exists in the different kinds of features and promote the compatibility of deep features and handcrafted features, thereby providing convenience to combine them. (3) A feature aggregation method, namely crucial cues aggregation, is proposed to aggregate crucial cues hidden inside handcrafted features and deep feature maps into a compact and high discriminating representation. Comparative experiments demonstrated that our method has provided the outstandingly retrieval performance on some benchmark datasets. The mean average precision of our method is 2.9%, 4.6%, 3.2%, 6.0% and 3.7% higher than that of most methods on the Oxford5K, Paris6K, Oxford105K, Paris106K and Holidays datasets, respectively.}, archive = {J_EAAI}, author = {Zhou Lu and Guang-Hai Liu and Zuo-Yong Li and Bo-Jian Zhang}, doi = {10.1016/j.engappai.2025.110416}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110416}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Image retrieval using deep saliency edge feature}, volume = {149}, year = {2025}, }
• (2025). SlimDL: Deploying ultra-light deep learning model on sweeping robots. EAAI, 149, 110415. (www)

  @article{ , abstract = {Advanced object detection methods have yielded impressive progress in recent years. However, the computational constraints of edge mobile devices present significant deployment challenges for state-of-the-art algorithms. We propose a deep learning deployment framework with two stages: model adaptation and compression. Our method enhance “You Only Look Once version 5” (YOLOv5) with lightweight modules, which improves detection performance while reducing computational load. Additionally, we present a pruning algorithm, employing adaptive batch normalization and iterative pruning. Our evaluation on “Microsoft Common Objects in Context” (MSCOCO) dataset and custom SweepRobot datasets demonstrates that our method consistently outperforms state-of-the-art approaches. On the SweepRobot dataset, our method doubled YOLOv5’s detection speed on the sweeping robot from 15.69 frames per second (FPS) to 30.77 FPS, maintaining 97.3% performance at 20% of the computational cost. Even on Graphics Processing Unit equipped devices, our method achieved 1.8% and 2.8% higher Average Precision compared to direct scaling and pruning with the original pruning algorithm.}, archive = {J_EAAI}, author = {Xudong Sun and Yu Wang and Zhanglin Liu and Shaoxuan Gao and Wenbo He and Chao Tong}, doi = {10.1016/j.engappai.2025.110415}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110415}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SlimDL: Deploying ultra-light deep learning model on sweeping robots}, volume = {149}, year = {2025}, }
• (2025). A novel group decision-making method for incomplete interval-valued intuitionistic multiplicative linguistic preference relations. EAAI, 149, 110412. (www)

  @article{ , abstract = {By conducting pairwise comparisons, decision-makers can construct interval-valued intuitionistic multiplicative linguistic preference relations (IVIMLPRs) to express the asymmetrically uncertain preferred and non-preferred qualitative judgments. Based on the consistency and consensus analysis, this paper proposes a new group decision-making (GDM) method with incomplete IVIMLPRs. Firstly, a reasonable and rational concept for IVIMLPR is defined. Inspired by the consistent intuitionistic multiplicative linguistic preference relations (IMLPRs), the consistency of IVIMLPRs is expressed by considering the corresponding lower and upper IMLPRs. After that, the acceptably consistent IVIMLPR is further introduced. Based on these concepts, two optimization models are constructed to estimate the missing linguistic variables and adjust an unacceptably consistent IVIMLPR, respectively. To obtain the priority weights from IVIMLPR in a reliable way, the consistency modeling method is employed. Before calculating the collective IVIMLPR, the weights of decision-makers are determined. Subsequently, the consensus analysis is conducted. If the consensus of an IVIMLPR is insufficient, a mathematical model is established to enhance the consensus level. Finally, the applications of the proposed GDM approach are offered and the comparative analysis is discussed. Compared with some existing methods, the proposed decision-making algorithm can perform a rational and effective process in the field of artificial intelligence computing.}, archive = {J_EAAI}, author = {Tao Li and Liyuan Zhang}, doi = {10.1016/j.engappai.2025.110412}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110412}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel group decision-making method for incomplete interval-valued intuitionistic multiplicative linguistic preference relations}, volume = {149}, year = {2025}, }
• (2025). High frequency volatility forecasting and risk assessment using neural networks-based heteroscedasticity model. EAAI, 149, 110397. (www)

  @article{ , abstract = {High frequency volatility forecasting is essential for timely risk management and informed decision-making in dynamic financial markets. However, accurate forecasting is challenging due to the rapid nature of market movements and the complexity of underlying economic factors. This paper introduces a novel architecture combining Generalized Autoregressive Conditional Heteroscedasticity (GARCH) and Multi-layer Perceptron (MLP)-based models for enhanced volatility forecasting and risk assessment, where input variables are processed through GARCH-type models for volatility forecasting. The proposed GARCH-based MLP-Mixer (GaMM) model incorporates the stacking of multi-layer perceptrons, enabling deep representation learning, facilitating the extraction of temporal and feature information through operations along both time and feature dimensions, and addressing the complexity of high-frequency time-series data. The proposed model is evaluated on three high frequency financial times series datasets over three different years. The computational results demonstrate the proposed model’s superior performance over sixteen forecasting methods in three error metrics, Value-at-risk, and statistical tests for high frequency volatility forecasting and risk assessment tasks.}, archive = {J_EAAI}, author = {Aryan Bhambu and Koushik Bera and Selvaraju Natarajan and Ponnuthurai Nagaratnam Suganthan}, doi = {10.1016/j.engappai.2025.110397}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110397}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {High frequency volatility forecasting and risk assessment using neural networks-based heteroscedasticity model}, volume = {149}, year = {2025}, }
• (2025). Reinforcement learning based multi-perspective motion planning of manned electric vertical take-off and landing vehicle in urban environment with wind fields. EAAI, 149, 110392. (www)

  @article{ , abstract = {Electric vertical-takeoff and landing (eVTOL) aircraft, known for their maneuverability and flexibility, offer a promising alternative to traditional transportation systems. However, these aircraft face significant challenges from various perspectives, including the need to increase energy efficiency, enhance passenger experience, and mitigate noise impact on urban environments. While mathematical modeling-based approaches have been employed for flight motion planning, they often struggle to adapt to dynamic and complex environments. In this work, we introduce a three-dimensional motion planning method based on deep reinforcement learning (DRL), tailored for manned eVTOL flights through urban wind fields. Our approach considers three crucial aspects: aircraft energy consumption, passenger experience, and noise impact on urban environment. We modify the Proximal Policy Optimization (PPO) algorithm and design comprehensive reward function that considers these objectives. By incorporating energy efficiency, passenger experience, and noise impact into our reward function, our method demonstrates improved policy learning compared to existing approaches. Comparative experiments conducted under various wind conditions show that our method outperforms commonly used techniques, effectively optimizing multiple objectives in challenging urban environments. Code of our work are available at https://github.com/cgchrfchscyrh/eVTOL_RL/tree/main.}, archive = {J_EAAI}, author = {Songyang Liu and Weizi Li and Haochen Li and Shuai Li}, doi = {10.1016/j.engappai.2025.110392}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110392}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reinforcement learning based multi-perspective motion planning of manned electric vertical take-off and landing vehicle in urban environment with wind fields}, volume = {149}, year = {2025}, }
• (2025). Leverage points in the bitcoin ecosystem: Impact on price and climate change using dynamic systems and spherical fuzzy sets. EAAI, 149, 110390. (www)

  @article{ , abstract = {With the significant increase in Bitcoin's price to around $108,000 in 2025 and its growth of 100% in 2024, scrutiny of the Bitcoin ecosystem and its influencing factors is of particular importance. The research aims to provide a comprehensive model for analyzing the Bitcoin ecosystem in forecasting the factors affecting it.This research used the Total Interpretive Structural Modeling method (TISM) to come up with ideas, the Spherical Fuzzy Decision Making Trial And Evaluation method (SF-DEMATEL) to find and look into how the indices related to each other, and the Spherical Fuzzy Step-Wise Weight Assessment Ratio Analysis method (SF-SWARA) to figure out how important the indices were. The findings identify control rules (0.222) and network security (0.196) as the most fundamental factors.Also, price (0.107) and Bitcoin value (0.089) are the most important among the influencing factors, while other cryptocurrencies have a lesser impact on the Bitcoin ecosystem (0.037). Analyses conducted using System Dynamics (SD) modeling showed that policies that increase control rules or improve tool quality by 50% can directly affect Bitcoin's price and climate change. The leverage point's analysis showed that stakeholders can contribute to the long-term sustainability of the Bitcoin ecosystem by strategically managing control rules, tool quality, network capacity, and turnover. The research presents a hybrid approach (SF & SD) to analyze the complexities and uncertainties of the Bitcoin ecosystem.The research implications can aid market participants in comprehending the Bitcoin ecosystem, forecasting Bitcoin prices, and assisting policymakers in implementing suitable regulations for the sustainable growth of this sector.}, archive = {J_EAAI}, author = {Saeed Alinejad and Zahra Khoshsepehr and Javad Nazarian-Jashnabadi and Samira Ebrahimi}, doi = {10.1016/j.engappai.2025.110390}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110390}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Leverage points in the bitcoin ecosystem: Impact on price and climate change using dynamic systems and spherical fuzzy sets}, volume = {149}, year = {2025}, }
• (2025). Influence factor-based transformation method for translating mass function to probability in Dempster–Shafer evidence theory. EAAI, 149, 110385. (www)

  @article{ , abstract = {Dempster–Shafer evidence theory provides an effective mathematical tool to represent uncertain information by assigning information into power set. Among its associated studies, a pivotal challenge is the transformation of mass functions into probability distributions which can enhance the robustness and reliability of decision-making. In this paper, influence factor is constructed by considering the impact of transformation between multi-element propositions and single-element propositions. Then based on influence factor, the novel transformation method is proposed. In addition, some numerical examples are used to explain effectiveness of new method by analyzing the probability information capacity of different methods. Finally, this paper applies the novel method to target recognition and validates its effectiveness as well as its enhanced support for decision-making through the utilization of real-world datasets.}, archive = {J_EAAI}, author = {Haocheng Shao and Lipeng Pan and Jiahui Chen and Xiaozhuan Gao and BingYi Kang}, doi = {10.1016/j.engappai.2025.110385}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110385}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Influence factor-based transformation method for translating mass function to probability in Dempster–Shafer evidence theory}, volume = {149}, year = {2025}, }
• (2025). REDIBAGG: Reducing the training set size in ensemble machine learning-based prediction models. EAAI, 149, 110382. (www)

  @article{ , abstract = {Machine learning-based algorithms have gained wide acceptance over the years due to their high generalization capabilities for a wide range of classification applications. Although these algorithms demonstrate potential and promising performance, they are often limited by speed, particularly when training on large databases. Big data sets with many instances have storage requirements and execution times that can be excessive. This study proposes reducing the sample size generated by the bootstrap resampling method in Ensemble Machine Learning-based models and evaluates their generalization capability on unknown data. Reduced bootstrap samples are employed in the training phase of the Bagging ensemble model. This approach reduces execution times and, consequently, storage requirements. The proposed method was tested on classification tasks, effectively reducing training subset size without compromising performance. Experimental results demonstrate that this approach achieves execution times reductions of up to 70% for some data sets. This reduction has no impact on accuracy, whereas maintaining levels comparable to classical Bagging and its variants. On average, the training subset size was reduced by 25% compared to the original size.}, archive = {J_EAAI}, author = {Esther-Lydia Silva-Ramírez and Juan-Francisco Cabrera-Sánchez and Manuel López-Coello}, doi = {10.1016/j.engappai.2025.110382}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110382}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {REDIBAGG: Reducing the training set size in ensemble machine learning-based prediction models}, volume = {149}, year = {2025}, }
• (2025). Integrating permutation feature importance with conformal prediction for robust explainable artificial intelligence in predictive process monitoring. EAAI, 149, 110363. (www)

  @article{ , abstract = {As artificial intelligence (AI) systems are increasingly deployed in high-stakes environments, the need for explanations that convey uncertain information has become evident. Conventional explainable AI (XAI) methods often overlook uncertainty, focusing solely on point predictions. To address this gap, we propose using permutation feature importance (PFI) combined with predictive uncertainty evaluation measures. This novel approach examines the significance of features by relating them to the model’s confidence in its predictions. By using split conformal prediction (SCP) to quantify predictive uncertainty and integrating the outcomes to PFI, we aim to enhance the robustness and interpretability of machine learning (ML) algorithms. More importantly, we examine three scenarios for conformal prediction-based PFI explanations: permuting feature values in the test data, the calibration data, and both. These scenarios assess the impact of feature permutations from different perspectives, revealing feature sensitivity and the importance of features in various settings. We also perform a series of sensitivity analyses, particularly exploring calibration data size and computational efficiency, to demonstrate the robustness and scalability of our approach for industrial applications. Our comprehensive evaluation offers insights into feature impact on predictions and their associated confidence levels. We validate our proposed approach through a real-world predictive process monitoring use case in manufacturing.}, archive = {J_EAAI}, author = {Nijat Mehdiyev and Maxim Majlatow and Peter Fettke}, doi = {10.1016/j.engappai.2025.110363}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110363}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating permutation feature importance with conformal prediction for robust explainable artificial intelligence in predictive process monitoring}, volume = {149}, year = {2025}, }
• (2025). A robust sparse identification of nonlinear dynamics approach by combining neural networks and an integral form. EAAI, 149, 110360. (www)

  @article{ , abstract = {One widely used methodology for uncovering governing equations from data is sparse regression for nonlinear dynamics, commonly known as Sparse Identification of Nonlinear Dynamics ( SINDy ). However, noisy and limited data remain a significant challenge for the success of the SINDy approach. In this work, we propose a robust strategy to discover nonlinear governing equations from both noisy and scarce data. Specifically, we employ neural networks to learn an implicit representation from measurement data, thereby ensuring that the network output remains close to the measurements while also admitting a dynamical system interpretation for its time evolution. Moreover, we identify this dynamical system in the spirit of the SINDy framework. By leveraging the neural network’s implicit representation, we employ automatic differentiation to obtain the derivative information required by SINDy . To further enhance the robustness of our approach, we incorporate an integral constraint on the output of the implicit networks. In addition, we extend our method to handle data acquired from multiple initial conditions. Through several examples, we demonstrate the proposed method’s effectiveness in discovering governing equations under noisy, data-scarce conditions and compare its performance against existing methods.}, archive = {J_EAAI}, author = {Ali Forootani and Pawan Goyal and Peter Benner}, doi = {10.1016/j.engappai.2025.110360}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110360}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust sparse identification of nonlinear dynamics approach by combining neural networks and an integral form}, volume = {149}, year = {2025}, }
• (2025). Cross-domain facial expression recognition: Bi-directional fusion of active and stable information. EAAI, 149, 110357. (www)

  @article{ , abstract = {Facial expression recognition (FER) algorithms often encounter obstacles in cross-domain scenarios, attributed to variations in collection conditions such as lighting, weather, age, gender, and skin color of subjects. Unlike existing approaches that primarily focus on extracting globally invariant features and aligning domain distributions, we propose a novel framework that fundamentally shifts the approach to cross-domain FER. Our proposed algorithm, termed Bi-Directional Fusion of Active and Stable Information (FER-DAS), uniquely combines three innovative components: the Active Assessment Strategy (AAS), Cross-Domain Dynamic Class Threshold (CD-DCT), and Weighted Cross-Domain Alignment (WCDA). The AAS component selectively identifies and enhances active samples in the target domain, providing precise annotations for improved model robustness. Samples with the highest uncertainty are deemed active, indicating low prediction confidence and high informational value for model training. These are then filtered using a predefined threshold to ensure only the most informative samples are included in training iterations. In contrast to conventional static threshold techniques, our dynamic class threshold strategy (CD-DCT) adaptively filters stable samples across domains, thereby ensuring that only the most reliable information is utilized in training. The WCDA strategy further refines this process by dynamically assessing and weighting the contribution of target domain samples to class centers, effectively mitigating domain distribution discrepancies. Extensive experiments on multiple benchmark datasets confirm that FER-DAS sets a new standard in cross-domain FER, consistently outperforming existing state-of-the-art methods.}, archive = {J_EAAI}, author = {Yanan Zhu and Jiaqiu Ai and Weibao Xue and Mingyang Wu and Sen Yang and Wei Jia and Min Hu}, doi = {10.1016/j.engappai.2025.110357}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110357}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-domain facial expression recognition: Bi-directional fusion of active and stable information}, volume = {149}, year = {2025}, }
• (2025). Depression and anxiety detection method based on serialized facial expression imitation. EAAI, 149, 110354. (www)

  @article{ , abstract = {Facial recognition techniques are widely employed for automatic detection of depression and anxiety. However, current studies overlook the impact of varying spatial resolutions on model performance and lack a mechanism to share attention regions across sequential data. To advance research in this area, we conducted the Voluntary Facial Expression Mimicry Experiment (VFEM) and constructed the VFEM dataset. We also introduce the SFE-Former, a sequential facial expression recognition model designed for detecting depression and anxiety. SFE-Former features a mechanism that shares attention regions across sequence data, allowing each data point to enhance its features by leveraging shared information. Additionally, the model integrates features from different scales using fusion and weighting strategies. The experimental results indicate that SFE-Former achieved impressive accuracy rate: 0.893 for depression detection, 0.889 for anxiety detection, and 0.780 for co-occurrence detection of depression and anxiety. Meanwhile, SFE-Former also obtained state-of-the-art (SOAT) results on AVEC2014 dataset. This work can enhance the accuracy of identifying patients with depression and anxiety, providing doctors with reliable auxiliary diagnosis. The source code for SFE-Former is accessible at https://github.com/lulin-6k/SFE-Former .}, archive = {J_EAAI}, author = {Lin Lu and Yan Jiang and Xingyun Li and Hao Wang and Qingzhi Zou and Qingxiang Wang}, doi = {10.1016/j.engappai.2025.110354}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110354}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Depression and anxiety detection method based on serialized facial expression imitation}, volume = {149}, year = {2025}, }
• (2025). Goal-oriented graph generation for transmission expansion planning. EAAI, 149, 110350. (www)

  @article{ , abstract = {The electrification strategies that are being designed to meet sustainability objectives and rising energy demands pose significant challenges for power systems worldwide and require Transmission Expansion Planning (TEP). This study adopts a risk-informed approach to TEP, formulated as a multi-objective optimization problem that concurrently minimizes systemic risks and expansion costs. Given the intractability of this problem with conventional solvers, we turn to artificial intelligence techniques. In particular, we conceptualize power grids as graphs and introduce a goal-oriented graph generation methodology using deep reinforcement learning. We extend welfare-Q learning, a modified variant of Q-learning tailored to yield high rewards across multiple dimensions, by incorporating geometric deep learning for function approximation. This allows us to account for system security while minimizing grid expansion costs. Notably, system risk is evaluated by incorporating a Graph Neural Network (GNN) cascading failure meta-model into the proposed approach. The TEP method is applied to the IEEE 118-bus system, and the efficacy of this novel technique is compared against the state of the art. We conclude that the deep reinforcement learning method can compete with established methods for multi-objective optimization, identifying expansion strategies that improve system security at reduced costs. Furthermore, we test the robustness of the meta-model against topology changes in the transmission network, demonstrating its applicability to novel grid configurations.}, archive = {J_EAAI}, author = {Anna Varbella and Blazhe Gjorgiev and Federico Sartore and Enrico Zio and Giovanni Sansavini}, doi = {10.1016/j.engappai.2025.110350}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110350}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Goal-oriented graph generation for transmission expansion planning}, volume = {149}, year = {2025}, }
• (2025). Dealing with zero-inflated data: Achieving state-of-the-art with a two-fold machine learning approach. EAAI, 149, 110339. (www)

  @article{ , abstract = {In many cases, a machine learning model must learn to correctly predict a few data points with particular values of interest in a broader range of data where many target values are zero. Zero-inflated data can be found in diverse scenarios, such as lumpy and intermittent demands, power consumption for home appliances being turned on and off, impurities measurement in distillation processes, and even airport shuttle demand prediction. The presence of zeroes affects the models’ learning and may result in poor performance. Furthermore, zeroes also distort the metrics used to compute the model’s prediction quality. This paper showcases two real-world use cases (home appliances classification and airport shuttle demand prediction) where a hierarchical model applied in the context of zero-inflated data leads to considerable performance improvements. In particular, for home appliances classification, the weighted average of Precision, Recall, F1, and Area Under the Receiver Operating Characteristic Curve (AUC ROC) was increased by 39%, 49%, 88%, and 48%, respectively. Furthermore, it is estimated that the proposed approach is also four times more energy efficient than the state-of-the-art (SOTA) approach against which it was compared to. Two-fold modeling approaches significantly outperform regular regression, especially when predicting the occurrence of demand events. SOTA results were achieved using Gradient Boosting trees to determine whether an event will occur and Visual Geometry Group (VGG) or Support Vector Regressor (SVR) models for the subsequent classification/regression. The code has been released at two separate repositories.}, archive = {J_EAAI}, author = {Jože M. Rožanec and Gašper Petelin and João Costa and Gregor Cerar and Blaž Bertalanič and Marko Guček and Gregor Papa and Dunja Mladenić}, doi = {10.1016/j.engappai.2025.110339}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110339}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dealing with zero-inflated data: Achieving state-of-the-art with a two-fold machine learning approach}, volume = {149}, year = {2025}, }
• (2025). Hybrid EfficientNet feed forward neural network for ransomware detection in blockchain. EAAI, 149, 110292. (www)

  @article{ , abstract = {Ransomware, a type of malware, threatens to encrypt or block access to data until a ransom is paid, causing severe financial and operational impacts. Detecting ransomware in blockchain environments is crucial due to its increasing prevalence and the need for robust security measures. This research proposes an EfficientNet Feed Forward Neural Network (EFFNN) model for ransomware detection in blockchain. Initially, sequence-based statistical features are extracted from blockchain data and normalized using linear normalization. Feature selection is performed with Kumar Hassebrook and Czekanowski’s similarity measures to enhance relevant feature identification. The Hybrid EFFNN, integrating EfficientNet and Deep Feed Forward Neural Networks (DFFNN), is employed for detection. The model achieved high performance, with a True Negative Rate (TNR) of 0.960, accuracy of 0.955, and True Positive Rate (TPR) of 0.821, demonstrating its efficacy in identifying ransomware in blockchain systems.}, archive = {J_EAAI}, author = {Balajee Maram and Neelima Gullipalli and Rudra Kalyan Nayak and Ramamani Tripathy and Satish Muppidi and Madan Lal Saini}, doi = {10.1016/j.engappai.2025.110292}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110292}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid EfficientNet feed forward neural network for ransomware detection in blockchain}, volume = {149}, year = {2025}, }
• (2025). Unlocking language barriers: Assessing pre-trained large language models across multilingual tasks and unveiling the black box with explainable artificial intelligence. EAAI, 149, 110136. (www)

  @article{ , abstract = {Large Language Models (LLMs) have revolutionized many industrial applications and paved the way for fostering a new research direction in many fields. Conventional Natural Language Processing (NLP) techniques, for instance, are no longer necessary for many text-based tasks, including polarity estimation, sentiment and emotion classification, and hate speech detection. However, training a language model for domain-specific tasks is hugely costly and requires high computational power, thereby restricting its true potential for standard tasks. This study, therefore, provides a comprehensive analysis of the latest pre-trained LLMs for various NLP-related applications without fine-tuning them to evaluate their effectiveness. Five language models are thus employed in this study on six distinct NLP tasks (including emotion recognition, sentiment analysis, hate speech detection, irony detection, offensiveness detection, and stance detection) for 12 languages from low- to medium- and high-resource. Generative Pre-trained Transformer 4 (GPT-4) and Gemini Pro outperform state-of-the-art models, achieving average F1 scores of 70.6% and 68.8% on the Tweet Sentiment Multilingual dataset compared to the state-of-the-art average F1 score of 66.8%. The study further interprets the findings obtained by the LLMs using Explainable Artificial Intelligence (XAI). To the best of our knowledge, it is the first time any study has employed explainability on pre-trained language models.}, archive = {J_EAAI}, author = {Muhamet Kastrati and Ali Shariq Imran and Ehtesham Hashmi and Zenun Kastrati and Sher Muhammad Daudpota and Marenglen Biba}, doi = {10.1016/j.engappai.2025.110136}, journal = {Engineering Applications of Artificial Intelligence}, month = {6}, pages = {110136}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unlocking language barriers: Assessing pre-trained large language models across multilingual tasks and unveiling the black box with explainable artificial intelligence}, volume = {149}, year = {2025}, }
• (2025). Explainable fault diagnosis method based on statistical features for gearboxes. EAAI, 148, 110503. (www)

  @article{ , abstract = {Machine learning techniques, known for their high accuracy, have been extensively applied in gearbox fault diagnosis. However, the machine learning methods used for gearbox fault diagnosis primarily focus on diagnostic accuracy, and their explainability during the diagnostic process remains a significant challenge. In response to this issue, this paper proposes a novel explainable fault diagnosis method. The method comprises statistical feature refinement, boosting trees, and the Shapley additive explanations interpreter. Specifically, the approach initially filters fault features using the diagnostic accuracy and diagnostic contribution of simple machine learning methods to reduce the complexity of the model explainability analysis. Subsequently, various boosting tree models are trained, and their contributions are calculated, with the model exhibiting the highest F1-score selected for the explainability analysis of the diagnostic process. Finally, after selecting the model, we conduct tests and perform global and local attribution analysis, feature analysis, and misclassified sample analysis based on feature contributions. This analysis elucidates the causal relationship between the statistical features extracted from vibration signals and gear fault states. Experimental results demonstrate that there is a strong coupling and highly nonlinear relationship between gear vibration signals and fault states. The proposed method not only achieves high diagnostic accuracy but also decouples the strong coupling between gear faults and their corresponding vibration features. Most importantly, it elucidates the nonlinear mapping relationship between these faults and gear vibration features. Additionally, the misclassified sample analysis method presented in this paper can eliminate redundant sensor signals, thereby enhancing the model's attention to critical signals.}, archive = {J_EAAI}, author = {Huakang Tang and Honglei Wang and Chengjiang Li}, doi = {10.1016/j.engappai.2025.110503}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110503}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable fault diagnosis method based on statistical features for gearboxes}, volume = {148}, year = {2025}, }
• (2025). Parallel segmentation network for real-time semantic segmentation. EAAI, 148, 110487. (www)

  @article{ , abstract = {Real-time semantic segmentation holds extensive application prospects in autonomous driving and robot navigation. Recently, real-time semantic segmentation networks mainly adopt encoder-decoder architecture and multi-branch architecture. However, both approaches have their own advantages and limitations. Encoder-decoder models are generally better at extracting contextual information, but may face challenges in capturing fine details and local spatial information. On the other hand, the multi-branch structure excels at capturing boundary and spatial detail information, but it requires an efficient and flexible feature fusion strategy to prevent information redundancy. To leverage the strengths of both approaches, we propose a Parallel Segmentation Network (PaSeNet) which adopts the unsymmetrical encoder-decoder structure to introduce novel ideas for research and applications in real-time semantic segmentation. Specifically, we design a main branch with a spatial information enhancement path during the encoding phase and introduce mask autoencoder based on self-supervised learning as an auxiliary branch to supplement the main branch in extracting details as well as local spatial information. Additionally, we propose the Grouped Aggregation Pyramid Pooling Module to optimize the extraction of contextual information. In the decoding phase, we introduce the Coordinate-Attention-Guided Decoder to effectively integrate diverse information from different branches. A large number of experiments on the Cityscapes, Cambridge-driving Labeled Video database (CamVid), NightCity and instance Segmentation in Aerial Images Dataset demonstrate that our method achieves competitive results. Specifically, PaSeNet-Base obtains 79.9% mean Intersection Over Union (mIOU) at 55.6 Frames Per Second (FPS) on Cityscapes test dataset and 80.2% mIOU at 96.8 FPS on CamVid test dataset.}, archive = {J_EAAI}, author = {Guanke Chen and Haibin Li and Yaqian Li and Wenming Zhang and Tao Song}, doi = {10.1016/j.engappai.2025.110487}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110487}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Parallel segmentation network for real-time semantic segmentation}, volume = {148}, year = {2025}, }
• (2025). Hardware-in-the-loop verification and real-time simulation of innovative fuzzy super-twisting control technique of doubly-fed induction generators. EAAI, 148, 110484. (www)

  @article{ , abstract = {The choice of energy command for a doubly-fed induction generator (DFIG) is essential to ensure high energy fineness and good operation. Direct power command (DPC) is one of the approaches used to command power due to its simplicity, ease of realization, and quick dynamic response. However, despite these advantages, the DPC is described by problems that hinder its spread, such as reduced durability if the DFIG parameters change. To augment the competence of the DPC of a DFIG, a new algorithm is proposed that relies on combining both fuzzy logic (FL) and the super-twisting algorithm (STA) to command the DFIG powers, where two FL-STA regulators are used to replace traditional DPC controllers. Also, the pulse width modulation approach converts the reference values created by the FL-STA controllers into pulses to run the machine inverter. The suggested approach is a modification of the DPC, where simplicity, robustness, high competence, and ease of realization are among its most important features. First, the designed algorithm was realized using MATLAB and compared to the DPC, where some tests were used to study the performance and robustness of this method. The simulation results showed the advantage of the DPC-FL-STA over the DPC, as the current harmonics were minimized by rates estimated at 77.98 %, 79.26 %, and 80.50 % in all tests. In addition, the overshoot value of active power has been improved by 27.42 %, 14.55 %, and 22.39 % compared to the DPC. Secondly, the results were verified using empirical work based on Hardware-in-the-loop testing. In the empirical work, two different wind speed profiles were used to study the competence and robustness of the DPC-FL-STA compared to the DPC. The results of the experimental work confirm the results and the efficacy of the DPC-FL-STA in enhancing the fineness of stream and power. These results make the DPC-FL-STA a promising solution for the future.}, archive = {J_EAAI}, author = {Mourad Yessef and Habib Benbouhenni and Ahmed Lagrioui and Hatim Ameziane and Nicu Bizon and Badre Bossoufi and Ayman Alhejji}, doi = {10.1016/j.engappai.2025.110484}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110484}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hardware-in-the-loop verification and real-time simulation of innovative fuzzy super-twisting control technique of doubly-fed induction generators}, volume = {148}, year = {2025}, }
• (2025). A residual deep learning framework for sleep apnea diagnosis from single lead electrocardiogram signals: An explainable artificial intelligence approach. EAAI, 148, 110481. (www)

  @article{ , abstract = {Sleep apnea is a common sleep disorder that occurs due to repetitive obstruction of the airflow in human body and affect human lives. Conventional machine and deep learning-based methods hardly have the transparency in the decision making and work as a “black-box”. To alleviate this problem, this article aims to present an explainable deep learning framework for diagnosis of sleep apnea using single channel electrocardiogram signals. The recorded electrocardiogram signals are preprocessed and converted into two-dimensional time-frequency images via continuous wavelet transform for necessary feature extraction followed by classification via residual neural network. The prime motive to employ time-frequency representation is to produce temporal and spectral information simultaneously. Also, time-frequency graphs are beneficial to analyze and represent non stationary signals possessing multiple time varying frequencies. The contributions of the proposed research are threefold: Firstly, the proposed framework not only diagnose sleep apnea in terms of classification accuracy, but also obtain the graphical explanation in a supervised manner. Secondly, relevance maps, generated through layer wise relevance propagation, are analyzed at each convolutional layer for multiclass sleep apnea diagnosis. Thirdly, an optimal explainable deep learning framework has been proposed to yield a fine-tuned and precise time frequency plot. Substantial experiments reveal that the proposed method achieves a significant accuracy of 98.16%, outperforming state-of-the-art sleep apnea diagnosis methods, along with specificity 99.05% and sensitivity 97.64%. Moreover, the method yields an accuracy of 97.52 % in another database. The presented topology can also be further used to other medical modalities in different biomedical systems.}, archive = {J_EAAI}, author = {Biswarup Ganguly and Rajobrata Dasgupta and Debangshu Dey}, doi = {10.1016/j.engappai.2025.110481}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110481}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A residual deep learning framework for sleep apnea diagnosis from single lead electrocardiogram signals: An explainable artificial intelligence approach}, volume = {148}, year = {2025}, }
• (2025). Decoupled graph spatial-temporal transformer networks for traffic flow forecasting. EAAI, 148, 110476. (www)

  @article{ , abstract = {Recent research has extensively applied the Transformer architecture to traffic flow prediction, proposing various solutions to effectively capture the nonlinear spatial-temporal correlations of traffic data. To better model traffic flow data, some works have highlighted the need to simultaneously consider the spatial-temporal heterogeneity during the learning of spatial-temporal correlations of traffic flow data. However, none of the existing methods have successfully achieved the goal of effectively capturing spatial-temporal heterogeneity while learning spatial-temporal correlations. To address these challenges, we propose a model for traffic flow prediction, named Decoupled Graph Spatial-Temporal Transformer Networks (DSTTN). The core of this model is a Spatial-Temporal Decoupled Representation Learning module designed to decouple spatial-temporal embedding and apply it to Transformer Networks, integrating with the Multi-Head Attention Mechanism. Through the Spatial-Temporal Decoupled Representation Learning Attention Mechanism, the model effectively achieves spatial-temporal decoupled learning of input feature embedding after spatial-temporal embedding structure embedding, thereby enhancing the ability of the model to learn spatial-temporal features for representation learning. This enables the model to extract spatial-temporal correlations and model spatial-temporal heterogeneity based on the Graph Transformer model. Experiments conducted on six real-world traffic datasets demonstrate that our model outperforms twelve comparative models, validating the ability of the model to perform more accurate prediction tasks. Additionally, ablation experiments confirm the effectiveness of the proposed spatial-temporal decoupling structure based on Transformer.}, archive = {J_EAAI}, author = {Wei Sun and Rongzhang Cheng and Yingqi Jiao and Junbo Gao}, doi = {10.1016/j.engappai.2025.110476}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110476}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Decoupled graph spatial-temporal transformer networks for traffic flow forecasting}, volume = {148}, year = {2025}, }
• (2025). Investigate dual chiller characteristics and improve the performance by integrating machine learning and genetic algorithms. EAAI, 148, 110472. (www)

  @article{ , abstract = {In modern buildings, the chiller has become the primary source of energy consumption. Improving chiller performance not only represents an opportunity to mitigate the environmental impact but also to reduce operational costs. Significant benefits are offered to both the building sector and society. This paper investigates the characteristics of a dual-chiller system and proposes performance improvements. This investigation utilized a dataset spanning nearly two years from a hospital in Taiwan. Machine learning is employed to predict the total chiller system power and perform weather clustering. The prediction models achieve an R 2 exceeding 0.9, meeting our minimum acceptance criteria. Using the k-means algorithm, outside air temperature and humidity are categorized into five types, integrated into a Genetic Algorithm (GA) for chiller parameter optimization. The GA focuses on reducing total power consumption, optimizing the part-load ratio (PLR) and operating parameters, where the system load percentage ranging from 12.5% to 80%. Evaluation against the GA results shows an average 11% reduction in total power consumption ( P t o t ) and a 13.8% improvement in chiller plant coefficient of performance ( C O P c p ).}, archive = {J_EAAI}, author = {Elsa Chaerun Nisa and Yean-Der Kuan and Chin-Chang Lai}, doi = {10.1016/j.engappai.2025.110472}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110472}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Investigate dual chiller characteristics and improve the performance by integrating machine learning and genetic algorithms}, volume = {148}, year = {2025}, }
• (2025). A deep learning-based adaptive denoising approach for fine identification of rock microcracks from noisy strain data. EAAI, 148, 110471. (www)

  @article{ , abstract = {Most of the existing deep learning approaches about crack identification are difficult to obtain a satisfactory result when confronting highly noisy distributed fibre optic sensing (DFOS) data. To address this limitation, this research develops a hybrid attention residual shrinkage network (HARSNet) to enhance feature extraction ability from highly noisy DFOS data and achieve a high rock (micro) crack identification accuracy. Considering that it is challenging to set appropriate threshold values to eliminate data noise, a hybrid attention module is developed as trainable modules for the HARSNet to automatically and adaptively capture the thresholds relevant to data noise, so that the professional expertise on threshold determination is not required. Then a soft thresholding layer embedded in the HARSNet uses the captured thresholds to automatically eliminate noise to make the crack-related features more discriminative. The effectiveness of the proposed HARSNet in improving crack identification accuracy is examined through experimental comparisons with the other five state-of-the-art deep learning models. Results indicate that the proposed HARSNet outperforms the five deep learning models by yielding a maximal accuracy improvement of 12.9% on the highly noisy DFOS dataset with a signal-to-noise ratio of 0, which is satisfying in rock (micro) crack identification from highly noisy DFOS data.}, archive = {J_EAAI}, author = {Shuai Zhao and Dian-Rui Mu and Dao-Yuan Tan}, doi = {10.1016/j.engappai.2025.110471}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110471}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep learning-based adaptive denoising approach for fine identification of rock microcracks from noisy strain data}, volume = {148}, year = {2025}, }
• (2025). Strength and durability predictions of ternary blended nano-engineered high-performance concrete: Application of hybrid machine learning techniques with bio-inspired optimization. EAAI, 148, 110470. (www)

  @article{ , abstract = {Higher-strength, longer-lived, and environmentally friendly high-performance concrete (HPC) necessitates advanced formulation of materials and predictive modeling techniques. The traditional designs of high-performance concrete often fail to optimize the mix parameters suitably, mainly from the point of sustainability view, and lack the capacity to model complex nonlinear relationships among material properties and performance metrics. The current paper introduces the integrated framework of nano-engineered materials that comprises nano-silica, graphene oxide (GO), and carbon nanotubes (CNTs) along with advanced artificial intelligence (AI) methods in the predictive modeling and optimization of high-performance concrete (HPC). The Machine Learning (ML) models consisting of Random Forest (RF), Gradient Boosting Machines (GBM), and Deep Neural Networks (DNNs) were implemented to predict compressive and tensile strengths, elastic modulus, and durability metrics such as chloride penetration and sulfate resistance sets. Dimensionality reduction techniques, such as Principal Component Analysis (PCA) and Partial Least Squares (PLS), improved model interpretability and performance sets. Optimization of mix designs was carried out using bio-inspired algorithms, namely Genetic Algorithms (GA) and Particle Swarm Optimization (PSO), to balance mechanical performance, material efficiency, and environmental sustainability. The results suggest strength improvement up to 15% and durability up to 30% while reducing cement content and associated carbon emissions. This is the framework, which exploits artificial intelligence techniques in predictive modeling and optimization to achieve scalable and sustainable next-generation high-performance concrete designs. It is at the intersection of theoretical potential and practical application that the work is advancing the role of artificial intelligence in construction material science toward contributing to sustainable infrastructure development process.}, archive = {J_EAAI}, author = {Vikrant S. Vairagade and Boskey V. Bahoria and Haytham F. Isleem and Nilesh Shelke and Nischal P. Mungle}, doi = {10.1016/j.engappai.2025.110470}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110470}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Strength and durability predictions of ternary blended nano-engineered high-performance concrete: Application of hybrid machine learning techniques with bio-inspired optimization}, volume = {148}, year = {2025}, }
• (2025). Evaluation of medium-lift forest fire helicopter using q-rung orthopair fuzzy set based alternative ranking technique based on adaptive standardized intervals approach. EAAI, 148, 110468. (www)

  @article{ , abstract = {The use of helicopters in combating forest fires is critical due to their ability to rapidly access remote and rugged terrains, enabling timely interventions that significantly curb fire spread. Additionally, helicopters can deliver substantial amounts of water and firefighting agents directly to affected areas, enhancing suppression efforts while minimizing damage to ecosystems and nearby communities. The main objective of this study is to evaluate the forest fire helicopter fleet by developing a decision model to prioritize helicopters for more effective firefighting. The motivation of the study is to propose a systematic approach that ensures the optimal utilization of resources in fire management. A q-Rung Orthopair Fuzzy Set based Alternative Ranking Technique based on Adaptive Standardized Intervals (ARTASI) method is introduced as an innovative framework for evaluating medium-lift forest fire helicopters. The methodology is applied in a case study involving seven helicopters domestically produced, leased, or designated as reserve assets assessed based on firefighting-specific criteria. The results highlight cost and water-carrying capacity as critical evaluation factors. Among the alternatives, the “Ka-32” helicopter is identified as the most suitable, reflecting Turkey's successful efforts to enhance its firefighting capabilities with models like the “T-70”. Sensitivity analyses, including variations in criteria weights, the q parameter, and alternative decision-making methods, confirm the robustness and applicability of the proposed approach for helicopter selection in forest fire management.}, archive = {J_EAAI}, author = {Hakan Ayhan Dağıstanlı and Cevriye Temel Gencer}, doi = {10.1016/j.engappai.2025.110468}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110468}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evaluation of medium-lift forest fire helicopter using q-rung orthopair fuzzy set based alternative ranking technique based on adaptive standardized intervals approach}, volume = {148}, year = {2025}, }
• (2025). Multi-fidelity sub-label-guided transfer network with physically interpretable synthetic datasets for rotor fault diagnosis. EAAI, 148, 110467. (www)

  @article{ , abstract = {The development of deep-learning-based diagnostic models for prognostics and health management relies on data from the target system. However, when the system does not exist (e.g., during the design phase), acquiring health data is impossible, making accurate fault diagnosis models difficult to develop. To resolve this challenge, this study proposes a fault diagnosis framework for rotating machinery that uses physically interpretable synthetic data generation and sub-label-guided transfer learning. An empirical model based on domain knowledge is used to generate large quantities of low-cost, low-fidelity data, although system configuration and expected operating condition information are incorporated into multi-body dynamics simulations to create a small amount of accurate, high-fidelity data. Low-fidelity data are labeled with a primary label that represents dominant fault modes and with sub-labels corresponding to detailed categories of faults, whereas high-fidelity data are assigned only a primary label for the dominant fault mode. Multi-fidelity data are preprocessed to create two-dimensional orbit plots and one-dimensional handcrafted features, which are then used as inputs for the transfer learning model. To effectively train the model, while considering the correlation between multi-fidelity data, this study designs a sub-label-guided transfer network (SlgTN) that combines sub-label guidance with transfer learning. In the pretraining process, feature learning is performed using low-fidelity data, with two parallel classification layers designed to effectively learn both the primary label and the sub-label. Center loss and cross entropy loss are incorporated. Experimental results from applying the method to a Bently-Nevada rotor kit demonstrate the effectiveness and validity of the proposed method.}, archive = {J_EAAI}, author = {Dongmin Lee and Jun Gyu Lee and Minseok Choi and Cheonha Park and Chang Wan Kim and Gang Niu and Hyunseok Oh}, doi = {10.1016/j.engappai.2025.110467}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110467}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-fidelity sub-label-guided transfer network with physically interpretable synthetic datasets for rotor fault diagnosis}, volume = {148}, year = {2025}, }
• (2025). A deep neural network for small object detection in complex environments with unmanned aerial vehicle imagery. EAAI, 148, 110466. (www)

  @article{ , abstract = {Deep learning-based object detectors perform effectively on edge devices but encounter challenges with small and flat objects in complex environments, especially under low-light conditions and in high-altitude images captured by unmanned aerial vehicles (UAVs). The primary issue is the pixel similarity between objects and their backgrounds, making detection challenging. While existing detectors struggle to detect small and flat objects in these scenarios, the advent of you only look once (YOLO) algorithms have shown promise. However, they still have limitations in detecting small and flat objects under these conditions. Due to a shortage of suitable datasets covering complex environments and lighting conditions, the field lacks comprehensive research on detecting small and flat objects in UAV-assisted images. To address these issues, we develop a dataset with nine classes tailored to small object detection (SOD) challenges. We propose a dynamic model based on the you only look once network v5 (version 6.2) architecture to overcome the above-mentioned limitations. We introduce the Luna-enhancement mechanism and four novel modules, which enhance the detector's capacity to detect objects in complex environments. Our approach aims to improve the accuracy and robustness of detecting small and flat objects in complex environments, benefiting applications like aerial surveillance, search and rescue, and autonomous navigation. The experimental results demonstrate that our proposed model achieves a mean average precision (mAP_0.5) of 74.8% on the common objects in context (COCO) dataset, 76.3% on the VisDrone2019 dataset, 90.6% on the dataset for object detection in aerial images (DOTA-v1.5) dataset, and 71.5% on our SOD-Dataset, with improvements of 7.7%, 6.9%, 4.4% and 10.9%, respectively. For mAP_0.5:0.95, the model achieves 57.2%, 58.2%, 68.2%, and 51.7% on the COCO, VisDrone2019, DOTA-v1.5, and SOD-Dataset, with improvements of 5.5%, 16.4%, 3.4%, and 12.1% compared to the baseline algorithm. Furthermore, ablation experiments and visualization analysis provide additional evidence of the importance of each model component. The code and dataset are publicly available at https://github.com/dhuvisionlab/YOLO-SOD .}, archive = {J_EAAI}, author = {Sayed Jobaer and Xue-song Tang and Yihong Zhang}, doi = {10.1016/j.engappai.2025.110466}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110466}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep neural network for small object detection in complex environments with unmanned aerial vehicle imagery}, volume = {148}, year = {2025}, }
• (2025). A hybrid machine learning approach for predicting fiber-reinforced polymer-concrete interface bond strength. EAAI, 148, 110458. (www)

  @article{ , abstract = {The interfacial bond strength between fiber-reinforced polymer (FRP) sheets and concrete is crucial for structural design. This study presented a novel approach using ensemble learning models to predict bond strength and analyze input parameters' influence. No previous research used gene expression programming (GEP) for developing bond strength models in single shear tests. This research introduced GEP to develop an expression for estimating bond strength, comparing its performance with existing empirical models used in design codes. Six ensemble machine learning models were tested: extreme gradient boosting (XGBoost), light gradient boosting (LightGBM), categorical boosting (CatBoost), adaptive boosting (AdaBoost), random forest (RF), and explainable boosting machine (EBM), using 855 samples. CatBoost demonstrated superior performance with R2 = 0.98, RMSE = 1.61 kN, and MAE = 1.18 kN. The study utilized EBM's interpretability for parametric analysis through local and global explanations. Results showed FRP material and geometric properties had greater impact on bond strength than concrete properties. The novel GEP-developed empirical expression achieved higher accuracy compared to existing empirical models, with R2 = 0.812, RMSE = 4.63 kN, and MAE = 3.58 kN. The GEP model primarily relied on FRP's material and geometric properties, aligning with parametric analysis findings. Based on the results, both the CatBoost ensemble learning model and GEP model are recommended for estimating FRP-concrete interfacial bond strength.}, archive = {J_EAAI}, author = {Sarmed Wahab and Babatunde Abiodun Salami and Hassan Danish and Saad Nisar and Ali H. AlAteah and Ali Alsubeai}, doi = {10.1016/j.engappai.2025.110458}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110458}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid machine learning approach for predicting fiber-reinforced polymer-concrete interface bond strength}, volume = {148}, year = {2025}, }
• (2025). Residual-like multi-kernel block and dynamic attention for deep neural networks. EAAI, 148, 110456. (www)

  @article{ , abstract = {Traditional network architectures struggled with a uniform approach to receptive field (RF) sizes, leading to suboptimal performance across scales. Although recent advances have addressed the problem by utilizing different RF sizes, a balance between accuracy and complexity remains elusive. In addition, the existing group attention mechanism that simply uses the squeeze-and-excitation method neglects the spatial position information in the feature selection and fusion process. Therefore, this research introduces a lightweight and efficient architecture named Split-Dense Adaptive Network (SDANet) to cope with these limitations. In the proposed network, a residual-like multi-kernel method is implemented to enable better feature extraction under diverse RF sizes. Next, a new grouped attention module processes features dynamically and highlight the location information. Also, the constructed feature augmentation structure strengthens the model's representation. Furthermore, a new channel split and merge strategy is utilized for computation reduction. Compared with state-of-the-art methods, our model achieved better generalization ability, less computational complexity, and superior precision based on various public datasets. The introduced network shows a promising general applicability in the field of computer vision, and further inspires research on supervised deep learning.}, archive = {J_EAAI}, author = {Hanxiang Wang and Yanfen Li and Tan N. Nguyen and L. Minh Dang}, doi = {10.1016/j.engappai.2025.110456}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110456}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Residual-like multi-kernel block and dynamic attention for deep neural networks}, volume = {148}, year = {2025}, }
• (2025). Identification of zinc stripping defects from cathode plate based on deep learning. EAAI, 148, 110448. (www)

  @article{ , abstract = {During hydro-zinc smelting, the cathode plates are attached by with residual zinc or discarded due to damaged insulation strips and edging strips. Such defects limit the recycling of cathode plates. Current manual observation leads to low accuracy and speed of recognition owing to perception biases. Therefore, this work applied computer vision and deep learning semantic segmentation technology to realize the defect recognition of cathode plates. Firstly, a semantic segmentation dataset on cathode plates was constructed for training and testing the model. Then a network of attention mechanism and multiscale feature fusion (AMNet) was proposed to detect the defects. In AMNet, the encoder-decoder jump connection architecture was designed to fuse low-level and high-level features. A channel attention module was incorporated to enhance focus on the channels with important information, and the newly proposed multiscale feature extraction module was used to solve the problem of target multiscale capture. Through related parameter selection experiments, the final AMNet achieved 95.12% and 97.73% for Mean Intersection over Union (MIoU) and mean pixel accuracy (MPA), respectively. These values are 3.24 and 1.74 percentage points higher than DeepLabv3+.}, archive = {J_EAAI}, author = {Tao Liu and Yibin Liu and Jian Chen and Jin Gong}, doi = {10.1016/j.engappai.2025.110448}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110448}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identification of zinc stripping defects from cathode plate based on deep learning}, volume = {148}, year = {2025}, }
• (2025). Non-repetitive motion planning for autonomous vehicles under perceptual uncertainty. EAAI, 148, 110446. (www)

  @article{ , abstract = {Under perceptual uncertainty stemming from occlusion, autonomous vehicles lacking complete information are highly susceptible to collisions with pedestrians suddenly emerging from occluded areas. This problem is often tackled by repetitive training to obtain the distribution of uncertain information, but it is inadequate for addressing untrained new environments. We propose a framework for guiding vehicles in recognizing uncertainty. The framework uses dynamic probabilities to represent pedestrian uncertainty, with different states of pedestrians corresponding to games under different branches of nature choices. The cognitive probabilities will be updated over time until reaching certainty. After establishing cognition at a specific moment, the concept of equilibrium fusion is introduced to address the challenges of strategy-making in the face of multiple potential games under uncertainty. Our proposed framework demonstrates the capability to achieve effective collision avoidance planning in non-repetitive training environments, with behavioral patterns closely resembling those observed in human driving styles, and achieve results comparable to repetitive training.}, archive = {J_EAAI}, author = {Yuqing Cao and Hao Sun and Guangjing Yang}, doi = {10.1016/j.engappai.2025.110446}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110446}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Non-repetitive motion planning for autonomous vehicles under perceptual uncertainty}, volume = {148}, year = {2025}, }
• (2025). Blending traditional and novel techniques: Hybrid type-1 fuzzy functions for forecasting. EAAI, 148, 110445. (www)

  @article{ , abstract = {Time series forecasting has progressed from relying solely on traditional linear or non-linear methods to hybrid approaches that integrate complementary strengths. Traditional models often excel in specific scenarios but struggle with the complexities of real-world data, where linear trends and non-linear dynamics coexist. To address these limitations, this study introduces hybrid forecasting methods that combine autoregressive techniques with type-1 fuzzy functions. A second version of the model integrates a subsampling bootstrap approach, enabling statistical inference and confidence interval estimation within a fuzzy framework. The proposed methods were tested on datasets from four international stock exchanges, each encompassing five distinct years (2010–2014). Forecasting was conducted across three different test set sizes (10, 20, and 40 observations), resulting in a total of 60 experimental scenarios. The hybrid models consistently outperformed benchmark methods in 75% of cases, demonstrating their ability to capture both linear trends and non-linear complexities effectively. The bootstrap-enhanced version further improved reliability, delivering sharper probabilistic forecasts and robust confidence intervals, particularly in volatile financial environments. To ensure precise and adaptive parameter tuning, artificial intelligence-based optimization techniques, such as particle swarm optimization, were employed. This integration of artificial intelligence not only enhanced forecasting accuracy but also demonstrated the transformative potential of hybrid models for real-world applications. These findings provide a valuable tool for decision-making in dynamic and uncertain domains, such as financial market analysis.}, archive = {J_EAAI}, author = {Ali Zafer Dalar and Erol Egrioglu}, doi = {10.1016/j.engappai.2025.110445}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110445}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Blending traditional and novel techniques: Hybrid type-1 fuzzy functions for forecasting}, volume = {148}, year = {2025}, }
• (2025). An uncertainty-based collaborative weakly supervised segmentation network for positron emission tomography-computed tomography images. EAAI, 148, 110442. (www)

  @article{ , abstract = {Weakly supervised segmentation has emerged as an alternative to mitigate the necessity for large volumes of annotated data in semantic segmentation, particularly crucial in medical image analysis where pixel-level labeling is time-consuming and labor-intensive. At present, many weakly supervised methods for single modality medical image segmentation have been proposed, but there is a lack of research on multimodality, especially for Positron emission tomography-Computed tomography images. In Positron emission tomography-Computed tomography images, objects may be easily distinguishable in one modality but indistinguishable in another modality. Therefore, we propose an Uncertainty-based Collaborative Weakly Supervised Segmentation Network. First, we propose a Self-Refine module to output a more precise Class Activation Map for a single modality. Then, an uncertainty collaborative learning strategy is proposed, which follows the principle of “who claim, who burden the evidence”. Uncertainty collaborative learning strategy leverages the uncertainty dispersion module to make the modalities considered to have tumors dominate in the final segmentation results fusion. If both modalities are considered to have a tumor, the cross-modal consistency constraint is utilized to obtain more precise segmentation results. Finally, we extend the uncertainty collaborative learning strategy to the fully supervised segmentation task. We do experiments on two datasets of soft tissue sarcoma and liver tumors. The experimental results prove that the proposed method is superior to other weakly supervised methods in Positron emission tomography-Computed tomography images. In addition, experiments on fully supervised segmentation also demonstrate that the uncertainty collaborative learning strategy proposed can improve the segmentation results of different segmentation networks. The code is available at https://github.com/HarriesDZS/UCWS-Net .}, archive = {J_EAAI}, author = {Zhaoshuo Diao and Huiyan Jiang}, doi = {10.1016/j.engappai.2025.110442}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110442}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An uncertainty-based collaborative weakly supervised segmentation network for positron emission tomography-computed tomography images}, volume = {148}, year = {2025}, }
• (2025). A patch-based deep learning framework with 5-B network for breast cancer multi-classification using histopathological images. EAAI, 148, 110439. (www)

  @article{ , abstract = {Despite the fact that convolutional networks have long been the preferred architecture for investigating breast cancer (BC), new research has revealed that transformer-based architectures perform better in specific circumstances. Currently, Vision Transformer (ViT) has shown to be the most successful transformer-based architecture in vision tasks by employing patch encoding. Drawing inspiration from the patch-based modeling, this study proposed a Patch-Based deep learning network with direct operation on patches as input and the segregation of blending of spatial and channel parameters without maintaining the same size and resolution across the structure as well as simply employing conventional convolutions to carry out the mixing phases. This study further introduced a novel 5-B Network at the end of the pointwise convolutional layer for tiny feature extraction. The 5-BNet comprises five branches that process information concurrently. The primary distinction among these branches lies in the size of their convolutional kernels. In order to capture high-level image features, the 5-BNet gradually reduces the filter size of each convolution layer in every concurrent branch. Based on the proposed model, an end-to-end training for breast cancer (BC) multi-classification using the publicly available BreakHis (Benign class and Malignant Class) is carried out. In addition, an Eight-class multi-classification set by incorporating the benign four classes and the malignant four classes to further evaluate the robustness of the proposed model in multi-classification tasks. Also, the proposed model visualized internal composition to demonstrate how it understood the different patterns of the input images is illustrated. The proposed model obtained 98.1 ± 1.0% accuracy for eight classes and 98.01 ± 1.0% for four classes on all magnifications. The experimental results show that the proposed method achieves the highest breast classification accuracy when compared to cutting-edge models.}, archive = {J_EAAI}, author = {Jehoiada Jackson and Linda E. Jackson and Chiagoziem C. Ukwuoma and Maame D. Kissi and Ariyo Oluwasanmi and Qin Zhiguang}, doi = {10.1016/j.engappai.2025.110439}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110439}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A patch-based deep learning framework with 5-B network for breast cancer multi-classification using histopathological images}, volume = {148}, year = {2025}, }
• (2025). Detection of concealed object using terahertz images: A comprehensive review. EAAI, 148, 110432. (www)

  @article{ , abstract = {Terahertz (THz) imaging has emerged as a potential technique for concealed object detection due to its ability to penetrate a variety of materials and deliver detailed pictures while emitting no ionizing radiation. This review work thoroughly investigates the wide variety of approaches and architectures used in the detection of hidden targets using THz images. The study explores enhancements in THz sources, detectors, and imaging modalities to analyze the evolution of THz technology. In the context of THz-based concealed object detection, several signal processing approaches such as image enhancement, feature extraction, and machine learning(ML) algorithms are discussed. This paper provides a thorough overview of techniques and architectures in the literature, it examines target detection models and their workflows, analyzing techniques and evaluating model performances. The paper involves examining, contrasting, and noting various approaches, highlighting the strengths and weaknesses of different approaches. It aims to provide researchers with a clear understanding of how each technique can impact performance, aiding their decision-making for future use. The study also discusses potential future directions and emphasizes the importance of hybrid models across different processing methods, contributing insights for the ongoing development of research in this area.}, archive = {J_EAAI}, author = {Phibansabeth Nongkseh and Samarendra Nath Sur and Debdatta Kandar}, doi = {10.1016/j.engappai.2025.110432}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110432}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detection of concealed object using terahertz images: A comprehensive review}, volume = {148}, year = {2025}, }
• (2025). Deep learning-enhanced quantum optimization for integrated job scheduling in container terminals. EAAI, 148, 110431. (www)

  @article{ , abstract = {This study explored the challenge of multi-equipment collaborative scheduling and management within container terminals, focusing on essential equipment such as reach stackers, automated guided vehicles (AGVs), quay cranes, and yard cranes. The job shop scheduling problem (JSSP) in container handling was modeled using a quadratic unconstrained binary optimization (QUBO) formulation. A novel scheduling approach is developed based on the synergistic combination of an artificial intelligence (AI) algorithm and quantum computing (QC) (or AI-QC scheduler). This strategy integrates deep reinforcement learning (DRL) and a quantum approximate optimization algorithm (QAOA) to minimize the makespan (total completion time) with accurate coordination of port equipment. Based on operating scenarios in port environments, the proposed AI-QC scheduler offers significant advantages over traditional methods such as first-in-first-out (FIFO), shortest processing time (SPT), and genetic algorithms (GA), particularly in handling stochastic disturbances in processing times. The hybrid technique based on deep learning, quantum computing, and (DRL-QAOA) has self-optimization capabilities, adjusting to changes in market demand in port operations. The AI-QC scheduler can outperform benchmark algorithms in allocating and scheduling port resources, achieving a lower makespan and enhanced operational efficiency across various scenarios. Robust performance indicates optimizing equipment utilization and enhancing terminal throughput, particularly in unpredictable scenarios. Hence, deep-learning AI-driven quantum optimization strategies can improve the competitiveness and efficiency of container terminal operations in a rapidly evolving global market.}, archive = {J_EAAI}, author = {Truong Ngoc Cuong and Hwan-Seong Kim and Le Ngoc Bao Long and Sam-Sang You and Nguyen Duy Tan}, doi = {10.1016/j.engappai.2025.110431}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110431}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning-enhanced quantum optimization for integrated job scheduling in container terminals}, volume = {148}, year = {2025}, }
• (2025). Enhancing autism spectrum disorder early detection with parent-child dyads block-play protocol and attention-enhanced hybrid deep learning framework. EAAI, 148, 110430. (www)

  @article{ , abstract = {Autism Spectrum Disorder (ASD) is a rapidly growing neurodevelopmental disorder. Early intervention is crucial for the development of young children with ASD, yet traditional clinical screening methods often lack objectivity. We introduce a novel Parent-Child Dyads Block-Play (PCB) protocol that captures distinct behavioral patterns in ASD toddlers during naturalistic interactions with their parents. This protocol systematically captures and quantifies parent–child interactions during the block-play task, providing a structured and naturalistic environment to observe ASD-relevant behaviors. Drawing on kinesiological and neuroscientific insights, our approach analyzes movement dynamics to reliably differentiate ASD from typically developing (TD) toddlers. In a dataset of 129 toddlers (40 ASD, 89 TD), we analyze the videos using a hybrid deep learning framework that integrates a two-stream graph convolution network (2sGCN) with an attention-enhanced extended long short-term memory (AxLSTM), enabling the capture of both spatial and temporal aspects of movement. Our 2sGCN-AxLSTM framework efficiently analyzes human dynamic behavioral patterns and is able to distinguish between ASD and typical developmental disorders with an unprecedented 89.6% accuracy. This high level of accuracy holds promise for practical clinical use, as it could facilitate timely interventions and potentially improve developmental outcomes. By focusing on real-life parent–child interactions, the proposed PCB protocol provides a valuable tool that can complement traditional assessments, facilitating timely interventions, and potentially improving developmental outcomes.}, archive = {J_EAAI}, author = {Xiang Li and Lizhou Fan and Hanbo Wu and Kunping Chen and Xiaoxiao Yu and Chao Che and Zhifeng Cai and Xiuhong Niu and Aihua Cao and Xin Ma}, doi = {10.1016/j.engappai.2025.110430}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110430}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing autism spectrum disorder early detection with parent-child dyads block-play protocol and attention-enhanced hybrid deep learning framework}, volume = {148}, year = {2025}, }
• (2025). Recursively learning fine-grained spatial–temporal features for video-based person re-identification. EAAI, 148, 110429. (www)

  @article{ , abstract = {Video-based person Re-identification (Re-ID) is a field within artificial intelligence that aims to match the same person under different cameras. Its core challenge lies in how to effectively reinforce the spatial–temporal feature learning. Existing methods usually perform careful spatial feature enhancement and simple temporal feature aggregation, or the opposite, thus these methods may easily miss fine-grained temporal or spatial clues due to occlusion. To address the above issues, we propose a recursive spatial–temporal feature learning framework to enhance spatial features and recursively integrate them along the timeline. Specifically, the Trigeminal Attention Fusion (TAF) model is proposed to perform spatial complementary learning through self-relation and cross-relation attention. The TAF module consists of three branches: the Convolutional Neural Network (CNN) Branch, the Keypoint Branch, and the Global Branch. The CNN Branch uses self-relation attention to extract enhanced local features. The Keypoint Branch employs cross-relation attention to capture pedestrian keypoint-based features and handle occlusion under global guidance. After that, the Temporal Attention Alignment (TAA) model is designed to recursively propagate temporal information between adjacent frames. Furthermore, we design a bottom-up and top-down training strategy to improve the feature learning ability of the model, which can mine high-quality video-level features from frame-level features via bottom-up inference and refine frame-level features under the top-down guidance of video-level semantic feedback. Extensive experiments on four public Re-ID benchmarks demonstrate that our framework outperforms several state-of-the-art methods and performs well even in occlusion cases.}, archive = {J_EAAI}, author = {Haifei Ma and Canlong Zhang and Zhixin Li and Zhiwen Wang}, doi = {10.1016/j.engappai.2025.110429}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110429}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Recursively learning fine-grained spatial–temporal features for video-based person re-identification}, volume = {148}, year = {2025}, }
• (2025). Implicit embedding based multi modal attention network for cricket video summarization. EAAI, 148, 110428. (www)

  @article{ , abstract = {Personalized video summarization involves generation of user specific video summary grounded on individual user preferences. The existing approaches rely on uni-modal information and generate static video summary in different application domains. It presents potential gap towards requirement of multimodal approach towards dynamic video summarization in cricket sports domain. Cricket video summarization has great potential towards sports and commercial application of Artificial intelligence (AI). This paper presents a Multi-Modal Attention (M-MA) architecture consisting of user attention through Active Implicit User Preference Acquisition (AI-UPA) and Audio Based Attention for generation of user specific dynamic Cricket video summary (highlights). This work also presents a User Recognition and Profiling (URP) Module to reduce the system overhead and facilitate quick summary generation for multiple access instances. In this paper, we present a novel face emotion dataset named Cricket Stimuli Emotion (CS-Emo) dataset. This paper proposes two novel personalized summary evaluation metrics based on qualitative and hybrid evaluation paradigms. The AI-UPA model achieved validation accuracy of 83.33% after last epoch (75 epoch cycle). The M-MA method achieved an Avg. Qual Score (QS) score of 9.16, with an advancement of 0.46 score value from the closest performing existing related method (8.7) making a performance gain of 4.6%. The user involvement in preference acquisition, summary generation and summary evaluation process facilitated for user satisfactory and preference guided summarization. The result analysis over quantitative and proposed evaluation metrics indicate the supremacy of our approach and establishes a benchmark towards personalized video summarization domain.}, archive = {J_EAAI}, author = {Ipsita Pattnaik and Pulkit Narwal}, doi = {10.1016/j.engappai.2025.110428}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110428}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Implicit embedding based multi modal attention network for cricket video summarization}, volume = {148}, year = {2025}, }
• (2025). A multi-module explainable artificial intelligence framework for project risk management: Enhancing transparency in decision-making. EAAI, 148, 110427. (www)

  @article{ , abstract = {The remarkable advancements in machine learning (ML) have led to its extensive adoption in Project Risk Management (PRM), leveraging its powerful predictive capabilities and data-driven insights that support proactive decision-making. Nevertheless, the “black-box” nature of ML models obscures the reasoning behind predictions, undermining transparency and trust. To address this, existing explainable artificial intelligence (XAI) techniques, such as Local Interpretable Model-agnostic Explanations (LIME), Global Priors-based LIME (G-LIME), and SHapley Additive exPlanations (SHAP), have been applied to interpret black-box models. Yet, they face considerable limitations in PRM, including their inability to model cascading effects and multi-level dependencies among risk factors, suffering from inconsistencies due to random sampling, and failure to capture non-linear interactions in high-dimensional risk data. In response to these shortcomings, this paper proposes the Multi-Module eXplainable Artificial Intelligence framework for Project Risk Management (MMXAI-PRM), a novel approach designed to address the unique demands of PRM. The framework consists of three modules: the Risk Relationship Insight Module (RRIM), which models risk dependencies using a Knowledge Graph (KG); the Risk Factor Influence Analysis Module (RFIAM), which introduces a Conditional Tabular Generative Adversarial Network-aided Local Interpretable Model-agnostic Explanations using Kernel Ridge Regression (CTGAN-LIME-KR) to ensure explanation consistency and handle non-linearity; and the Visualization and Interpretation Module (VIM), which synthesizes these insights into an interpretable, chain-based representation. Extensive experiments demonstrate that MMXAI-PRM delivers more consistent, stable, and accurate explanations than existing XAI methods. By improving interpretability, it enhances trust in AI-driven risk predictions and equips project managers with actionable insights, advancing decision-making in PRM.}, archive = {J_EAAI}, author = {Bodrunnessa Badhon and Ripon K. Chakrabortty and Sreenatha G. Anavatti and Mario Vanhoucke}, doi = {10.1016/j.engappai.2025.110427}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110427}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-module explainable artificial intelligence framework for project risk management: Enhancing transparency in decision-making}, volume = {148}, year = {2025}, }
• (2025). Adversarial contrastive domain-generative learning for bacteria raman spectrum joint denoising and cross-domain identification. EAAI, 148, 110426. (www)

  @article{ , abstract = {Raman spectroscopy, as a label-free detection technology, has been widely utilized in the clinical diagnosis of pathogenic bacteria. However, Raman signals are naturally weak and sensitive to the condition of the acquisition process. The characteristic spectra of a bacteria can manifest varying signal-to-noise ratios and domain discrepancies under different acquisition conditions. Consequently, existing methods often face challenges when identifying unobserved acquisition conditions, i.e., the testing acquisition conditions are unavailable during model training. In this article, a generic framework, namely, an adversarial contrastive domain-generative learning framework, is proposed for joint Raman spectroscopy denoising and cross-domain identification. The proposed method is composed of a domain generation module and a domain task module. Through adversarial learning between these two modules, it utilizes only a single available source domain spectral data to generate extended denoised domains that are semantically consistent with the source domain and extract domain-invariant representations. Experimental results show that the proposed method significantly enhances diagnostic performance, with an average recognition accuracy improvement of over +5.0% under unknown acquisition conditions compared to existing methods. Notably, the proposed method also performs simultaneous denoising of the spectra, enhancing the signal-to-noise ratio of the original signal by an average gain of +3.0, without requiring noise-free ground truth. These results suggest that the proposed method holds great potential as a diagnostic tool for real-world clinical cases.}, archive = {J_EAAI}, author = {Haiming Yao and Wei Luo and Xue Wang}, doi = {10.1016/j.engappai.2025.110426}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110426}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adversarial contrastive domain-generative learning for bacteria raman spectrum joint denoising and cross-domain identification}, volume = {148}, year = {2025}, }
• (2025). Enhancing neighborhood-based co-clustering contrastive learning for multi-entity recommendation. EAAI, 148, 110425. (www)

  @article{ , abstract = {To enhance recommendation performance, multi-behavior has become popular for addressing data sparsity in recommendation systems (RS). Meanwhile, due to its self-supervised nature which is not reliant on large amounts of labeled data, contrastive learning (CL) has emerged as a critical technique for solving the cold start-induced data sparsity problem. In the context of multi-behavior RS based on CL, existing methods mainly construct positive and negative sample pairs from implicit user behaviors. However, these methods often rely on black-box neural networks (a common artificial intelligence technique) to simulate the relationships between users and items, making the target nodes susceptible to unreliable neighborhood representations during high-order graph convolution. Specifically, (1) with respect to different user-relationship classifications, although users may belong to different clusters, they could also belong to multiple clusters simultaneously (for example, user u might belong to both sport-cluster and book-lover-cluster). For users associated with multiple clusters, the RS should provide clear guidance on how to effectively recommend items. (2) Because different clusters of users have different preferences, the feature extraction should also differ. If clustering feature extraction originates from unreliable neighbor nodes, it is prone to generating noisy data. To address these challenges, we propose the Enhancing Neighborhood-based Co-clustering Contrastive Learning (NCCL) model, which simultaneously leverages user-level and item-level CL from two perspectives: user-soft-clustering relationships (horizontal) and item-hard-clustering (vertical). Specifically, NCCL combines user-item and item-entity clusters. In the horizontal embedding, CL is performed based on user-centric user-item interactions, whereas in the vertical embedding, CL is conducted using item-centric item-entity aggregations. NCCL aims to incorporate multi-entity data into CL to mitigate data sparsity issues. The model extracts feature values of users and items from different granular perspectives, eliminating noise occasioned by neighboring clusters, which enhances the RS accuracy. Extensive experiments on four public datasets demonstrate that the proposed NCCL significantly outperforms the state-of-the-art techniques, with average improvements of 19.38% in Hit Ratio (HR) and 38.66% in Normalized Discounted Cumulative Gain (NDCG).}, archive = {J_EAAI}, author = {Juan Liao and Aman Jantan and Zhe Liu}, doi = {10.1016/j.engappai.2025.110425}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110425}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing neighborhood-based co-clustering contrastive learning for multi-entity recommendation}, volume = {148}, year = {2025}, }
• (2025). Robust augmentation-based contrastive clustering with negative data mining. EAAI, 148, 110414. (www)

  @article{ , abstract = {Contrastive learning has been widely applied in unsupervised clustering tasks due to its superior performance in learning discriminative representations. However, the data augmentations commonly used in current contrastive clustering methods tend to combine weak and strong augmentations, which make it challenging to produce truly hard augmented examples due to the property of linear transformations. In addition, the simplicity of negative sample setting in contrastive learning generally leads to false positive samples, which aggravates the clustering performance. To alleviate these concerns, we propose a contrastive clustering method based on robust augmentation and negative data mining. Firstly, a robust augmentation method is introduced, which combines nonlinear augmentations with traditional linear augmentation methods to form contrastive pairs, to compensate for the limitations of linear augmentation in capturing complex features. Subsequently, we propose a hypothesis, i.e., if numerous samples of one class which widely separated from its cluster prototype are closest to the other cluster prototype, the two classes are considered to be the visual confusion categories (VCCs). Obviously, VCCs are particularly prone to boundary overlap in the clustering process. Therefore, we employ a confidence estimation criterion to obtain pseudo labels and increase the weight of negative samples/categories with VCCs labels, thereby enhancing the quality of clustering decision boundaries. Comprehensive experiments performed on four commonly used image benchmarks show that the proposed method is efficient.}, archive = {J_EAAI}, author = {Liu Sun and Ming He and Shuai Qin and Nianbin Wang}, doi = {10.1016/j.engappai.2025.110414}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110414}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust augmentation-based contrastive clustering with negative data mining}, volume = {148}, year = {2025}, }
• (2025). DCMA-net: A dual channel multi-scale feature attention network for crack image segmentation. EAAI, 148, 110411. (www)

  @article{ , abstract = {Cracks are a common structural damage to pavement that significantly threatens traffic safety, so regular inspections of road conditions are essential for maintaining traffic safety. Artificial intelligence (AI) technology has demonstrated significant potential in various computer vision tasks and has been widely applied to the field of crack detection in recent years. However, the crack detection remains a challenging task due to the intricacy of crack types, the presence of intensity inhomogeneities, and the difficulty in detecting edge areas within a complex background. To address these challenges, this paper proposes a novel dual channel multi-scale feature attention segmentation network (DCMA-Net) model for crack detection. Specifically, this two feature extraction channels are exploited to obtain global and local information. A multi-scale feature extraction module (MFE) is integrated into the residual structure to capture depth multi-scale information. Additionally, a hybrid attention mechanism is employed to connect the coder and decoder, allowing the network to focus on the target area from different scales and dimensions to enrich crack feature representation. To verify the superiority of the proposed method, we evaluate it on three crack datasets and compare it with state-of-the-art crack detection techniques. The experimental results demonstrate the superior segmentation performance of our proposed network compared to existing advanced methods.}, archive = {J_EAAI}, author = {Yidan Yan and Junding Sun and Hongyuan Zhang and Chaosheng Tang and Xiaosheng Wu and Shuihua Wang and Yudong Zhang}, doi = {10.1016/j.engappai.2025.110411}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110411}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DCMA-net: A dual channel multi-scale feature attention network for crack image segmentation}, volume = {148}, year = {2025}, }
• (2025). ShapG: New feature importance method based on the shapley value. EAAI, 148, 110409. (www)

  @article{ , abstract = {With wide application of Artificial Intelligence (AI), it has become particularly important to make decisions of AI systems explainable and transparent. In this paper, we proposed a new Explainable Artificial Intelligence (XAI) method called ShapG (Explanations based on Shapley value for Graphs) for measuring feature importance. ShapG is a model-agnostic global explanation method. At the first stage, it defines an undirected graph based on the dataset, where nodes represent features and edges are added based on calculation of correlation coefficients between features. At the second stage, it calculates an approximated Shapley value by sampling the data taking into account this graph structure. The sampling approach of ShapG allows to calculate the importance of features efficiently, i.e. to reduce computational complexity. Comparison of ShapG with other existing XAI methods shows that it provides more accurate explanations for two examined datasets. We also compared other XAI methods developed based on cooperative game theory with ShapG in running time, and the results show that ShapG exhibits obvious advantages in its running time, which further proves efficiency of ShapG. In addition, extensive experiments demonstrate a wide range of applicability of the ShapG method for explaining complex models. We find ShapG an important tool in improving explainability and transparency of AI systems and believe it can be widely used in various fields.}, archive = {J_EAAI}, author = {Chi Zhao and Jing Liu and Elena Parilina}, doi = {10.1016/j.engappai.2025.110409}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110409}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ShapG: New feature importance method based on the shapley value}, volume = {148}, year = {2025}, }
• (2025). Interpretable vertical federated learning with privacy-preserving multi-source data integration for prognostic prediction. EAAI, 148, 110408. (www)

  @article{ , abstract = {Digital healthcare is entering a new era characterized by the availability of extensive biomedical data. This wealth of data not only facilitates clinical diagnosis research but also enables the customization of appropriate treatments by incorporating various aspects of patient data. Advanced deep learning methods have the potential to improve clinical diagnosis and treatment efficient significantly. However, a major challenge lies in collecting datasets that are sufficiently large and diverse to train robust prognostic prediction models. Although various types of multi-source data can provide comprehensive insights for medical diagnosis, patient data are often collected by different institutions. Collaborations among multiple institutions involve sharing such data, which raises significant concerns regarding privacy protection and ownership. To address these challenges, this study proposes an interpretable ver tical fed erated learning with me ta- o ptimization (VerFedMeo) that integrates private multi-source data and improves the accuracy of cancer prognosis prediction. Specifically, VerFedMeo employs a partial likelihood function approximation as a its learning objective and derives gradient formulas to achieve more accurate approximations. Meta-optimization has been developed to enhance the model’s learning capacities by automatically identifying optimal hyperparameters. The experimental results demonstrate that VerFedMeo approach significantly outperforms existing methods, achieving improvements of 5.8% in F1-score and 3.5% in area under the curve(AUC) compared to the suboptimal method. Moreover, we presented an interpretable consequence of the potential summaries that could be used by clinicians to predict disease prognosis.}, archive = {J_EAAI}, author = {Qingyong Wang}, doi = {10.1016/j.engappai.2025.110408}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110408}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpretable vertical federated learning with privacy-preserving multi-source data integration for prognostic prediction}, volume = {148}, year = {2025}, }
• (2025). A novel structural adaptive seasonal grey bernoulli model in natural gas production forecasting. EAAI, 148, 110407. (www)

  @article{ , abstract = {Energy data are characterized by nonlinearity, seasonality and growth, which render prediction challenging, making it difficult for Chinese government to formulate relevant policies. Considering that the grey Bernoulli model can deal with nonlinear and growing data, and utilizing the newly proposed seasonal factor, we propose a novel structural adaptive seasonal grey Bernoulli model, which is limited to application to nonlinear, seasonal, and growing time sequence. Firstly, the traditional seasonal factors are logarithmically transformed to ensure more refined seasonal factor sequence. Then, the grey Bernoulli model and the grey structural adaptive model are combined in a multiplicative manner to create a new model, thereby refining its structural adaptability. And the Grey Wolf Optimization algorithm is used to optimize the parameters in the model. Finally, the natural gas production prediction is implemented to verify the feasibility of this model. The mean absolute percentage errors of the proposed model are below 5% both in the experiment and in the robustness test. Compared with other models, this model demonstrates an improvement of 29.75% in prediction accuracyand 15.80% in robustness test. Additionally, the Diebold & Mariano statistics exhibit values that are essentially significant at the 10% level. The experimental results substantiate the validity, superiority, and robustness of the model proposed in this paper, which can be effectively applied to the prediction of seasonal natural gas production.}, archive = {J_EAAI}, author = {Shuli Yan and Mengna Peng and Lifeng Wu and Pingping Xiong}, doi = {10.1016/j.engappai.2025.110407}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110407}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel structural adaptive seasonal grey bernoulli model in natural gas production forecasting}, volume = {148}, year = {2025}, }
• (2025). Dynamic modeling and control of pneumatic artificial muscles via deep lagrangian networks and reinforcement learning. EAAI, 148, 110406. (www)

  @article{ , abstract = {Pneumatic artificial muscles (PAMs), as typical soft actuators characterized by hysteresis and nonlinearity, pose a challenging task in modeling and control. This paper proposes a Deep Lagrangian Networks Reinforcement Learning (DeLaNRL) controller that combines deep Lagrangian networks (DeLaN) with reinforcement learning to achieve precise motion control of PAMs. By leveraging the DeLaN model, the dynamic model is constrained to adhere to the Lagrangian first principle, enhancing the model’s compliance with physical constraints. Furthermore, to improve the generality and adaptability of the model to various input data, the Self-scalable tanh (Stan) function is employed as the activation function within the DeLaN model. To validate the effectiveness of the proposed modeling approach, the model is tested on both sampled and unknown motions. The results demonstrate the effectiveness and generalization capability of the DeLaN model with the Stan activation function. Subsequently, the reinforcement learning controller is applied to the learned dynamics model, resulting in control strategies capable of precise motion control. To further demonstrate the effectiveness of the proposed controller, experiments are conducted on both simulation and the experiment platform for reaching and tracking tasks. The simulation results indicate that the control error is less than 0.91 millimeters, while on the experimental platform, the control error is less than 3.7 millimeters. These results confirm that the proposed DeLaNRL controller exhibits high control performance.}, archive = {J_EAAI}, author = {Shuopeng Wang and Rixin Wang and Yanhui Liu and Ying Zhang and Lina Hao}, doi = {10.1016/j.engappai.2025.110406}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110406}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic modeling and control of pneumatic artificial muscles via deep lagrangian networks and reinforcement learning}, volume = {148}, year = {2025}, }
• (2025). Dynamic gaussian graph operator: Learning parametric partial differential equations in arbitrary discrete mechanics problems. EAAI, 148, 110405. (www)

  @article{ , abstract = {Deep learning methods have access to be employed for solving physical systems governed by parametric partial differential equations (PDEs) due to massive scientific data. It has been refined to operator learning that focuses on learning non-linear mapping between infinite-dimensional function spaces, offering interface from observations to solutions. However, state-of-the-art neural operators are limited to constant and uniform discretization, thereby leading to deficiency in generalization on arbitrary discretization schemes for computational domain. In this work, we propose a novel operator learning algorithm, referred to as Dynamic Gaussian Graph Operator (DGGO) that expands neural operators to learning parametric PDEs in arbitrary discrete mechanics problems. The Dynamic Gaussian Graph (DGG) kernel learns to map the observation vectors defined in general Euclidean space to metric vectors defined in high-dimensional uniform metric space. The DGG integral kernel is parameterized by Gaussian kernel weighted Riemann sum approximating and using dynamic message passing graph to depict the interrelation within the integral term. Fourier Neural Operator is selected to localize the metric vectors on spatial and frequency domains. Metric vectors are regarded as located on latent uniform domain, wherein spatial and spectral transformation offer highly regular constraints on solution space. The efficiency and robustness of DGGO are validated by applying it to solve numerical arbitrary discrete mechanics problems in comparison with mainstream neural operators. Ablation experiments are implemented to demonstrate the effectiveness of spatial transformation in the DGG kernel. The proposed method is utilized to forecast stress field of hyper-elastic material with geometrically variable void as engineering application.}, archive = {J_EAAI}, author = {Chu Wang and Jinhong Wu and Yanzhi Wang and Zhijian Zha and Qi Zhou}, doi = {10.1016/j.engappai.2025.110405}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110405}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic gaussian graph operator: Learning parametric partial differential equations in arbitrary discrete mechanics problems}, volume = {148}, year = {2025}, }
• (2025). A dynamic cross-domain recommendation model with target-aware complementary preference transfer and information fusion. EAAI, 148, 110404. (www)

  @article{ , abstract = {The cold-start problem is one of the most challenging problem in recommender systems. Despite considerable researches, an effective resolution has remained elusive. Recent advancements in cross-domain recommendation(CDR) have offered promising solutions to address the cold-start problem in the target domain by leveraging auxiliary information from the source domain. In particular, meta-learning-based methods have achieved significant performance improvements, yet they overlook the critical guiding role of target item features in the learning process. Besides, they either learn a common bridge or a personalized bridge, thus ignoring the potential complementarity between the two. To address these limitations, we propose a novel method, namely A Dynamic Cross-Domain Recommendation Model with Target-Aware Complementary Preference Transfer and Information Fusion (DnmCDR). Specifically, a target-aware personalized bridge module is proposed to model user’s personalized preferences and incorporate target information to guide the learning process. Afterwards, a common bridge is learned to model the common preferences of all users. Moreover, we introduce a dynamic target-aware fusion module to dynamically fuse the outputs of the personalized bridge and the common bridge. Extensive experiments on three distinct CDR tasks demonstrate the effectiveness of proposed DnmCDR. The source code of our method is available at https://github.com/CreaterLL/DnmCDR .}, archive = {J_EAAI}, author = {Lele Duan and Ling Zhu and Peng Ren}, doi = {10.1016/j.engappai.2025.110404}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110404}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dynamic cross-domain recommendation model with target-aware complementary preference transfer and information fusion}, volume = {148}, year = {2025}, }
• (2025). Deep label relevance and label ambiguity based multi-label feature selection for text classification. EAAI, 148, 110403. (www)

  @article{ , abstract = {Multi-label text classification, where each document can be associated with multiple labels simultaneously, poses unique challenges in feature selection due to the complex relationships between features and labels. In this paper, we propose a novel D eep Label R elevance and L abel A mbiguity (DLRLA) based multi-label feature selection method designed for multi-label text data. Our approach constructs a quasi-relevance matrix integrating low-order, high-order feature-label relevance and label ambiguity. The low-order relevance captures the direct association between individual features and labels, while the high-order relevance accounts for the interactions between feature combinations and labels, collectively termed as deep label relevance. Label ambiguity, measured using information entropy, quantifies the uncertainty associated with each label. The quasi-relevance matrix is then evaluated using Grey Relation Optimization to rank and select the most informative features based on multiple relevance criteria. Additionally, feature-feature relevance is incorporated to reduce the candidate set of high-order features, mitigating computational complexity. Elastic Net Regression, a linear regularized model, estimates feature-label relevance, enabling efficient feature selection while addressing multicollinearity. For multi-label classification, we leverage the Multi-Label K-Nearest Neighbors algorithm, where the key parameters (number of neighbours k and smoothing factor s) are optimized using Particle Swarm Optimization. The proposed DLRLA method is extensively evaluated on ten multi-label text benchmark datasets, considering six performance evaluation metrics. Comparative analyses with seven state-of-the-art methods are conducted. Furthermore, a stability analysis of DLRLA is performed across all datasets and evaluation metrics, showcasing its robustness and consistency.}, archive = {J_EAAI}, author = {Gurudatta Verma and Tirath Prasad Sahu}, doi = {10.1016/j.engappai.2025.110403}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110403}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep label relevance and label ambiguity based multi-label feature selection for text classification}, volume = {148}, year = {2025}, }
• (2025). Determining the superiority of a robust cloud fault tolerance mechanism using a spherical cubic fuzzy set-based decision approach. EAAI, 148, 110402. (www)

  @article{ , abstract = {Ensuring the availability of robust fault tolerance mechanisms is crucial for providing reliable cloud computing services. The complexity of cloud system components, combined with the wide range of fault tolerance frameworks proposed in numerous studies, makes identifying the optimal cloud fault tolerance framework a significant challenge. These frameworks, typically based on either reactive fault tolerance (RFT) or proactive fault tolerance (PFT) mechanisms, can be evaluated using distinct attributes. However, determining the superiority of one framework over another is not straightforward due to several factors: the multiplicity of performance attributes, trade-offs among these attributes, decisions regarding their relative importance, observed variations in attribute data across different frameworks, and nature of the subjective evaluation. To address this challenge, this paper proposes a decision-making approach using multiple attributes decision-making (MADM) methods, including the Fuzzy Decision by Opinion Score Method (FDOSM) and the Fuzzy Weighted with Zero Inconsistency Criterion (FWZIC) method, extended and formulated within Spherical Cubic Fuzzy Sets (SCFS) and integrated with the Preference Ranking Organization Method for Enrichment of Evaluations (PROMETHEE). The developed SCFS–FWZIC method prioritizes the performance attributes of cloud fault tolerance frameworks, while the SCFS–FDOSM method is developed to transform the evaluation values of each framework into scores. PROMETHEE is then employed to rank 19 frameworks under the RFT category and 7 frameworks under the PFT category to identify the optimal one. The results indicate that AFTRC under RFT and ASSURE under PFT ranked highest due to their essential attributes, while SAFTP under RFT and PFHC 2 under PFT received the lowest ranks. Sensitivity, correlation, and comparison analyses were conducted to validate and assess the stability and robustness of the proposed methods. The implications of this study are likely to benefit a variety of stakeholders, including organizations and managers.}, archive = {J_EAAI}, author = {Mohannad T. Mohammed and Mohamed Safaa Shubber and Sarah Qahtan and Hassan A. Alsatta and Nahia Mourad and A.A. Zaidan and B.B. Zaidan}, doi = {10.1016/j.engappai.2025.110402}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110402}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Determining the superiority of a robust cloud fault tolerance mechanism using a spherical cubic fuzzy set-based decision approach}, volume = {148}, year = {2025}, }
• (2025). Systematic review on machine learning and computer vision in precision agriculture: Applications, trends, and emerging techniques. EAAI, 148, 110401. (www)

  @article{ , abstract = {Machine learning (ML) and computer vision (CV) play a crucial role in precision agriculture (PA) by enabling data-driven solutions that enhance crop yields, resource efficiency, and sustainability. These technologies have transformed PA by addressing complex agricultural challenges through real-time monitoring and predictive analytics. This systematic review aims to investigate the recent advancements in ML and CV methodologies within PA, focusing on crop, harvesting, soil and water management. Guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology, the study analyzes 213 articles from the past five years, providing insights into publication trends, key application area, and the distribution of research efforts across various crops and vision data types. The review highlights the predominant use of Convolutional Neural Networks (CNNs) in PA, particularly in crop detection, disease and pest detection, and weed detection. Emerging deep learning techniques, such as Vision Transformers (ViTs) and Generative Adversarial Networks (GANs), are also explored for their potential in visual data analysis. Findings show that maize is the most frequently studied crop, Red-Green-Blue (RGB) imagery is the primary vision data type and self-collected datasets are preferred over public ones. Key challenges include limited access to high-quality and diverse datasets, and the need for adaptable solutions across diverse agricultural contexts. By summarizing these advancements and challenges, this review provides direction for future research and highlights opportunities to expand ML and CV applications in sustainable and efficient agricultural practices.}, archive = {J_EAAI}, author = {Y.N. Kuan and K.M. Goh and L.L. Lim}, doi = {10.1016/j.engappai.2025.110401}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110401}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Systematic review on machine learning and computer vision in precision agriculture: Applications, trends, and emerging techniques}, volume = {148}, year = {2025}, }
• (2025). Binarizing historical architectural drawings with shallow convolutional autoencoders. EAAI, 148, 110400. (www)

  @article{ , abstract = {We developed convolutional autoencoder models to binarize 26 severely stained 1938 architectural drawings of the Department of Agriculture and Commerce Building in Manila. Due to the presence of deep-seated stains that have accumulated over time on the original linen drawings, traditional thresholding techniques failed in binarizing them. We propose utilizing the capabilities of deep learning models, particularly convolutional autoencoders, to automate the binarization of all 26 drawings. Our goal is to efficiently train a model using a segment from a single drawing, thereby eliminating the requirement for separate training on each individual drawing. We manually binarized a section of one drawing to serve as the desired output within our training dataset. Moreover, since deep learning models are computationally heavy, we proposed simplifying them into shallow models by trimming down the architecture to a single hidden layer. We created four pairs of models (deep and shallow) with input sizes 32 × 32 , 64 × 64 , 128 × 128 , and 256 × 256 . All the models effectively binarized the drawing by successfully separating the pronounced stains from the drawn lines, which the traditional binarization techniques have failed to do. The models achieved F1 scores from 0.961 to 0.977, intersection-over-union scores from 0.925 to 0.955, and peak-signal-to-noise ratio values from 23.1 to 25.4. Notably, we achieved higher performance metrics on shallow autoencoders than their deep counterparts. Our workflow also succeeded in binarizing a separate collection of line art drawings on vellum by another architect, which were at a lower resolution.}, archive = {J_EAAI}, author = {Mark Jeremy G. Narag and Gerard Rey Lico and Maricor Soriano}, doi = {10.1016/j.engappai.2025.110400}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110400}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Binarizing historical architectural drawings with shallow convolutional autoencoders}, volume = {148}, year = {2025}, }
• (2025). SFG-net: Semantic relationship and hierarchical fusion-based graph network for enhanced skeleton-based gait recognition. EAAI, 148, 110399. (www)

  @article{ , abstract = {Gait recognition has emerged as a crucial biometric identifier due to its non-invasive and unobtrusive characteristics. Unlike silhouette-based methods, which include appearance information, skeleton-based gait recognition offers gait data without visual clues. However, traditional models in this field often rely on handcrafted features and adjacency matrices formed from physically connected edges, posing a significant challenge in extracting semantically meaningful joints and edges. To address this challenge, a novel Semantic relationship and hierarchical Fusion-based Graph Network (SFG-Net) utilizing a Hierarchical-joint Connectivity Graph (HC-Graph) is proposed. SFG-Net divides each joint node into multiple subsets, facilitating the extraction of both proximal and distant edges, and constructs an HC-Graph to represent these edges within the semantic spaces of the human skeleton. Furthermore, a Hierarchical Attention (HA) mechanism is introduced to emphasize dominant hierarchical edge sets within the HC-Graph. The temporal dynamics of the gait data are captured using Multi-scale Temporal Convolution (MSTC). To further enhance discriminative power, features at different levels are concatenated, capturing both dynamic and structurally semantic features. Experimental results on benchmark gait recognition datasets show that the proposed SFG-Net significantly outperforms current state-of-the-art methods, exhibiting superior robustness and accuracy across various challenging scenarios.}, archive = {J_EAAI}, author = {Priyanka D. and Mala T.}, doi = {10.1016/j.engappai.2025.110399}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110399}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SFG-net: Semantic relationship and hierarchical fusion-based graph network for enhanced skeleton-based gait recognition}, volume = {148}, year = {2025}, }
• (2025). Multi-scale camouflaged feature mining and fusion network for liver tumor segmentation. EAAI, 148, 110398. (www)

  @article{ , abstract = {Automatic segmentation of liver tumors is crucial for assisting clinical doctors in diagnosis and treatment decision-making. However, due to the high heterogeneity and blurred boundaries of liver tumors, existing methods often fail to accurately segment the complete tumor areas, resulting in a high rate of false-negative segmentations. Therefore, we proposed a multi-scale camouflaged feature mining and fusion network, named MCFMFNet, to effectively extract tumor features of varying sizes, positions, and morphologies for accurate liver tumor segmentation. Specifically, we introduced a multi-scale feature aggregation module that utilized a set of multi-scale kernels to extract tumor features under different receptive fields and an element-wise operation method for feature fusion to effectively explore more representative tumor features. Additionally, we designed a camouflaged feature mining module that guided the model to focus on tumor areas that were easily misclassified as background, thereby mitigating the issue of false-negative in the tumor segmentation results. Finally, we validated the effectiveness of the MCFMFNet using a multi-center dataset and a public available dataset. The experimental results demonstrate that, in comparison with existing methods, our approach achieved the highest dice similarity coefficient (0.8413 and 0.7365) and the lowest false negative rates (0.1274 and 0.1961) on private and public available datasets, respectively. The code is available at https://github.com/Meiyan88/MCFMFnet .}, archive = {J_EAAI}, author = {Lei Yang and Jiawei Zhang and Tao Wang and Qianjin Feng and Sirui Fu and Meiyan Huang}, doi = {10.1016/j.engappai.2025.110398}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110398}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale camouflaged feature mining and fusion network for liver tumor segmentation}, volume = {148}, year = {2025}, }
• (2025). Generative adversarial networks based synthetic biology: A promising approach to sars-cov-2 mutations prediction. EAAI, 148, 110395. (www)

  @article{ , abstract = {Synthetic biology, blending biological science and engineering, confronts challenges like the complexity of living systems, difficulties in comprehending the full functionality of genes, the intricate and highly sensitive nature of biological experiments, management of complex data derived from experiments, and ethical concerns. Generative adversarial networks (GANs) have emerged as a potent tool, for overcoming these challenges and enhancing data generation in synthetic biology. This study delves into GAN's application in synthetic biology, focusing on generating synthetic Omicron data for Sever Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). The GAN-generated synthetic Omicron data undergoes evaluation using Machine Learning (ML) and bioinformatics techniques. A Convolutional Neural Network (CNN) trained on synthetic data exhibits remarkable accuracy, reaching 99.52%, in distinguishing synthetic variants from real Variants of Concern (VOCs). Saliency analysis sheds light on the key features distinguishing the synthetic Omicron data from the original one. Bioinformatic analyses, including multiple sequence alignment, and phylogenetic and immunoinformatic analysis, provide comprehensive insights into the biological activities and taxonomical aspects of the synthetic Omicron variant. In-silico immunoinformatic prediction revealed no significant differences in immunological responses between original and synthetic Omicron. Generated mutations in the genome of the synthetic variant might lead to changes in the transmissibility of the Omicron variant. The analysis of synthetic Omicron data contributes to a profound comprehension of genetic variations and their implications. The combined use of ML and bioinformatics tools presents a holistic approach to assessing the quality and characteristics of synthetic biological data. Code, data, and trained models are available on: https://github.com/hashemibahar/generated-omicron .}, archive = {J_EAAI}, author = {Bahar Hashemi and Ahmad Farhad Talebi and Amin Janghorbani}, doi = {10.1016/j.engappai.2025.110395}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110395}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Generative adversarial networks based synthetic biology: A promising approach to sars-cov-2 mutations prediction}, volume = {148}, year = {2025}, }
• (2025). Smart building evacuation by tracking multi-camera network and explainable re-identification model. EAAI, 148, 110394. (www)

  @article{ , abstract = {Real-time crowd data from surveillance devices is essential for the emergency decision-making and management inside complex buildings. Traditional evacuation monitoring with single-camera tracking often leads to erratic information, so multi-camera tracking for building occupants is critical to enhance evacuation safety and emergency response. This research proposes a novel real-time multi-camera tracking framework for the detection, tracking and re-identification (Re-ID) of evacuees across multi-camera. The framework consists of (1) a multi-camera network, (2) human detection model, (3) tracking model, (4) an explainable attention-aided Re-ID (AAR) model, and (5) a module of feature matching and re-distribution algorithm. The attention-aided Re-ID model presents outstanding performance on both the standard big benchmarks and our custom dataset. Moreover, a simple evacuation drill is conducted to demonstrate real-time multi-camera tracking, showing good accuracy in Re-ID and personnel counting, where the overall Re-ID tracking accuracy exceeds 75% and the personnel counting accuracy is approaching 100%. Lastly, the class activation map (CAM) illustrates the model explainability and limitations. The proposed multi-camera tracking framework helps develop a more automated monitoring system and an intelligent digital twin for building emergency safety management.}, archive = {J_EAAI}, author = {Yifei Ding and Xinghao Chen and Yuxin Zhang and Xinyan Huang}, doi = {10.1016/j.engappai.2025.110394}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110394}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Smart building evacuation by tracking multi-camera network and explainable re-identification model}, volume = {148}, year = {2025}, }
• (2025). Bidirectional rapidly exploring random tree path planning algorithm based on adaptive strategies and artificial potential fields. EAAI, 148, 110393. (www)

  @article{ , abstract = {Path planning is central to the operation of intelligent systems such as robots, drones, and autonomous vehicles, where path performance and time efficiency directly impact overall system performance. Although sampling-based path planning methods have achieved significant success in this field, their performance remains limited in crowded environments. This paper combines and improves the bidirectional exploration method of BI-RRT* (Bidirectional Rapidly-exploring Random Tree Star) and the expansion guidance of APF-RRT* (Artificial Potential Field Rapidly-exploring Random Tree Star), proposing a bidirectional rapidly exploring random tree algorithm based on adaptive mechanisms and artificial potential fields (AB-APF-RRT*). This method improves both the sampling and expansion methods of RRT*(Rapidly-exploring Random Tree Star) . In terms of sampling, the probabilities in different regions are modified using the line connecting the start and goal points, and dynamic goal bias and opposing bias strategies are introduced to guide the trees towards the target and each other. In terms of expansion, based on the bidirectional exploration of the two trees, optimized artificial potential fields and ray-casting navigation strategies are applied to guide the trees towards the goal while avoiding obstacles and dynamically adjusting the step size. To enhance the smoothness of the path, a cubic spline interpolation method is further applied. Ultimately, a comparison with several popular sampling-based path planning algorithms demonstrates that this method excels in both performance and time efficiency.}, archive = {J_EAAI}, author = {Zhaokang Sheng and Tingqiang Song and Jiale Song and Yalin Liu and Peng Ren}, doi = {10.1016/j.engappai.2025.110393}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110393}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bidirectional rapidly exploring random tree path planning algorithm based on adaptive strategies and artificial potential fields}, volume = {148}, year = {2025}, }
• (2025). Motion-inspired Spatial–Temporal transformer for accurate vessel trajectory prediction. EAAI, 148, 110391. (www)

  @article{ , abstract = {Accurate maritime vessel trajectory prediction is essential for safety and efficient port scheduling. Traditionally, such predictions rely on model-based methods, but complex forces and unpredictable marine disturbances impose difficulty on modeling. Recently, the collection of data from the Automatic Identification System (AIS) has facilitated the advancement of data-driven methods, especially deep learning. However, data-driven methods are often overfitted and may violate the law of motion due to the inherent noise in AIS data. To address these challenges, this paper presents a novel hybrid approach: Motion-Inspired Spatial–Temporal Transformer (MSTFormer). By combining motion modeling with the expressive power of neural networks, MSTFormer improves trajectory prediction while better capturing dynamic motion patterns. Specifically, MSTFormer employs a Transformer network enhanced with Augmented Trajectory Matrix (ATM) and Dynamic-Aware Attention mechanism to effectively extract dynamic spatial–temporal trajectory features and predict vessel motion accelerations. The predicted positions are then calculated by the accelerations through Euler’s method and spherical motion model. Experimental results on real-world datasets show that our strategy not only effectively improves predictive capability, but also outperforms the backbone on cornering data. Specifically, for the 0.4h prediction horizon, the MSTFormer achieved a 26.7% improvement in Mean Squared Error compared to the best performing baseline model. For datasets with frequent turns, it achieved a 28.1% increase. The ablation analysis further confirms the efficacy of the proposed method. The source code is available at https://github.com/simple316/MSTFormer .}, archive = {J_EAAI}, author = {Huimin Qiang and Zhiyuan Guo and Zhong Chu and Shiyuan Xie and Xiaodong Peng}, doi = {10.1016/j.engappai.2025.110391}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110391}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Motion-inspired Spatial–Temporal transformer for accurate vessel trajectory prediction}, volume = {148}, year = {2025}, }
• (2025). Improving deep random vector functional link networks through computational optimization of regularization parameters. EAAI, 148, 110389. (www)

  @article{ , abstract = {Deep Random Vector Functional Link (dRVFL) networks are a class of randomization-based deep neural networks known for their rapid learning capabilities and universal approximation potential. Despite these advantages, the dRVFL model faces substantial computational and memory challenges as training data, hidden layers, or input feature dimensions grow. These issues are intensified by regularization, as optimal performance requires repeatedly solving large-scale linear systems to tune the regularization parameter. Traditional matrix inversion methods for this task are computationally intensive, memory-demanding, and prone to numerical instability. This study introduces a computationally efficient two-stage approach to address these limitations. In the first stage, a randomized algorithm-based low-rank approximation computes a compressed representation of the feature matrix in O ( m n k ) for an m × n matrix ( k ≪ min ( m , n ) ), requiring only two passes over the data, significantly reducing costs. In the second stage, the derived solution is used to optimize λ via the Golden Section search and Menger curvature-based technique, enabling efficient exploration of the continuous domain with minimal function evaluations. Furthermore, the theoretical framework derives an upper bound on the error of the approximated regularized solution, combining deterministic and probabilistic insights. The proposed method, validated on multiple classification datasets, demonstrates up to 6% improvement in accuracy compared to state-of-the-art feedforward neural network algorithms. Statistical evaluations further confirm its robustness and superior performance over seven baseline algorithms. Overall, the proposed method enhances the scalability, robustness, and efficiency of dRVFL for large-scale applications.}, archive = {J_EAAI}, author = {Chinnamuthu Subramani and Ravi Prasad K. Jagannath and Venkatanareshbabu Kuppili}, doi = {10.1016/j.engappai.2025.110389}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110389}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving deep random vector functional link networks through computational optimization of regularization parameters}, volume = {148}, year = {2025}, }
• (2025). A multiple aging factor interactive learning framework for lithium-ion battery state-of-health estimation. EAAI, 148, 110388. (www)

  @article{ , abstract = {As lithium-ion batteries are widely used in electric vehicles, it has become critical to accurately estimate the state-of-health of the battery. While neural networks have been proven to be effective for state-of-health estimation, such networks primarily focus on feature modeling of raw data without exploiting inherent correlation among multiple dimensional information in the raw data, limiting the estimation accuracy. We thereby propose an interactive learning network for state-of-health estimation. The novel network simultaneously models features and learns correlations among multiple dimensional information using multiple layer perceptron in an interactive manner. We then extract multiple aging factors from the raw voltage and current data as network inputs, which enables knowledge associated to state-of-health of the battery to be encoded in the proposed network. In addition, benefiting from aging factors in lower dimensions than the raw data, computational overhead of the network are significantly reduced. Comprehensive experiments are conducted on two widely-adopted datasets. The experimental results confirm that our proposed network performs accurate state-of-health estimation within a mean absolute error of less than 3% in both of the two datasets, outperforming previous recurrent neural network and Transformer-based methods. Moreover, computational load comparison further demonstrates the potential of the proposed framework in battery management systems.}, archive = {J_EAAI}, author = {Zhengyi Bao and Tingting Luo and Mingyu Gao and Zhiwei He and Yuxiang Yang and Jiahao Nie}, doi = {10.1016/j.engappai.2025.110388}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110388}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multiple aging factor interactive learning framework for lithium-ion battery state-of-health estimation}, volume = {148}, year = {2025}, }
• (2025). Deep spectral clustering network for incomplete multi-view clustering. EAAI, 148, 110387. (www)

  @article{ , abstract = {Spectral clustering is a helpful technique for clustering non-convex data, which extends the clustering to data with multiple partial views. Still, it has a higher running time thanks to cubic time complexity, while extending spectral embeddings to unseen samples is non-trivial and prevents model reuse. In light of this, we propose a Deep Spectral Clustering Network for Incomplete Multi-view Clustering (DSCN-IMC). At its core, DSCN-IMC jointly learns the approximate spectral embeddings and the soft cluster assignments using a deep neural network in an unsupervised and end-to-end fashion. Specifically, we integrate the graph convolutional layers and manifold loss into our network to achieve robustness against incomplete multi-view data. After that, an orthnormalization layer is exploited to fulfill the orthogonal constraint of spectral embeddings, thereby achieving the lower running time of spectral clustering. Building upon this, a self-supervised clustering module is introduced to obtain cluster assignments of unseen samples and facilitates model reuse. Our method’s efficacy is demonstrated through experiments conducted on nine datasets, wherein we compare its performance against nine state-of-the-art baselines.}, archive = {J_EAAI}, author = {Ao Li and Sanlin Mei and Cong Feng and Tianyu Gao and Hai Huang}, doi = {10.1016/j.engappai.2025.110387}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110387}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep spectral clustering network for incomplete multi-view clustering}, volume = {148}, year = {2025}, }
• (2025). Dual-image differential transformer for periodic lace surface defect detection. EAAI, 148, 110386. (www)

  @article{ , abstract = {In lace fabric defect detection, traditional single-image methods struggle to differentiate defects from background textures due to the intricate texture structure of lace and the subtle nature of small defects. To address this challenge, we propose an innovative detection framework based on a contrastive learning strategy named the dual-image differential transformer detection network (DID-Former). This framework compares defect-free reference images with target images, mitigating the adverse effects of complex lace textures on detection performance. The core innovations of DID-Former encompass two key components: First, we introduce the object region representation (ORR) mechanism to enhance the feature representation of defect regions, thereby improving the model’s sensitivity to small defects. Second, we propose the dual-image differential attention (DID-attention) mechanism, which investigates the relationship between the two images to detect texture variations hidden in the complex background, thereby improving the detection accuracy of small defects. Experimental results demonstrate that DID-Former significantly outperforms traditional methods on a real-world lace fabric dataset, achieving a mean Intersection over Union (mIoU) of 83.34%, an F1 of 90.17%, and a Recall of 87.91%. These results validate the effectiveness and robustness of the proposed framework in complex texture environments, particularly in detecting small defects. Although the model’s performance may degrade under high noise interference or extreme textures, DID-Former offers an efficient and promising solution for defect detection in textured fabrics, with substantial potential for practical applications.}, archive = {J_EAAI}, author = {Wenhua Jiao and Xiaoyi Cai and Yuyu Tian and Mingcheng Zuo}, doi = {10.1016/j.engappai.2025.110386}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110386}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-image differential transformer for periodic lace surface defect detection}, volume = {148}, year = {2025}, }
• (2025). Just-in-time software defect location for mobile applications based on pre-trained programming encoder model and hierarchical attention network. EAAI, 148, 110381. (www)

  @article{ , abstract = {Just-in-Time Software Defect Localization plays an essential role in identifying and fixing bugs of mobile applications (APPs). However, state-of-the-art methods usually locate defective codes at the file level instead of the line level, have low performance, and few researches focus on the Software Defect Localization (SDL) of APPs. We propose a code line-level deep learning model (CodeLineDL), which is based on Pre-trained Programming Encoder Model and Hierarchical Attention Network (HAN) to more effectively locate defective files and code lines. First, dynamic word vectors containing contextual code semantics are generated for each commit-level code. Then, Bidirectional Gated Recurrent Unit units are used to capture the surrounding tokens of the commit-level source code lines. Next, HAN is employed to learn the source code's hierarchy to pinpoint the location of defective files within the commit. Finally, an attention mechanism is adopted to calculate the risk scores of tokens for each line of code in the defective file, using the sum of the risk scores of the tokens in that line to generate a ranking of risk lines. To evaluate the effectiveness of CodeLineDL, we assembled a new dataset comprising defective files and their corresponding defect lines of 15 APPs projects. The experimental results show that our method is superior to the previous methods. For the file-level SDL, the CodeLineDL model has an improvement in accuracy and Popt value by 112.81%–291.10% and 0.72%–8.02% respectively. For the line-level SDL, the CodeLineDL model has an improvement in accuracy by 9.74%–42.19%.}, archive = {J_EAAI}, author = {Xiaozhi Du and Xue Yuan and Zongbin Qiao}, doi = {10.1016/j.engappai.2025.110381}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110381}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Just-in-time software defect location for mobile applications based on pre-trained programming encoder model and hierarchical attention network}, volume = {148}, year = {2025}, }
• (2025). Utilizing big data and categorical boosting modeling methodology to interpret the load-deflection behavior of steel fiber-reinforced concrete beams. EAAI, 148, 110377. (www)

  @article{ , abstract = {Extensive research has been conducted on steel fiber reinforced concrete (SFRC) beams, yet studies focused on predicting their load-deflection curves using machine learning (ML) remain limited. This study addresses these gaps by employing advanced ML techniques—random forest, extreme gradient boosting, and CatBoost (Categorical Boosting)—to predict SFRC beam behavior. Using a dataset of 358 beams, rigorous data preprocessing ensured integrity. Hyperparameters were optimized via grid search and cross-validation. The CatBoost model excelled, achieving low error metrics and high correlation coefficients (92.7–95.1%). Analysis identified beam span and fiber tensile strength as key predictors of critical displacements (yield, ultimate, and failure) and loads (yield, ultimate, and failure). Findings indicate beam span (L) significantly influences critical displacements and loads due to its impact on shear and moment values. Fiber tensile strength is crucial for predicting ultimate and failure displacements, especially in later loading stages. Although fiber volume fraction has a lower impact on critical displacement prediction, it ranks higher for critical load prediction, showing variable influence across prediction types. Fiber type ranks low in importance for both displacement and load predictions. This study demonstrates the potential of ML techniques, particularly CatBoost, in advancing the prediction and optimization of SFRC beam designs, offering valuable insights for modern infrastructure development.}, archive = {J_EAAI}, author = {Ahmet Tuken and Yassir M. Abbas and Nadeem A. Siddiqui}, doi = {10.1016/j.engappai.2025.110377}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110377}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Utilizing big data and categorical boosting modeling methodology to interpret the load-deflection behavior of steel fiber-reinforced concrete beams}, volume = {148}, year = {2025}, }
• (2025). A lightweight and high-accuracy framework for printed circuit board defect detection. EAAI, 148, 110375. (www)

  @article{ , abstract = {Efficient and accurate defect detection in Printed Circuit Boards (PCB) is crucial for ensuring the reliability and production efficiency of electronic products. Traditional methods often struggle to achieve an effective balance between detection accuracy and speed. This study proposes Efficient Network for PCB detection (EffNet-PCB), a lightweight detection model based on You Only Look Once v8 (YOLOv8), optimized for real-time high-precision defect detection. Key innovations include the Advanced Feature Fusion Module (AFFM) for enhanced multi-scale feature extraction, the enhanced high-level screening-feature fusion pyramid network (HSFPN-E) to reduce information redundancy, and the Improved Detection Head (IDH) for dynamic and accurate classification and localization. Additionally, the Multi-Scale Convolutional Block Attention Module (MSCBAM) refines sensitivity to varying defect scales. Experimental validations on the Peking University Market PCB (PKU-Market-PCB) and DeepPCB datasets demonstrate that the proposed model surpasses popular object detection networks, achieving a 42.9% reduction in parameters and a 22.4% decrease in computational complexity compared to the baseline model. These results establish EffNet-PCB as a robust, efficient solution for industrial PCB defect detection.}, archive = {J_EAAI}, author = {Yingqiang Hou and Xindong Zhang}, doi = {10.1016/j.engappai.2025.110375}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110375}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight and high-accuracy framework for printed circuit board defect detection}, volume = {148}, year = {2025}, }
• (2025). Semi-supervised federated learning for collaborative security threat detection in control system for distributed power generation. EAAI, 148, 110374. (www)

  @article{ , abstract = {In response to the frequent occurrence of security incidents in existing control system for distributed power generation, the current security threat detection technologies face issues such as low detection accuracy, as well as inadequate data utilization due to objective reasons such as the high cost and low efficiency of manual data annotation, and the unreliability of automatic pseudo-annotation by models. we propose a semi-supervised federated learning-based security threat detection framework called federated uncertainty-aware pseudo-label selection (FedUPS). Firstly, this method utilizes Convolutional Neural Networks (CNNs) to effectively capture threats in control system for distributed power generation. Secondly, considering the characteristics of data spanning across devices and regions, we introduce federated learning methods to ensure the privacy and security of local data. Furthermore, to address the issue of unlabeled data within the system, we employ an Uncertainty-aware Pseudo-label Selection (UPS) framework to pseudo-label unlabeled data, ensuring the credibility of these pseudo-labels and incorporating them into the model training process. Lastly, in response to the dynamic changes in sample size and local model accuracy, we design a dynamic weighted aggregation method to optimize parameter updates and the model training process. Extensive experimental results demonstrate that our proposed method achieves an improvement in the optimal accuracy rate ranging from 2.46% to 12.28% within 50 iterations of training on the Network-Based Internet of Things (N-BaIoT) dataset. This method exhibits both effectiveness and practicality in detecting security threats in distributed power regulation systems.}, archive = {J_EAAI}, author = {Yuan Li and Yuancheng Li}, doi = {10.1016/j.engappai.2025.110374}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110374}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised federated learning for collaborative security threat detection in control system for distributed power generation}, volume = {148}, year = {2025}, }
• (2025). A plug-and-play fully on-the-job real-time reinforcement learning algorithm for a direct-drive tandem-wing experiment platforms under multiple random operating conditions. EAAI, 148, 110373. (www)

  @article{ , abstract = {This study addresses the motion control problem of the Direct-Drive Tandem-Wing Experiment Platform (DDTWEP), focusing on designing effective direct and transitional operating strategies for pitch, roll, and yaw under nonlinear, unsteady aerodynamic interference caused by high-frequency oscillations and closely spaced tandem wings by leveraging advanced artificial intelligence (AI) techniques. The Concerto Reinforcement Learning Extension (CRL2E) algorithm, a novel AI approach, is proposed to tackle this challenge, featuring the innovative Physics-Inspired Rule-Based Policy Composer strategy and experimental validation. The results demonstrate that the CRL2E algorithm maintains safety and efficiency throughout the training process, even with randomly initialized policy weights. In DDTWEP's plug-and-play, fully on-the-job motion control problem, the algorithm achieves a performance improvement of at least fourteen-fold and up to sixty-six-fold within the first five hundred interactions compared to Soft Actor-Critic (SAC), Proximal Policy Optimization (PPO), and Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithms. Furthermore, to further verify the rationality and performance of the module and algorithm design, this study introduces two perturbations: Time-Interleaved Capability Perturbation and Composer Perturbation, and develops multiple algorithms for comparative experiments. The experimental results show that compared to existing Concerto Reinforcement Learning (CRL) frameworks, the CRL2E algorithm achieves an 8.3%–60.4% enhancement in tracking accuracy, a 36.11%–57.64% improvement in convergence speed over the CRL with Composer Perturbation algorithm, and a 43.52%–65.85% improvement over the CRL with Time-Interleaved Capability Perturbation and Composer Perturbation algorithms, indicating the rationality of the CRL2E algorithm design. Regarding generalizability, the CRL2E algorithm demonstrates significant applicability in quadrotor flight control, highlighting its potential versatility. From a technical affinity perspective, the CRL2E algorithm is well-suited for integrating pretraining techniques, demonstrating excellent safety and efficiency in addressing cross-task plug-and-play and fully on-the-job fine-tuning problems. Regarding deplorability, hardware requirements were analyzed through ten thousand runs on diverse edge computing platforms, computational models, and operating systems to guide real-world deployment. Based on the experimental results, a real-time hardware-in-the-loop simulation system was constructed to validate the algorithm's effectiveness under realistic conditions. Additionally, an innovative yaw mechanism and its corresponding system model are introduced in this study to enhance the complexity of the system dynamics. These contributions provide valuable insights for addressing motion control challenges in complex mechanical systems.}, archive = {J_EAAI}, author = {Zhang Minghao and Song Bifeng and Yang Xiaojun and Wang Liang}, doi = {10.1016/j.engappai.2025.110373}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110373}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A plug-and-play fully on-the-job real-time reinforcement learning algorithm for a direct-drive tandem-wing experiment platforms under multiple random operating conditions}, volume = {148}, year = {2025}, }
• (2025). Traffic congestion recognition based on convolutional neural networks in different scenarios. EAAI, 148, 110372. (www)

  @article{ , abstract = {Traffic congestion has become a common problem worldwide, leading to the waste of time and resources. Accurate recognition of traffic congestion is crucial to solve these problems. Traditional traffic congestion recognition methods can only be accurate in a specific scenario. To solve the generalization of the model for traffic congestion recognition under different lighting conditions and weather. This paper proposes a traffic congestion recognition framework based on convolutional neural networks to recognize traffic congestion in different scenarios such as daytime, nighttime, rainy days, and foggy days. Since foggy obscure critical information in traffic images, and sensitivity analysis reveals that foggy conditions have a significant impact on traffic images, this paper proposes a Residual Attention-based Gated Fusion Defogging Network (RAGFNet) to process foggy traffic images. You Only Look Once (YOLO) demonstrates outstanding performance and efficiency in the field of image recognition, so the eighth version of You Only Look Once combined with local feature extraction (LFE-YOLOv8) is used to recognize traffic images. The model we used has the highest accuracy of 98.2% compared to other mainstream models, Recall is 97.7% and F1 value is 98.6%. Finally, to validate the applicability of the model in Intelligent Transportation Systems. A real-time evaluation was performed using Frames Per Second (FPS); congested areas were visualized using Gradient-weighted Class Activation Mapping (Grad-CAM) to explain the model and make it easier for traffic engineers to understand the model's decisions.}, archive = {J_EAAI}, author = {Chao Wang and Qiang Shang and Kun Liu and Wenxue Zhang}, doi = {10.1016/j.engappai.2025.110372}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110372}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Traffic congestion recognition based on convolutional neural networks in different scenarios}, volume = {148}, year = {2025}, }
• (2025). Efficient quantized transformer for atrial fibrillation detection in cross-domain datasets. EAAI, 148, 110371. (www)

  @article{ , abstract = {Atrial Fibrillation (AF) detection from electrocardiogram (ECG) signals has become vital for early diagnosis and continuous management in personal healthcare. Developing machine learning models for AF classification is challenging due to the various sources of ECG signals across datasets, which affects model generalization. To address this issue, we have proposed a lightweight transformer-based model using two public datasets and have evaluated its cross-domain generalization. The model has been further tested on a distinct dataset from the China Physiological Signal Challenge (CPSC) 2018. As an additional validation step, we have also tested it on real-world ECG data collected from wearable home monitoring devices, emphasizing its adaptability to diverse recording conditions. Despite these challenges, our transformer model achieves an accuracy of 85.6% in wearable data, significantly improving AF detection by analyzing complex ECG patterns and helping early diagnosis and continuous management of AF. Additionally, we have implemented a quantization technique that compresses the model to low bit precision for deployment on resource-constrained devices. The model, quantized to 3 bits for both weights and activations, maintains its accuracy while achieving a model size that is more than 9x smaller. Moreover, a fully binarized version with 5-bit quantized activations offers a 19x reduction in size, with only a minimal accuracy decrease of less than 2%.}, archive = {J_EAAI}, author = {Maedeh H. Toosi and Mahdi Mohammadi-nasab and Siamak Mohammadi and Mostafa E. Salehi}, doi = {10.1016/j.engappai.2025.110371}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110371}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient quantized transformer for atrial fibrillation detection in cross-domain datasets}, volume = {148}, year = {2025}, }
• (2025). Event-triggered predefined-time tracking control for high-order nonlinear systems with time-varying actuator failures and uncertain disturbances. EAAI, 148, 110368. (www)

  @article{ , abstract = {For a class of higher-order nonlinear system control problems with time-varying actuator failures and external disturbances, this paper designs efficient control strategies that allow the system to be stabilized in a predefined time. First, for such systems, this paper designs an effective predefined-time control strategy using the backstepping control method combined with the adaptive radial basis neural network technique, which makes the stabilization time of the system simple and adjustable. Secondly, while using the command filtering technique to solve the “complexity explosion” problem in the design of controllers for high-order nonlinear systems, this paper designs a novel predefined-time filtering error compensation mechanism to eliminate the impact of filtering errors on the stability of the system. Finally, an event-triggered mechanism is introduced, which effectively saves the communication resources. The effectiveness of the control strategy proposed in this paper is demonstrated by the simulation experiments.}, archive = {J_EAAI}, author = {Yue Wang and Jie Gao and Junchan Zhao and Xingyu Wu}, doi = {10.1016/j.engappai.2025.110368}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110368}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Event-triggered predefined-time tracking control for high-order nonlinear systems with time-varying actuator failures and uncertain disturbances}, volume = {148}, year = {2025}, }
• (2025). Accurate solar power prediction with advanced hybrid deep learning approach. EAAI, 148, 110367. (www)

  @article{ , abstract = {Solar power, a clean and sustainable energy source, is becoming increasingly vital for meeting global electricity demand, yet its inherent unpredictability presents a significant challenge for power system stability and critical applications. Accurate solar power prediction is crucial for efficient energy management, grid stability, and seamless integration of renewable sources into existing energy networks. This study introduces a novel hybrid deep learning approach that leverages the complementary strengths of a recurrent neural network (RNN), a transformer model, and a long short-term memory (LSTM) network to improve solar power prediction accuracy. The proposed model incorporates concatenating process with a multi-head self-attention mechanism, enabling it to capture intricate dependencies across varying temporal scales, thereby enhancing time series forecasting performance. To evaluate the robustness of the hybrid model, solar power prediction was conducted under diverse and unpredictable weather scenarios, including sunny, cloudy, and rainy conditions. The experimental results revealed that the proposed RNN-Transformer-LSTM approach significantly outperforms traditional methods, achieving superior prediction accuracy. Key performance metrics including mean absolute error (0.0687), mean square error (0.0084), root mean square error (0.0917), and normalized root mean square error (0.0306), indicate that the hybrid model consistently delivers highly reliable predictions. This research demonstrates the value of advanced deep learning models in renewable energy forecasting, improving grid reliability, optimizing power distribution, and supporting the shift toward sustainable energy.}, archive = {J_EAAI}, author = {Dongran Song and Muhammad Shams Ur Rehman and Xiaofei Deng and Zhao Xiao and Javeria Noor and Jian Yang and Mi Dong}, doi = {10.1016/j.engappai.2025.110367}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110367}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Accurate solar power prediction with advanced hybrid deep learning approach}, volume = {148}, year = {2025}, }
• (2025). Dual-stage dynamic hierarchical attention framework for saliency-aware explainable diabetic retinopathy grading. EAAI, 148, 110364. (www)

  @article{ , abstract = {Diabetic Retinopathy (DR) is a serious complication of diabetes, which leads to severe vision impairment if left untreated. For effective treatment, it is crucial to detect the disease early and grade it accurately. In recent years, convolutional neural networks have shown promising results in automated DR grading, yet their black-box nature challenges their interpretability. To address this, we propose a novel deep learning framework leveraging hierarchical attention refinement to dynamically highlight lesion salient regions in retinal images. The proposed model employs a Residual Network-18 backbone network to capture basic semantic feature representation from retinal fundus images, followed by a channel-wise and spatial weighed attention encoding to generate an initial saliency representation. This is further enhanced with a hierarchical cross attention mechanism to produce enriched saliency maps. Concurrently, these saliency maps guide the final decision-making process, thereby enhancing interpretability and assisting in accurate DR grading. Experimental results validate the efficacy of the proposed model in enhancing the performance of DR grading across multiple evaluation metrics. Further, quantitative and qualitative analysis of the generated saliency maps demonstrated substantial enhancements in pinpointing lesion areas within fundus images, leading to enhanced explainability and diagnostic accuracy of the model’s predictions.}, archive = {J_EAAI}, author = {Shilpa Elsa Abraham and Binsu C. Kovoor}, doi = {10.1016/j.engappai.2025.110364}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110364}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-stage dynamic hierarchical attention framework for saliency-aware explainable diabetic retinopathy grading}, volume = {148}, year = {2025}, }
• (2025). Improving the magnetic resonance images super-resolution with a dual-channel enhancement model incorporating complementary information. EAAI, 148, 110359. (www)

  @article{ , abstract = {Although significant progress has been made in image super-resolution using artificial intelligence, achieving high-quality super-resolution for magnetic resonance (MR) images remains challenging due to their unique imaging principles and processes. Inspired by MR parallel imaging, we propose a novel concept to improve the MR image super-resolution quality by leveraging the complementary spatial information inherently contained in multi-channel receive coils. A new MR image degradation model was developed to generate the training dataset that complies with the parallel imaging and Sensitivity Encoding (SENSE) reconstruction. A dual-channel enhancement model, named sensitivity encoding based super-resolution (SenseSR), is then devised with the main channel processing the single low-resolution image and the enhancement channel processing the multiple images from each coil channel. SenseSR is mainly featured with cascaded double enhancement blocks that can extract deeper features of the multiple coil-channel images and fuse them together into the main channel. Experiments were performed to test the performance and compare it with other benchmark models. The results demonstrate a significant improvement in MR image super-resolution quality, with an enhancement of peak signal-to-noise ratio ranging from 0.5 to 6.5 decibels (dB). Further experiments with different testing datasets and MR images collected in-situ demonstrated that SenseSR also has good generalization capability and robustness, indicating its potential for clinical applications. The code is available at https://github.com/MISR-Lab/SenseSR .}, archive = {J_EAAI}, author = {Chunqiao He and Hang Liu and Yue Shen and Deyin Zhou and Lin Wu and Hailin Ma and Tao Zhang}, doi = {10.1016/j.engappai.2025.110359}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110359}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving the magnetic resonance images super-resolution with a dual-channel enhancement model incorporating complementary information}, volume = {148}, year = {2025}, }
• (2025). Dynamic window sampling strategy for image captioning. EAAI, 148, 110358. (www)

  @article{ , abstract = {The task of image captioning aims to transform the visual information from images into semantically accurate and grammatically correct textual descriptions. In this paper, we concentrate on improving the word sampling process to enhance the training effectiveness of image captioning models. We first demonstrate that using beam search to sample sentences during reinforcement learning training performs worse than the probability-based sampling method. We then find that the probability-based sampling is prone to selecting inaccurate words due to an unstable probability distribution. Since current mainstream reinforcement learning training relies on sentence-level feedback, it is difficult to perceive word-level effects. Consequently, irrelevant words can interfere with learning the entire sentence. To address this issue, we propose a dynamic window sampling strategy for image captioning. The core idea is to dynamically determine the word candidate pool, i.e., the sampling window, based on overall prediction confidence. Compared to sampling words over the entire vocabulary, our approach maintains diversity while avoiding the sampling of irrelevant words as much as possible. Extensive experiments on benchmark datasets show that the proposed dynamic window sampling strategy can significantly improve model performance. Specifically, our method achieved CIDEr scores of 144.9% (single model) and 147.9% (ensemble of 4 models) on the offline test, and 142.1% (c5) and 144.0% (c40) on the official online test server. Overall, this paper contributes to artificial intelligence research by studying sampling strategies to improve image captioning performance. The source code is available at https://github.com/792218/DWSS.}, archive = {J_EAAI}, author = {Zhixin Li and Jiahui Wei and Tiantao Xian and Canlong Zhang and Huifang Ma}, doi = {10.1016/j.engappai.2025.110358}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110358}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic window sampling strategy for image captioning}, volume = {148}, year = {2025}, }
• (2025). DC-net: Decomposing and coupling saliency map for lesion segmentation in ultrasound images. EAAI, 148, 110355. (www)

  @article{ , abstract = {Accurate lesion segmentation in ultrasound images faces the unique challenge that adjacent tissues ( i.e. , background) share similar intensity and often exhibit richer texture patterns than the lesion region ( i.e. , foreground). This work presents a decomposition-coupling network, called DC-Net, to deal with this challenge in a (foreground-background) saliency map disentanglement-fusion manner. The DC-Net consists of decomposition and coupling subnets, and the former preliminarily disentangles original image into foreground and background saliency maps, followed by the latter for accurate segmentation under the assistance of saliency prior fusion. The coupling subnet involves three aspects of fusion strategies, including: (1) regional feature aggregation (via differentiable context pooling operator in the encoder) to adaptively preserve local contextual details with the larger receptive field during dimension reduction; (2) relation-aware representation fusion (via cross-correlation fusion module in the decoder) to efficiently fuse low-level visual characteristics and high-level semantic features during resolution restoration; (3) dependency-aware prior incorporation (via coupler) to reinforce foreground-salient representation with the complementary information derived from background representation. Furthermore, a harmonic loss function is introduced to encourage the network to focus more attention on low-confidence and hard pixels. The proposed method is evaluated on two ultrasound lesion segmentation tasks, achieving better performance than second best methods ( i.e. , dice similarity coefficient: 73.96 ± 0.36 (+4.18%)/76.25 ± 1.89 (+3.88%), mean intersection over union/jaccard index: 64.26 ± 0.48 (+5.07%)/65.28 ± 2.24 (+6.46%), precision: 75.82 ± 1.16 (+8.07%)/79.35 ± 1.92 (+3.07%), 95% hausdorff distance: 5.44 ± 0.10 (+5.39%)/6.22 ± 0.26 (+5.18%) on breast/thyroid ultrasound). Our proposed DC-Net can adapt to multiple ultrasound segmentation tasks and achieve remarkable performance improvement. Our code is available at https://github.com/smu-myx/DC-Net .}, archive = {J_EAAI}, author = {Zhenyuan Ning and Yixiao Mao and Xiaotong Xu and Qianjin Feng and Shengzhou Zhong and Yu Zhang}, doi = {10.1016/j.engappai.2025.110355}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110355}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DC-net: Decomposing and coupling saliency map for lesion segmentation in ultrasound images}, volume = {148}, year = {2025}, }
• (2025). Forecasting stock market time series through the integration of bee colony optimizer and multivariate empirical mode decomposition with extreme gradient boosting regression. EAAI, 148, 110353. (www)

  @article{ , abstract = {Stock price prediction is essential for the optimization of investment strategies, the mitigation of risks, and the facilitation of informed decision-making. Accurate forecasting is exceedingly difficult due to the nonlinear, nonstationary, and volatile nature of stock prices. This complexity is frequently not adequately addressed by conventional methods, underscoring the necessity of sophisticated hybrid models. This study uses stock price data from the Standard & Poor's 500 Index to develop a novel hybrid model, Multivariate Empirical Mode Decomposition-Artificial Bee Colony-Extreme Gradient Boosting Regression. Extreme Gradient Boosting Regression captures intricate patterns in the data, Artificial Bee Colony optimizes the hyperparameters of Extreme Gradient Boosting Regression to enhance model robustness, and Multivariate Empirical Mode Decomposition decomposes complex financial time-series data into manageable intrinsic mode functions. Close price, Momentum, Simple Moving Average, Moving Average Convergence Divergence, Relative Strength Index, and Trading volume comprise the dataset. These features are indispensable for identifying both short-term fluctuations and long-term trends. The presented model is significantly more effective than traditional models, as evidenced by its test set coefficient of determination of 0.9914. The proposed model's robustness is confirmed by comprehensive 5-fold cross-validation and ablation studies, which also emphasize the significance of its integrated components. Furthermore, the model's adaptability is further illustrated by its ability to generalize to other markets, as evidenced by its coefficient of determination values exceeding 0.99 on three other indexes. These results underscore the potential of artificial intelligence-driven hybrid models to enhance stock price forecasting, offering useful insights for policymakers, financial analysts, and investors.}, archive = {J_EAAI}, author = {Xuefeng Liu and Zhixin Wu and Jiayue Xin}, doi = {10.1016/j.engappai.2025.110353}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110353}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forecasting stock market time series through the integration of bee colony optimizer and multivariate empirical mode decomposition with extreme gradient boosting regression}, volume = {148}, year = {2025}, }
• (2025). A weighted graph network-based method for combining conflicting evidence. EAAI, 148, 110351. (www)

  @article{ , abstract = {Information fusion technology is crucial in intricate information systems, and Dempster–Shafer evidence(DSE) theory plays a significant role in it. However, most of the current research focuses on improving the conflict measurement method of high-conflict evidence in the DSE theory framework, while ignoring the comprehensive consideration of multiple conflicts of complex information. Considering the generality of graph network to complex system modeling, novel evidence measurement factors (EMF) and weighted Graph Convolution Network Dempster–Shafer evidence (wGCNDS) combination method, are proposed to optimize the combination of conflict evidence from the perspective of graph network. By constructing a weighted graph network, information transmission is realized and information fusion of associated nodes is completed. Numerical examples and real datasets verify the effectiveness and performance of wGCNDS.}, archive = {J_EAAI}, author = {Jinjian Lin and Kai Xie}, doi = {10.1016/j.engappai.2025.110351}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110351}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A weighted graph network-based method for combining conflicting evidence}, volume = {148}, year = {2025}, }
• (2025). Using physics-informed neural networks for solving navier-stokes equations in fluid dynamic complex scenarios. EAAI, 148, 110347. (www)

  @article{ , abstract = {Navier-Stokes equations used to model fluid dynamic processes are fundamental to address several real-world problems related to energy production, aerospace applications, automotive design, industrial process, etc. However, since in most cases they do not admit any analytical solution, numerical simulations are required in industrial contexts to assess fluid dynamic behaviors in specific setups. Computational Fluid Dynamics (CFD) methods, like those using finite volume or element approaches, are exploited to find Navier-Stokes solutions and carry out simulations. However, such methods require expensive hardware resources, relevant computational times, and manual efforts for the definition of dense meshes on which equations are evaluated iteratively for each time step of the simulation. Physics-Informed Neural Networks (PINNs), which are deep neural networks where physical laws are directly embedded into the training process, offer a promising approach for solving Navier-Stokes equations, thus alleviating hardware and time requirements. PINNs bypass some CFD limitations by using neural networks to produce solutions based on governing equations, thus reducing the need for large datasets, dense meshing, and iterative estimation over time. This paper evaluates the application of PINNs in near real-world scenarios, while considering various geometries. The study focuses on the achieved accuracy, by comparing PINN estimates with CFD solutions obtained via OpenFOAM, and the required training times; this includes evaluating different neural network architectures, activation functions, and numbers of sampling points. Additionally, several training strategies such as fine-tuning, multi-resolution learning, and parametrized training are proposed to enhance efficiency and obtain speed up. Results demonstrate that PINNs can achieve comparable accuracy to CFD methods (with a velocity magnitude mean absolute error inferior to 10 − 2 ) and significantly reduce computational costs. Our findings demonstrated that with appropriate training techniques PINNs can be effectively used in industrial applications requiring rapid and accurate fluid dynamic simulations, thus paving the way for their broader adoption in practical engineering problems.}, archive = {J_EAAI}, author = {Tommaso Botarelli and Marco Fanfani and Paolo Nesi and Lorenzo Pinelli}, doi = {10.1016/j.engappai.2025.110347}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110347}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Using physics-informed neural networks for solving navier-stokes equations in fluid dynamic complex scenarios}, volume = {148}, year = {2025}, }
• (2025). Algorithm for surface flow velocity measurement in trunk canal based on improved YOLOv8 and DeepSORT. EAAI, 148, 110344. (www)

  @article{ , abstract = {The velocity measurement of trunk canal and river plays an important role in agriculture and forestry irrigation scheduling, water resources management and flood prediction. Particle flow measurement technology can realize non-contact and high-precision flow measurement, but in practical application, the particle size is small, the shape is different and the dynamic change brings great challenges to the application of this method. To solve these problems, this paper proposed the surface velocity measurement method of trunk canal based on improved YOLOv8(You Only Look Once Version 8) and DeepSORT(Deep Simple Online and Realtime Tracking), and introduced tiny detection layer and channel attention mechanism to improve YOLOv8's detection capability of small targets. In DeepSORT, IBN-Net(Intent-Based Networking-Network) network structure and GIoU(Generalized Intersection over Union) matching are introduced to solve the problem of discontinuity or loss of target tracking in complex cases, which improves the accuracy and robustness of target tracking. The experimental results show that the improved YOLOv8 improves AP(Average Precision) and mAP(mean Average Precision) by nearly 5% and 0.2% respectively. The performance of the improved DeepSORT has been improved across the board, especially IDP and MOTA, which have improved by 25.2% and 5.6% respectively. The algorithm also has good accuracy in actual velocity measurement.}, archive = {J_EAAI}, author = {Yuhui Zhou and Xiaojie Wu and Yiming Li and Huimin Sun and Di Fan}, doi = {10.1016/j.engappai.2025.110344}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110344}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Algorithm for surface flow velocity measurement in trunk canal based on improved YOLOv8 and DeepSORT}, volume = {148}, year = {2025}, }
• (2025). Domain generalization for zero-calibration brain–computer interfaces with knowledge distillation-based phase invariant feature extraction. EAAI, 148, 110340. (www)

  @article{ , abstract = {The distribution shift of electroencephalography (EEG) data causes poor generalization of brain–computer interfaces (BCIs) in unseen domains. Some methods try to tackle this challenge by collecting a portion of user data for calibration. However, it is time-consuming, mentally fatiguing, and user-unfriendly. To achieve zero-calibration BCIs, most studies employ domain generalization techniques to learn invariant features across different domains in the training set. However, they fail to fully explore invariant features within the same domain, leading to limited performance. In this paper, we present an novel method to learn domain-invariant features from both inter-domain and intra-domain perspectives. For intra-domain invariant features, we propose a knowledge distillation framework to extract EEG phase-invariant features within one domain. As for inter-domain invariant features, correlation alignment is used to bridge distribution gaps across multiple domains. Experimental results on three public datasets validate the effectiveness of our method, showcasing state-of-the-art performance. To the best of our knowledge, this is the first domain generalization study that exploit Fourier phase information as an intra-domain invariant feature to facilitate EEG generalization. More importantly, the zero-calibration BCI based on inter- and intra-domain invariant features has significant potential to advance the practical applications of BCIs in real world. (The code is available on https://github.com/ZilinL/KnIFE ).}, archive = {J_EAAI}, author = {Zilin Liang and Zheng Zheng and Weihai Chen and Xinzhi Ma and Zhongcai Pei and Xiantao Sun}, doi = {10.1016/j.engappai.2025.110340}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110340}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Domain generalization for zero-calibration brain–computer interfaces with knowledge distillation-based phase invariant feature extraction}, volume = {148}, year = {2025}, }
• (2025). STDDAE: Identifying spatial domains in spatial transcriptomics by dual denoising autoencoder with attention mechanism. EAAI, 148, 110338. (www)

  @article{ , abstract = {Spatial transcriptomics provides a novel perspective for comprehending the intricate relationship between tissue structure and function, as well as for discovering new cell types and subtypes. However, it remains a significant challenge to accurately identify spatial domains with similar gene expression, which requires efficient combination of gene expression data, histology image information, and spatial location. To address this challenge, a novel dual denoising autoencoder with attention mechanism (STDDAE) is proposed. STDDAE integrates gene expression data, histology image information and spatial location, and the decoder consists of a master decoder and a follower decoder, which are jointly optimized to generate low-dimensional latent embeddings for precise spatial domain identification. The performance of STDDAE was evaluated across four datasets with varying resolutions and platforms. The experimental findings validated that STDDAE outperformed other cutting-edg methods in spatial domain identification, trajectory inference, and data denoising. Additionally, STDDAE successfully detected differentially expressed genes within identified spatial domains, which may be valuable in disease diagnosis, prognostic assessment, and treatment selection.}, archive = {J_EAAI}, author = {Yue Gao and Ying-Lian Gao and Cui-Na Jiao and Xu-Ran Dou and Feng Li and Jin-Xing Liu}, doi = {10.1016/j.engappai.2025.110338}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110338}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {STDDAE: Identifying spatial domains in spatial transcriptomics by dual denoising autoencoder with attention mechanism}, volume = {148}, year = {2025}, }
• (2025). Multiobjective optimization deep reinforcement learning for dependent task scheduling based on spatio-temporal fusion graph neural network. EAAI, 148, 110337. (www)

  @article{ , abstract = {Task scheduling problem in distributed high-performance computing systems (HPCS) is a complex problem that involves a large number of processors and a wide range of resources. Due to the heterogeneity of processors, it is difficult to effectively manage the scheduling of dependent tasks in a non-preemptive heterogeneous system. To address this issue, we propose a Multiobjective Optimization Deep Reinforcement Learning-based (MODRL) task scheduling framework to assign suitable processors for an input task sequence, which utilizes a spatio-temporal graph neural network to extract potential temporal and spatial characteristics and uses a Set Transformer to merge the current cluster features. Besides, we develop a multiobjective optimized training method to improve the generalization of the scheduling model for both goals on overall execution time and average resource occupation. The experimental results conducted on a real-world dataset demonstrate the effectiveness and competitiveness of the proposed method in comparison to well-known existing approaches in terms of both overall makespan and improved performance in load balancing.}, archive = {J_EAAI}, author = {Zhi Wang and Wenhan Zhan and Hancong Duan and Hualong Huang}, doi = {10.1016/j.engappai.2025.110337}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110337}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multiobjective optimization deep reinforcement learning for dependent task scheduling based on spatio-temporal fusion graph neural network}, volume = {148}, year = {2025}, }
• (2025). Q-rung orthopair fuzzy multi-criteria decision-making method for internet of things platforms selection. EAAI, 148, 110336. (www)

  @article{ , abstract = {The assessment and development of Internet of Things (IoT) platforms are closely linked, as they are crucial in ensuring that the system platform operates efficiently and effectively while providing new horizons in engineering management optimization. The IoT platforms enable precise coordination of diverse IoT elements, which helps simplify operational costs and improve decision-making processes. However, the evaluation data is often filled with uncertain data. Additionally, the weight information of decision-makers has a tremendous impact on the evaluation of IoT platforms. This paper aims to explore an integrated information fusion method and provide a trustworthy decision-making framework for assessing IoT platforms. This method employs a combination of q-rung orthopair fuzzy set (q-ROFS), multi-criteria decision-making (MCDM), score function, aggregation operator, and combined weight to explore information fusion in evaluating IoT platforms. Firstly, a new q-rung orthopair fuzzy (q-ROF) similarity measure considering hesitancy is proposed. Then, the q-ROF score function based on similarity measure is introduced. Moreover, a non-linear combined weight by merging subjective weight based on expert and objective weight based on weights by envelope and slope (WENSLO) is given. Later, q-ROF generalized mean aggregation operator is discussed with its proven properties. Experimental results demonstrate that our method yields rational and intuitive evaluation outcomes, comprehensively incorporating both subjective and objective weight information.}, archive = {J_EAAI}, author = {Xindong Peng and Linhui Yu and Xiu Wu}, doi = {10.1016/j.engappai.2025.110336}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110336}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Q-rung orthopair fuzzy multi-criteria decision-making method for internet of things platforms selection}, volume = {148}, year = {2025}, }
• (2025). Footprint of uncertainty in the context of bow-tie risk tool using fuzzy logic. EAAI, 148, 110335. (www)

  @article{ , abstract = {The Bow-tie method is a qualitative risk analysis tool known for its effectiveness in visualizing the relationships between causes, barriers, and consequences. In this paper, we present a novel approach to quantifying Bow-tie diagram outcomes by incorporating uncertainty in both input and output parameters. We utilize fuzzy inference to aggregate frequencies and probabilities of failure on demand (PFD). Two fuzzy logic systems are tested: the Type 1 Fuzzy Logic System (T1FLS), and the Interval Type 2 Fuzzy Logic System (IT2FLS), applied here for the first time. The primary innovation of this work lies in the application of IT2FLS, which introduces the concept of the footprint of uncertainty (FOU) to better account for uncertainty in the membership functions of linguistic variables. To validate these models, both were adjusted using expert knowledge to replicate the behavior of the Bow-tie combined with Layers of Protection Analysis (probabilistic model). Simulations compared the proposed methods to the probabilistic model, with sensitivity analyses examining variations in protection barrier failure probabilities. The T1FLS achieved a normalized root mean square error (NRMSE) of 9.54%, while the IT2FLS reached 12.82%. For the normalized root mean square logarithmic error (NRMSLE), T1FLS yielded 4.65%, and IT2FLS 6.30%. The methods showed 87.32% similarity in ranking protection barrier sensitivity indices. The findings suggest both fuzzy systems exhibit strong potential for accurately representing complex systems with inherent uncertainties, making them valuable tools for risk analysis.}, archive = {J_EAAI}, author = {V.A.O. Silva and R. Santana and R.I. Tsukada and S.S.V. Vianna and F.V. Silva}, doi = {10.1016/j.engappai.2025.110335}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110335}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Footprint of uncertainty in the context of bow-tie risk tool using fuzzy logic}, volume = {148}, year = {2025}, }
• (2025). Transfer learning approach to modeling multichannel gate-all-around nanosheet field-effect transistors under work function fluctuations. EAAI, 148, 110322. (www)

  @article{ , abstract = {Deep learning (DL) has significantly advanced various industries, including semiconductors, by providing sophisticated methods for analyzing emerging device data. Transfer learning (TL), a prominent DL topology, leverages knowledge from a pre-trained model to improve the performance of a specific target task. This study aims to apply TL to predict the impact of work function fluctuation (WKF) on the variability of threshold voltage ( σ V T H ) in gate-all-around silicon nanosheet field-effect transistors (GAA Si NS FETs). We used TL to transfer knowledge from models trained on 1-channel (1ch) GAA Si NS FETs to more complex 4-channel (4ch) counterparts. The prediction accuracy is evaluated by using the mean square error( M S E ), root mean square error ( R M S E ), and the mean absolute error( M A E ) metrics. Our TL approach achieved an error rate of less than 1%, demonstrating high accuracy in predicting WKF effects on σ V T H of 4ch GAA Si NS FETs using only 500 samples. This study presents a pioneering application of TL using a one-dimensional (1D) convolutional neural network (CNN) long short-term memory (LSTM) topology in semiconductor device technology. The results effectively address the challenge of predicting σ V T H variations in advanced GAA Si NS FETs because of WKF, which eliminates the need for extensive semiconductor device simulations.}, archive = {J_EAAI}, author = {Sagarika Dash and Yiming Li}, doi = {10.1016/j.engappai.2025.110322}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110322}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transfer learning approach to modeling multichannel gate-all-around nanosheet field-effect transistors under work function fluctuations}, volume = {148}, year = {2025}, }
• (2025). Multi-modal temporal action segmentation for manufacturing scenarios. EAAI, 148, 110320. (www)

  @article{ , abstract = {Industrial robots have become prevalent in manufacturing due to their advantages of accuracy, speed, and reduced operator fatigue. Nevertheless, human operators play a crucial role in primary production lines. This study focuses on the temporal segmentation of human actions, aiming to identify the physical and cognitive behavior of operators working alongside collaborative robots. While existing literature explores temporal action segmentation datasets, there is a lack of evaluation for manufacturing tasks. This work assesses six state-of-the-art action segmentation models using the Human Action Multi-Modal Monitoring in Manufacturing (HA4M) dataset, where subjects assemble an industrial object in realistic manufacturing scenarios. By employing Cross-Subject and Cross-Location approaches, the study not only demonstrates the effectiveness of these models in industrial settings but also introduces a new benchmark for evaluating generalization across different subjects and locations. The evaluation further includes new videos in simulated industrial locations, assessed with both fully and semi-supervised learning approaches. The findings reveal that the Multi-Stage Temporal Convolutional Network ++ (MS-TCN++) and the Action Segmentation Transformer (ASFormer) architectures exhibit high performance in supervised and semi-supervised learning settings, also using new data, particularly when trained with Skeletal features, advancing the capabilities of temporal action segmentation in real-world manufacturing environments. This research lays the foundation for addressing video activity understanding challenges in manufacturing and presents opportunities for future investigations.}, archive = {J_EAAI}, author = {Laura Romeo and Roberto Marani and Anna Gina Perri and Juergen Gall}, doi = {10.1016/j.engappai.2025.110320}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110320}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modal temporal action segmentation for manufacturing scenarios}, volume = {148}, year = {2025}, }
• (2025). Large language model-based planning agent with generative memory strengthens performance in textualized world. EAAI, 148, 110319. (www)

  @article{ , abstract = {This paper constructs a planning agent with generative memory (GM-Agent) based on Large Language Models (LLM) to address the complex planning problems in textualized world. GM-Agent perceives and makes decisions by representing the world as a quintuple, constructing a textualized world model (TWM), and generating memory. Its generative rewards and strategies can optimize the decision-making performance. Baseline results demonstrate that under the LandWar benchmark, compared with prompt-based generative agents and the traditional Reinforcement Learning from Human Feedback (RLHF) framework, GM-Agent can exhibit stronger continuous planning capabilities. The degree to which GM-Agent follows instructions has increased by 137.25%, and the degree to which its decisions are aligned with human experience has increased by 51.69%. The execution ability of mainline tasks and the convergence performance of the model demonstrate state-of-the-arts (SOTA). In comparisons with various benchmark artificial intelligences (AIs), GM-Agent performs excellently, especially showing obvious advantages in adversarial tasks, with a maximum 96.08% improvement in the performance of mainline tasks. Orthogonal experiments prove the effectiveness of the proposed reward scoring and advantage function. We have demonstrated the framework portability and application scenario generalization capabilities of GM-Agent in different LLM frameworks and benchmarks in different fields. It shows excellent performance in the complex field of embodied intelligence planning. However, there is still a certain gap compared with advanced technologies in scientific common sense and logical reasoning tasks. We have demonstrated the application potential of GM-Agent in multiple application scenarios, and found its deficiencies in terms of stability, observation subjectivity, feedback ability, and flexibility.}, archive = {J_EAAI}, author = {Junyang Liu and Wenning Hao and Kai Cheng and Dawei Jin}, doi = {10.1016/j.engappai.2025.110319}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110319}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Large language model-based planning agent with generative memory strengthens performance in textualized world}, volume = {148}, year = {2025}, }
• (2025). Enhanced few-shot object detection for remote sensing images based on target characteristics. EAAI, 148, 110315. (www)

  @article{ , abstract = {Few-shot object detection (FSOD) in remote sensing images (RSIs) is a challenging and hot issue due to the characteristics of targets in remote sensing images such as varying sizes, complex backgrounds, target occlusions, and unbalanced target categories. Two-stage fine-tuning approach (TFA) has been shown to be the most competitive approach for few-shot learning. However, previous methods of fine-tuning remote sensing images using single-feature improvements had a limited effect on the result. In this paper, we propose a novel few-shot object detection method that coordinates multiple methods to gradually improve detection accuracy by focusing on the characteristics of remote sensing targets. More specifically, our model contains three main components: a novel context-aware weighted feature fusion module (CA-WFFM) that enhances the discrimination between background and foreground structures in complex scenarios, an edge detection block (EDB), which combines handcrafted features, that enhances the stability of the network when the target is occluded or the sample is limited, and a selective random neighborhood oversampling (SRNO) strategy, that balances the base and novel category samples. These coordinated improvements were evaluated using the Drones In Optics Recognition (DIOR) and NWPU-VHR-10.v2 (VHR-10.v2) datasets with multiple training sample splits. Experimental results indicate that our method surpasses existing methods in novel class detection and also provides a new research approach for few-shot object detection. This involves designing algorithms based on target characteristics to enhance the overall detection capability of few-shot object detection models.}, archive = {J_EAAI}, author = {Jian Wang and Zeya Zhao and Jiang Shao and Xiaochun Zou and Xinbo Zhao}, doi = {10.1016/j.engappai.2025.110315}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110315}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced few-shot object detection for remote sensing images based on target characteristics}, volume = {148}, year = {2025}, }
• (2025). High-quality three-dimensional cartoon avatar reconstruction with gaussian splatting. EAAI, 148, 110305. (www)

  @article{ , abstract = {The growth of the augmented reality industry has increased demand for three-dimensional (3D) cartoon avatars, requiring expertise from computer graphics designers. Recent 3D Gaussian splatting methods have successfully reconstructed 3D avatars from videos, establishing them as a promising solution for this task. However, these methods primarily focus on real-world videos, limiting their effectiveness in the cartoon domain. In this paper, we present an artificial intelligence (AI)-based method for 3D avatar reconstruction from animated cartoon videos, addressing the physically unrealistic and unstructured geometries of cartoons, as well as the varying texture styles across frames. Our surface fitting module models the unstructured geometry of cartoon characters by integrating the surfaces observed from multiple views into a 3D avatar. We design a style normalizer that adjusts color distributions to reduce texture color inconsistencies in each frame of animated cartoons. Additionally, to better capture the simplified color distributions of cartoons, we design a frequency transform loss that focuses on low-frequency components. Our method significantly outperforms state-of-the-art methods, achieving approximately a 25% improvement in Learned Perceptual Image Patch Similarity (LPIPS) with a score of 0.052 over baselines across the Cartoon Neuman and ToonVid datasets, which comprise 10 videos with diverse styles and poses. Consequently, this paper presents a promising solution to meet the growing demand for high-quality 3D cartoon avatar modeling.}, archive = {J_EAAI}, author = {MinHyuk Jang and Jong Wook Kim and Youngdong Jang and Donghyun Kim and Wonseok Roh and InYong Hwang and Guang Lin and Sangpil Kim}, doi = {10.1016/j.engappai.2025.110305}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110305}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {High-quality three-dimensional cartoon avatar reconstruction with gaussian splatting}, volume = {148}, year = {2025}, }
• (2025). Dynamic prediction of sulfur dioxide concentration in a single-tower double-circulation desulfurization system based on chemical mechanism and deep learning. EAAI, 148, 110294. (www)

  @article{ , abstract = {With the fluctuation of load in coal-fired power plants, the sulfur dioxide (SO 2 ) concentration in the flue gas changes more and more frequently, due to the large delay and inertia characteristics of wet flue gas desulfurization (WFGD) systems, the SO 2 concentration in the flue gas emitted from the outlet is unstable. To accurately control the SO 2 emission concentration of the desulfurization system, a single-tower double-circulation wet flue gas desulfurization (SD-WFGD) system outlet SO 2 concentration prediction model was established. Firstly, considering the application of the established model in control systems and system optimization operation schemes, it is necessary to enhance the interpretability of the model. Therefore, based on the chemical reaction process of SO 2 absorption in the desulfurization system, SO 2 mechanism prediction models for the outlet of the absorption tower and absorber feed tank (AFT) tower were established, and parameter identification was carried out using quantum particle swarm optimization algorithm (QPSO) and historical operation data of the power station. Secondly, to improve the prediction accuracy of the mechanism model, a Convolutional Neural Network (CNN)-Long Short Term Memory (LSTM)-Attention data compensation model was established. In this process, the difficulty of manually adjusting the hyperparameters of the deep learning model was considered, during the model training process, the rime optimization algorithm was used to optimize the model hyperparameters in real time. To enable the data compensation model to obtain more data feature information, the input data was decomposed using the Variational Mode Decomposition (VMD) method and the same frequency modes were combined and reconstructed. After model training, different modal model outputs were superimposed to obtain the final compensation data. Finally, the mechanism model and data compensation model were combined to obtain a hybrid prediction model for SO 2 concentration. The model validation results showed that the error indicators root mean squared error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R 2 ) of the model are 0.2739 mg/m 3 , 0.8267%, and 0.9999, respectively.}, archive = {J_EAAI}, author = {Ruilian Li and Deliang Zeng and Tingting Li and Yan Xie and Yong Hu and Guangming Zhang}, doi = {10.1016/j.engappai.2025.110294}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110294}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic prediction of sulfur dioxide concentration in a single-tower double-circulation desulfurization system based on chemical mechanism and deep learning}, volume = {148}, year = {2025}, }
• (2025). Enhancing fine-grained geographic named entity recognition by multi-scale siamese reconstruction network. EAAI, 148, 110287. (www)

  @article{ , abstract = {Fine-grained geographic named entity recognition involves the identification of precise locations within geographic text. Existing methods face challenges in effectively addressing this task, namely: (1) the complexity and long-dependency nature of various fine-grained locations, which pose difficulties for recognition, and (2) the presence of numerous out-of-vocabulary fine-grained locations that are not present in the training data. In this paper, we propose a Multi-scale Siamese Reconstruction Network (MSRN) to tackle these challenges. To address the first challenge, MSRN employs a multi-scale convolutional network to aggregate interactions among coarse-grained and fine-grained span features. To tackle the second challenge, MSRN introduces a siamese reconstruction network that corrupts the representation and then reconstructs it, preventing the model from relying on rote memorization of biased locations in the training data. This approach enhances the model’s understanding of entity context. We conduct experiments on four real-world fine-grained geographic named entity recognition datasets, which contain a variety of complex long fine-grained locations and geographic entities not present in the training data. These experiments illustrate the challenges commonly encountered in real-world engineering applications and deliver a solid foundation for evaluating the robustness and applicability of our proposed method. The experimental results demonstrate the superiority of MSRN over previous models, achieving significant performance improvements ranging from 2.1 to 3.0. Further analysis confirms the effectiveness of MSRN in recognizing complex long-dependency entities, with a performance improvement of 3.0, as well as out-of-vocabulary locations, with a performance improvement of 6.6 compared to models with similar parameter sizes.}, archive = {J_EAAI}, author = {Guanhua Huang and Bofei Gao and Jiaze Chen and Yuchen Zhang and Zhouwang Yang}, doi = {10.1016/j.engappai.2025.110287}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110287}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing fine-grained geographic named entity recognition by multi-scale siamese reconstruction network}, volume = {148}, year = {2025}, }
• (2025). ASTTN: An adaptive Spatial–Temporal transformer network for traffic flow prediction. EAAI, 148, 110263. (www)

  @article{ , abstract = {Traffic flow prediction as a representative challenge in spatial–temporal modeling tasks, playing a fundamental role in intelligent transportation systems (ITSs). The complexity of spatial–temporal features from road networks and unexpected events complicates accurate predictions. To address the above problems, we propose a novel Adaptive Spatial–Temporal Transformer Network (ASTTN) designed to effectively capture dynamic spatial–temporal features and improve the model’s predictive adaptability under different systems. Considering the comprehensive temporal features, we design a Multi-View Temporal Attention Module to extract long-term and short-term temporal correlation. Additionally, by the limitation of the fixed adjacency matrix, our model proposes an Adaptive Spatial Graph Convolutional Network Module to dynamically capture spatial dependence by the adaptive mechanism based on self-attention. An adaptive Gated Fusion module is employed to achieve the dynamic fusion of temporal and spatial features. Experiments on four real-world traffic datasets show that our model significantly outperforms existing state-of-the-art baselines.}, archive = {J_EAAI}, author = {Zijie Xue and Linyu Huang and Qian Ning}, doi = {10.1016/j.engappai.2025.110263}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110263}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ASTTN: An adaptive Spatial–Temporal transformer network for traffic flow prediction}, volume = {148}, year = {2025}, }
• (2025). Spatio-temporal channel attention and membrane potential modulation for efficient spiking neural network. EAAI, 148, 110131. (www)

  @article{ , abstract = {Spiking Neural Networks (SNNs) are an energy-efficient alternative to Artificial Neural Networks (ANNs) due to their event-driven nature. However, common coding methods such as direct coding struggle to capture critical spatio-temporal dynamics. To address this, we propose a Spatio-Temporal Channel Attention (STCA) module to improve feature extraction during spike encoding. In addition, we introduce a Membrane Potential Modulator (MPM) to reduce information loss due to binary quantization. Together, STCA and MPM form the Gated Attention Coding Mechanism (GACM), which improves SNN training on both static and neuromorphic datasets. Experiments at the Canadian Institute for Advanced Research 10/100 (CIFAR10/100) and CIFAR10-Dynamic Vision Sensor (DVS) show that GACM has higher accuracy and significant efficiency over direct coding. In particular, we improved accuracy by 1.73% on the CIFAR100 and 0.81% on the CIFAR10 in fewer steps.}, archive = {J_EAAI}, author = {Xingming Tang and Tao Chen and Qian Cheng and Hangchi Shen and Shukai Duan and Lidan Wang}, doi = {10.1016/j.engappai.2025.110131}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110131}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatio-temporal channel attention and membrane potential modulation for efficient spiking neural network}, volume = {148}, year = {2025}, }
• (2025). Fault diagnosis method for imbalanced data based on adaptive diffusion models and generative adversarial networks. EAAI, 147, 110410. (www)

  @article{ , abstract = {In engineering practice, the challenge of collecting fault samples often leads to data imbalance, significantly affecting the performance of fault diagnosis. Data augmentation methods offer an effective solution to address this issue by supplementing fault data. However, these methods must meet the requirements of generating high-quality samples, achieving comprehensive mode coverage, and ensuring fast sample generation. Traditional Generative Adversarial Networks are prone to mode collapse, while diffusion models suffer from slow generation speeds, making it difficult for either to fully satisfy these demands. To overcome these challenges, this paper proposes a data augmentation method for fault diagnosis based on an adaptive diffusion model integrated with Generative Adversarial Networks. By introducing a length-adaptive forward diffusion chain to generate Gaussian mixture noise, this method not only ensures smoother and more stable gradients but also avoids computational redundancy and gradient vanishing problems. At each diffusion timestep, the discriminator learns to distinguish real and generated data across different noise ratios and timesteps, enhancing the diversity of fault samples and effectively mitigating mode collapse. Experimental results on two datasets demonstrate that the proposed method outperforms other data augmentation techniques in terms of generation efficiency and stability, effectively addressing the data imbalance problem and significantly improving fault diagnosis performance.}, archive = {J_EAAI}, author = {Xueyi Li and Xudong Wu and Tianyang Wang and Yining Xie and Fulei Chu}, doi = {10.1016/j.engappai.2025.110410}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110410}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis method for imbalanced data based on adaptive diffusion models and generative adversarial networks}, volume = {147}, year = {2025}, }
• (2025). An explainable machine learning method for predicting and designing crashworthiness of multi-cell tubes under oblique load. EAAI, 147, 110396. (www)

  @article{ , abstract = {Multi-cell tubes are widely used in energy-absorbing structures due to their excellent crashworthiness. However, oblique loading in real-world collisions can drastically change their deformation modes, leading to a sharp reduction in energy absorption. This paper proposed an explainable machine learning method combining Artificial Neural Network (ANN) and SHapley Additive exPlanations (SHAP) to quickly predict and design the crashworthiness of multi-cell tubes subjected to oblique loads. Explainable models for six cross-section types were developed, with geometry, material, and loading angle as inputs, and three crashworthiness metrics as outputs. These models demonstrate high accuracy, with the Coefficient of Determination (R 2 ) exceeding 0.91 and the Mean Absolute Percentage Error (MAPE) below 15.76, as well as significant computational efficiency, achieving a speed-up of up to 187,500 times relative to finite element (FE) simulations. Explainability analyses reveal critical nonlinear effects of loading angles, with section dimensions and material properties identified as key factors influencing crashworthiness. Based on the explainable global and local effects of design parameters, a fast optimization strategy for crashworthiness in multi-cell tubes was developed. This strategy was demonstrated using a single-cell tube (1CT) as an application example, with its design and optimization results validated by FE simulations. Compared to existing FE simulations and black-box models, this approach effectively combines fast predictions with explainable results, providing actionable insights for optimizing energy absorption in multi-cell tubes under oblique loads.}, archive = {J_EAAI}, author = {Jian Xie and Junyuan Zhang and Zheng Dou and Mengge Chang and Chang Su}, doi = {10.1016/j.engappai.2025.110396}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110396}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An explainable machine learning method for predicting and designing crashworthiness of multi-cell tubes under oblique load}, volume = {147}, year = {2025}, }
• (2025). An enhanced failure mode and effect analysis method based on preference disaggregation in risk analysis of intelligent wearable medical devices. EAAI, 147, 110384. (www)

  @article{ , abstract = {Conducting a risk analysis on potential failure modes that may damage the performance of intelligent wearable medical devices is imperative since the failure of the devices could directly impact human health. Failure mode and effect analysis (FMEA) is an evaluative instrument for potential failure modes in risk management. This paper presents an enhanced FMEA technique grounded in preference disaggregation analysis considering the interrelationships between failure modes to improve the precision of risk analysis. First, the initial evaluation of failure mode occurrence is updated by the overall influence-strength matrix among failure modes. The matrix formation considers the indirect interrelationships between failure modes, the positive/negative effects of failure modes, and the initial strength of failure modes. Then, a preference disaggregation method is applied to derive the weights of risk factors and the overall utilities of failure modes from historical decision examples. Failure modes are categorized from the most severe to the least severe according to their utilities. Smart bracelets, as a type of intelligent wearable medical devices, apply artificial intelligence technology in health monitoring. Through an illustrative case study of smart bracelets, the efficacy of the proposed approach is validated.}, archive = {J_EAAI}, author = {Huchang Liao and Xiaoyan Yin and Xingli Wu and Romualdas Bausys}, doi = {10.1016/j.engappai.2025.110384}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110384}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An enhanced failure mode and effect analysis method based on preference disaggregation in risk analysis of intelligent wearable medical devices}, volume = {147}, year = {2025}, }
• (2025). Quantitative method for structural health evaluation under multiple performance metrics via multi-physics guided neural network. EAAI, 147, 110383. (www)

  @article{ , abstract = {Quantifying structural health has become a fundamental issue in structural health monitoring. This paper proposes a method for structural health assessment and a corresponding structural health quantification index (health index) that integrates multiple performance parameters. Given the difficulty of capturing the changes in structural performance under excitation, we combine neural networks, hysteresis models, and physical laws to establish a multi-physics-guided neural network that captures the inherent parameters within hysteresis models, and a diversified database is constructed using OpenSees. The model parameters and multiple performance parameters extracted through this network are used to quantitatively analyse structural health. The effectiveness and generalisation of the proposed method are validated through conventional quasi-static tests on components, structures, and materials, including three sets of steel frames. The results demonstrate that the method accurately describes structural health deterioration, offering a novel approach to the quantitative evaluation of structural health.}, archive = {J_EAAI}, author = {Chen Liu and Huiyong Guo and Jin Di and Kaixuan Zheng}, doi = {10.1016/j.engappai.2025.110383}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110383}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Quantitative method for structural health evaluation under multiple performance metrics via multi-physics guided neural network}, volume = {147}, year = {2025}, }
• (2025). Characterizing the hydrodynamic and mechanical properties of hydraulic fractured shale plays using a kolmogorov-arnold-network-assisted data assimilation approach. EAAI, 147, 110380. (www)

  @article{ , abstract = {The effectiveness of hydraulic fracture stimulation is crucial for developing shale hydrocarbon reservoirs. Geophysical-based monitoring tools and numerical simulators require high capital and computational investments, making data-driven models preferred for shale reservoir characteristics. This study integrates the Kolmogorov-Arnold Networks (KAN) with a data assimilation algorithm, Ensemble Kalman Filter (EnKF), and structures a KAN-assisted-EnKF (K-EnKF) protocol. It characterizes the critical rock mechanical and hydrodynamical properties using the real-time pumping pressure as input. The KAN model predicts time-series pump pressure and hydraulic fracture characteristic data for specific stimulation stages. The data assimilation efficacy is significantly enhanced by employing the KAN model as the predictor model. Blind testing applications indicate that KAN outperforms Deep Neural Networks (DNN) by more than an order of magnitude fewer learnable parameters. The numerical experiments show that the EnKF model, when coupled with a KAN containing 1068 learnable parameters, achieves a Mean Squared Error (MSE) of 0.0313 in matching field-measured pump pressure data. The DNN assisted EnKF model (DNN-EnKF) requires 12,869 learnable parameters to achieve a similar pressure matching quality with an MSE of 0.04 without the occurrence of overlearning. Moreover, the KAN model enhances the computational efficiency of EnKF by over twenty times when the numerical simulator is employed as the predictor. The statistical analysis of the inversion results reveals that the proposed K-EnKF model yields hydrodynamic and rock mechanical properties with significantly lower uncertainty compared to the DNN-EnKF model, which consistently generates multiple solutions following a Gaussian distribution.}, archive = {J_EAAI}, author = {Ziqiang Zhou and Baojiang Sun and Qian Sun}, doi = {10.1016/j.engappai.2025.110380}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110380}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Characterizing the hydrodynamic and mechanical properties of hydraulic fractured shale plays using a kolmogorov-arnold-network-assisted data assimilation approach}, volume = {147}, year = {2025}, }
• (2025). Food multi-factor decoupling recognition based on progressive spatial-frequency attention distillation learning. EAAI, 147, 110379. (www)

  @article{ , abstract = {With the widespread application of image recognition technology in daily life, food image recognition faces challenges such as diverse categories and complex forms. Particularly when dealing with subtle differences between similar food items, imbalanced categories, feature ambiguities, and classification confusion caused by the coupling of multiple factors in food representation, existing models still have room for improvement in their recognition accuracy and generalization ability. Therefore, constructing a recognition model that can precisely differentiate food categories while effectively addressing the complexities of coupled factors has become a key issue in this field. In response to these challenges, we propose the innovative Progressive Spatial-Frequency Distillation Network (PSFDNet). By utilizing a unique multidimensional progressive learning strategy combined with an adaptive spatial-frequency attention mechanism, the model significantly enhances its feature extraction and discrimination capabilities within complex food structures. Additionally, we introduce the food correlation evaluation loss to decouple the mutual interference among food features effectively, thereby improving the accuracy and robustness of food image recognition. Extensive experiments verified the outstanding performance of PSFDNet across datasets, demonstrating a notable increase of 0.87% in the Top-1 recognition accuracy and a 50% increase in inference speed. Particularly in recognizing food images characterized by highly coupled features and extremely imbalanced categories, PSFDNet exhibited significant performance advantages over other methods.}, archive = {J_EAAI}, author = {Minkang Chai and Lu Wei and Zheng Qian and Ran Zhang and Ye Zhu and Baoqing Zhou}, doi = {10.1016/j.engappai.2025.110379}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110379}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Food multi-factor decoupling recognition based on progressive spatial-frequency attention distillation learning}, volume = {147}, year = {2025}, }
• (2025). Dynamic health prediction of plain reservoirs based on deep learning algorithms. EAAI, 147, 110378. (www)

  @article{ , abstract = {The healthy operation of reservoirs is crucial for fulfilling their functions and preventing harm to human populations and riverine ecosystems. This study addresses the issues of poor model generalization and low diagnostic accuracy resulting from imbalanced sample distribution in the health prediction of plain reservoirs. We innovatively propose a sample augmentation model (VAE-CGAN) that combines Variational Autoencoder (VAE) and Conditional Generative Adversarial Network (CGAN). Additionally, we present a classification model (CNN-BiLSTM) based on Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory Network (BiLSTM). The VAE-CGAN model learns the data distribution of authentic samples through alternating training of the encoder, generator, and discriminator, thereby achieving augmentation of fault samples and effectively addressing the issue of sample imbalance. The CNN-BiLSTM model utilizes CNN to capture global key features and BiLSTM to capture bi-directional features of time series, efficiently classifying the augmented and balanced data and accurately identifying the health status of the reservoir. In practical applications at two plain reservoirs in China, our method demonstrated superior robustness compared to three other mainstream deep learning models when handling data with varying degrees of imbalance, achieving an accuracy rate of 0.94, which is significantly higher than that of other models. Even under extreme imbalance ratios of 1:20, the accuracy rate improved from 0.80 to 0.93 through sample augmentation. This study not only enhances the dynamic perception and understanding of the operational health of reservoirs but also significantly bolsters the reliability of risk mitigation decisions, offering a novel technical approach for reservoir health management.}, archive = {J_EAAI}, author = {Zhaohui Zhu and Hao Wu and Zhicheng Zhang and Rui Wang and Qiang Yue}, doi = {10.1016/j.engappai.2025.110378}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110378}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic health prediction of plain reservoirs based on deep learning algorithms}, volume = {147}, year = {2025}, }
• (2025). Cross-resolution person re-identification based on double-layer graph convolution network. EAAI, 147, 110376. (www)

  @article{ , abstract = {Cross-resolution person re-identification aims to search for specific person under different camera views and different locations, which has been widely used in the field of intelligent security. To get higher accuracy of person re-identification and reduce the interference of complex background on person re-identification, a cross-resolution person re-identification based on double-layer graph convolution network was proposed in this paper. It mainly includes super-resolution reconstruction (SRR) network and person re-identification network. In the super-resolution network, we introduced a new SRR network combined Transformer (TNSR) with meta-learning up-sampling module, which can ensure the network focuses on the key areas of the image and gets higher resolution SRR images easier for identification. In the person re-identification network, the human key point detection model was introduced, which can effectively reduce the interference of complex background and extract the key point features of the human body. Then we used these key point features as the nodes of the graph convolution and formed the topological map of the human body. Finally, the proposed feature fusion method based on the double-layer graph convolution network was used to gather the information between adjacent nodes and fuse them into a new feature representation. Experimental results on commonly used person re-identification datasets of Concurrent Actions in Video Analysis and Recognition (CAVIAR), Modified Low-Resolution-Market1501(MLR-Market1501) and Modified Low-Resolution-The Chinese University of Hong Kong 03(MLR-CUHK03) showed that, compared with the state of the art person re-identification method, our method improved the Rank-1, Rank-5 and Rank-10 by 2.67%, 2.23% and 0.7% respectively.}, archive = {J_EAAI}, author = {Fuzhen Zhu and Donglin Li and Ce Sun and Bing Zhu}, doi = {10.1016/j.engappai.2025.110376}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110376}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-resolution person re-identification based on double-layer graph convolution network}, volume = {147}, year = {2025}, }
• (2025). Prediction of submarine soil dredging difficulty scale in cutter suction dredger construction with clustering-based deep learning. EAAI, 147, 110370. (www)

  @article{ , abstract = {Expensive marine geological surveys and subjective human assessments limit the accuracy of submarine soil information, making it challenging to predict dredging difficulty for cutter suction dredger (CSD) excavation. To this end, we propose a data-driven framework that comprises data preprocessing, unsupervised clustering, and time series prediction using CSD construction big data. First, a high-dimensional feature selection method is employed to identify key features significantly affecting cutter cutting torque (CCT) and winch swing torque (WST) from 256-dimensional construction data. Then, the K-means algorithm defines a dredging difficulty scale (DDS) for dividing dredging difficulty by clustering CCT and WST. Finally, a deep learning model integrating the convolutional neural network (CNN), long short-term memory (LSTM), and the attention mechanism is formulated. The CNN-LSTM-Attention model aims to predict CCT and WST in the context of multivariate time series and then map such predictions to different DDSs. The applicability of the proposed framework is validated using 14,400 construction data collected from Tian Jing Hao CSD. Results show that four DDSs defined can effectively represent various dredging difficulty levels. The CNN-LSTM-Attention model achieves a high real-time prediction accuracy of 95.83% for DDSs and maintains robust performance across various prediction steps, which outperform baseline models. The proposed framework provides a novel approach for dividing and predicting dredging difficulty without relying on soil information, helping operators to optimize operational instructions in advance.}, archive = {J_EAAI}, author = {Yong Chen and Qiubing Ren and Mingchao Li and Huijing Tian and Liang Qin and Dianchun Wu}, doi = {10.1016/j.engappai.2025.110370}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110370}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction of submarine soil dredging difficulty scale in cutter suction dredger construction with clustering-based deep learning}, volume = {147}, year = {2025}, }
• (2025). Optimized graph neural networks for spatial recognition to support automatic building information model semantic enrichment - Exploring node features enhancing strategies. EAAI, 147, 110365. (www)

  @article{ , abstract = {The integration of Graph Neural Networks (GNNs) with Building Information Modeling (BIM) unveils transformative potential for the semantic enrichment of BIM data, offering significant advancements to the Architecture, Engineering, and Construction (AEC) industry. This paper addresses the critical issue of spatial recognition and classification in BIM through the proposed node-enhanced, self-supervised graph neural network model, Node-Enhanced Graph- Bidirectional Encoder Representation from Transformer (NE-Graph-BERT) , which incorporates edge features. This model represents each space as a node, utilizing spatial relational edges as inputs and is trained on an independently developed knowledge base consisting of 14 space types and 4 relational features across three major categories of architectural space layouts, enabling automatic identification of spatial features. Experimental results indicate that: (1) the enhanced GNNs with integrated edge features achieves precise spatial classification of building functions on extensive building diagrams; (2) NE-Graph-BERT surpasses existing GNNs models in both Precision (97.08%) and F 1 -score (96.75%) ; (3) spatial classification accuracy and interoperability are notably enhanced, reaching up to 97.05% . This study advances the application of GNNs for semantic enrichment in BIM and offers practical insights for their implementation in real AEC projects, paving the way for enhanced efficiency and automation in applications including compliance review, spatial analysis, and building evacuation.}, archive = {J_EAAI}, author = {Yian Chen and Huixian Jiang}, doi = {10.1016/j.engappai.2025.110365}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110365}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimized graph neural networks for spatial recognition to support automatic building information model semantic enrichment - Exploring node features enhancing strategies}, volume = {147}, year = {2025}, }
• (2025). Interpreting injection molding quality defect using explainable artificial intelligence and analysis of variance. EAAI, 147, 110362. (www)

  @article{ , abstract = {Injection molding is a widely employed manufacturing process known for its efficiency, precision, and scalability in producing complex plastic components. However, persistent quality defects, particularly shrinkage, remain a significant challenge, directly impacting product reliability and performance. Traditional studies have relied heavily on Analysis of Variance (ANOVA) to identify and analyze the parameters influencing such defects. While ANOVA is effective in determining the significance of individual factors, it often falls short in capturing the nonlinear interactions and complex dependencies characteristic of injection molding processes. Our study addresses this limitation by adopting a hybrid methodology that integrates ANOVA with explainable artificial intelligence (XAI) techniques, specifically SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations). This innovative approach offers a deeper and more interpretable analysis of the factors affecting shrinkage. We investigate five key parameters: mold temperature, melt temperature, injection time, packing time, and packing pressure. The results reveal that the hybrid approach not only identifies significant parameters but also uncovers complex interactions overlooked by ANOVA alone. This nuanced understanding of process dynamics facilitates improved defect prediction and control, advancing the quality of injection-molded products. By bridging the gap in conventional methodologies, our research blends statistical precision with XAI's interpretability, providing a novel framework for optimizing injection molding processes and enhancing product reliability.}, archive = {J_EAAI}, author = {Faouzi Tayalati and Ikhlass Boukrouh and Abdellah Azmani and Monir Azmani}, doi = {10.1016/j.engappai.2025.110362}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110362}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpreting injection molding quality defect using explainable artificial intelligence and analysis of variance}, volume = {147}, year = {2025}, }
• (2025). Few-shot machine reading comprehension for bridge inspection via domain-specific and task-aware pre-tuning approach. EAAI, 147, 110361. (www)

  @article{ , abstract = {With the wide application of information technologies in the field of bridge engineering, many electronic bridge inspection reports have been generated. However, due to insufficient research on machine reading comprehension (MRC) in this field, a lot of bridge inspection information, e.g., structural basic data, inspected defects, and maintenance suggestions, has not been fully used. Especially, it is time-consuming and labor-intensive to pre-train a domain-specific language model from scratch or annotate large-scale question answering corpora, which also brings challenges to the MRC research in this field. To tackle the problems, this paper proposes a novel few-shot MRC approach for bridge inspection based on the idea of data augmentation. The proposed model uses a pre-trained model as backbone, along with introducing a pre-tuning stage to bridge the gaps between general-purpose pre-training and domain-specific MRC tasks. In order to reduce the workload of manual annotation, we present a novel pre-tuning data generation algorithm which is based on the domain-specific question classification and answer prediction neural models. After pre-tuning and fine-tuning, the proposed model achieves efficient bridge inspection MRC. The experimental results show that the proposed model outperforms the mainstream fine-tuning-based approaches and few-shot MRC baseline models in various settings. With 1024 fine-tuning samples, the F1 value and Exact Match (EM) value are 86.42%, 74.65%, respectively. Our research work can serve as a foundation for the construction of automatic question answering systems for intelligent bridge management and maintenance.}, archive = {J_EAAI}, author = {Ren Li and Luyi Zhang and Qiao Xiao and Jianxi Yang and Yu Chen and Shixin Jiang and Di Wang}, doi = {10.1016/j.engappai.2025.110361}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110361}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Few-shot machine reading comprehension for bridge inspection via domain-specific and task-aware pre-tuning approach}, volume = {147}, year = {2025}, }
• (2025). Hierarchical graph-based machine learning model for optimization of three-dimensional braced steel frame. EAAI, 147, 110356. (www)

  @article{ , abstract = {This paper proposes a method for topology optimization of three-dimensional braced steel frames under static seismic loads, aiming to minimize structural volume while satisfying response constraints. The method consists of a proposed hierarchical graph-based machine learning model that sequentially embeds graph representations of structural elements to create a comprehensive lumped mass model suitable for seismic analysis and optimization of building frames. The proposed model is utilized as a reinforcement learning agent, a class of machine learning, to observe the current building frame configurations, modify the frame to improve its performance, and adjust the model parameters using the obtained reward function. Numerical results demonstrate that the proposed graph-based model can improve its performance better than benchmark of a graph neural network utilized in previous research. When applied to three cases of unseen large three-dimensional building frames, the trained agent with the proposed model outperforms the genetic algorithm and simulated annealing in the optimization task with 11–46% lower objective function while utilizing the same computational cost. The generality, performance, and computational efficiency of the agent indicate its applicability to applied to the optimization of three-dimensional frames.}, archive = {J_EAAI}, author = {Chi-tathon Kupwiwat and Kazuki Hayashi and Makoto Ohsaki}, doi = {10.1016/j.engappai.2025.110356}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110356}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical graph-based machine learning model for optimization of three-dimensional braced steel frame}, volume = {147}, year = {2025}, }
• (2025). Novel q-rung orthopair fuzzy distance based similarity measure and score function in real life decision making. EAAI, 147, 110348. (www)

  @article{ , abstract = {The distance and similarity measure are the most promising information measures to estimate the distance and similarity between two sets. In the present study, novel distance and similarity measure are introduced and affirmed their proficiencies by theoretical explications and numerical examples. To demonstrate the applicability of the proposed similarity measure ( pSMr ), medical diagnosis and pattern recognition problems have been solved. This study aims to develop a novel ranking method under q-Rung Orthopair Fuzzy Environment (q-ROFE) based on the proposed measure to solve multi-criteria group decision making (MCGDM) problem. This method aims to identify the most promising alternative based on similarity to positive ideal solution (PIS). The focus is to comprehend the closeness of performance of an alternative from desired performance relative to its distance from ideal solutions. Furthermore, an improved q-Rung orthopair fuzzy score function ( SF ) has also been proposed elaborating all the significant properties. Moreover, Full consistency method (FUCOM) has been reclaimed under q-ROFE with the aid of q-Rung orthopair fuzzy Weighted Averaging Operator (q-ROFWA). To validate the acceptability of the proposed ranking method, a problem of healthcare waste management (HCWM) has been structured and solved. The feasibility and effectiveness of integrated model have been encountered with comparative and sensitivity examinations. The proposed measures outperform over existing measures to discriminate two q-ROF numbers, which has been validated by comparison. Moreover, the novel S F is competent enough to rank any type of q-ROF numbers regardless of their membership and non-membership grades and the dominance has been illustrated through comparisons.}, archive = {J_EAAI}, author = {Raili Basu and Sayanta Chakraborty and Apu Kumar Saha}, doi = {10.1016/j.engappai.2025.110348}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110348}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Novel q-rung orthopair fuzzy distance based similarity measure and score function in real life decision making}, volume = {147}, year = {2025}, }
• (2025). Self-optimized learning algorithm for multi-specialty multi-stage elective surgery scheduling. EAAI, 147, 110346. (www)

  @article{ , abstract = {As populations age and living standards improve, the demand for elective surgery has risen significantly. Yet, economic and policy constraints slow the expansion of medical resources, requiring hospitals to manage multi-specialty resources collectively. This makes modelling and scheduling elective surgeries complex. This study proposes an integrated scheduling model for multi-specialty and multi-stage elective surgeries. We use novel evaluation indicators to measure medical workload, waiting times and service continuity from hospital and patient perspectives, ensuring dimension consistency in the objective function. To address scheduling complexity, we design a self-optimized learning algorithm combining reinforcement learning (RL), genetic algorithm (GA), and heuristic rules, that is, a dual-encoding hybrid genetic algorithm based on Q-learning and scheduling rules (DQGA). Our results show that the hybrid algorithm achieves a 64.99% improvement in solution accuracy compared to GA and 66.07% to Q-learning-based genetic algorithm (QGA). Experimental findings highlight that our scheduling model not only enhances resource utilization efficiency but also provides practical, scalable solutions for real-world hospital scheduling problems, where traditional methods often struggle with larger problems.}, archive = {J_EAAI}, author = {Yufan Liu and Youhao Huang and Zongli Dai and Yueming Gao}, doi = {10.1016/j.engappai.2025.110346}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110346}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-optimized learning algorithm for multi-specialty multi-stage elective surgery scheduling}, volume = {147}, year = {2025}, }
• (2025). The applications of machine learning mechanisms in the compositions of internet of things services: A systematic study, current progress, and future research agenda. EAAI, 147, 110345. (www)

  @article{ , abstract = {The Internet of Things (IoT) is a modern domain involving billions of intelligent objects connecting the physical with the virtual world. Typically, these things are IoT devices whose functionality they provide through services. The techniques of service composition represent ways to realize user requests in IoT applications with diverse services. Resource-constrained IoT nodes inherently require solutions for their technologically limited resources; however, many IoT devices tap into cloud infrastructures to harness unlimited resources, among other advantages. This paper stresses the importance of service composition in the IoT landscape, keeping in mind the heterogeneous and dynamic nature of the devices. Research on service composition in IoT and Machine Learning (ML) algorithms is surprisingly scant, with just a few articles treading the thin line between the two subjects, leaving significant knowledge gaps. This article provides a comprehensive framework that links ML algorithms to service composition in IoT, analyzes their applications, challenges, solutions, and benefits, and identifies future research. This will serve to fill the gaps by providing adequate categorization, comprehensive analysis, and pointing out suggestions that need to be addressed further in the field. This review sums up the integration of ML techniques for IoT ecosystems into service composition. Essential ML techniques such as supervised learning, reinforcement learning, ensemble learning, and deep learning are investigated in this paper, and they offer different strengths and challenges for the optimality of the service composition. These ML-based approaches form the backbone of interoperability challenges and dynamic adaptation of service compositions to runtime-evolving contexts in IoT environments. Specifically, the integration of successful ML with consideration to continuous learning, robustness against uncertainties, adaptability, privacy and security, scalability, efficiency, user-centric design, and integration with decision-making frameworks are also discussed. The results show that by navigating these considerations, ML effectively empowers IoT systems for efficient, adaptive, and intelligent service orchestration, marking a significant improvement in the dynamic landscape of the IoT.}, archive = {J_EAAI}, author = {Jiazhong Lu and Weisha Zhang and Marzieh Hamzei and Nima Jafari}, doi = {10.1016/j.engappai.2025.110345}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110345}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The applications of machine learning mechanisms in the compositions of internet of things services: A systematic study, current progress, and future research agenda}, volume = {147}, year = {2025}, }
• (2025). Fault diagnosis method of mining vibrating screen mesh based on an improved algorithm. EAAI, 147, 110343. (www)

  @article{ , abstract = {Artificial intelligence fault diagnosis technology based on machine vision, due to its low cost and high efficiency, has become an indispensable part of production processes across various industries. Compared to traditional fault diagnosis methods, artificial intelligence diagnosis of common mechanical failures, such as ‘clogging’, ‘wear’, and ‘breakage’ in vibrating screen meshes within the mining screening sector, improves detection efficiency, accuracy, and sustainability. Since small target faults in large screening areas are challenging to detect through manual diagnosis, it reduces screening efficiency and shorter equipment lifespan, negatively impacting mining enterprises' safe and efficient production. A fault diagnosis model with a better speed-precision trade-off is proposed to improve detection precision based on the You Only Look Once version 5 single-stage object detection algorithm. This model is optimized in feature extraction and fusion by integrating autocode masking, re-parameterization, and omni-dimensional attention. The model's performance is primarily evaluated using precision, recall, balanced score, and mean average precision. The improved algorithm achieves a precision of 97.2%, a recall of 93.3%, a balanced score of 95.21%, and a mean average precision of 97.0%. Experimental results demonstrate that the improved algorithm increases the mean average precision by 3.1% compared to the original model. The results show that the improved algorithm is more effective than the original in fault diagnosis, with enhanced screen mesh detection precision. Thus, it ensures production safety and stable screening efficiency. Moreover, the proposed algorithm provides a reference for advancing intelligent and efficient fault diagnosis technology in the mining screening field.}, archive = {J_EAAI}, author = {Fusheng Niu and Jiahui Wu and Jinxia Zhang and ZhiHeng Nie and Guang Song and Xiongsheng Zhu and Shuo Wang}, doi = {10.1016/j.engappai.2025.110343}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110343}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis method of mining vibrating screen mesh based on an improved algorithm}, volume = {147}, year = {2025}, }
• (2025). A novel enhanced superlet synchroextracting transform ensemble learning for structural health monitoring using nonlinear wave modulation. EAAI, 147, 110341. (www)

  @article{ , abstract = {This study investigates the application of nonlinear wave modulation (NWM) using chirp signals for structural health monitoring (SHM). The implementation of NWM with monoharmonic signals (periodic signals that consist of a single frequency component) poses significant challenges due to the complexity of selecting optimal pump and carrier frequencies, leading to time-intensive processes. In contrast, analyzing NWM with chirp signals introduces additional complexities regarding signal processing compared to monoharmonic excitations. Time-frequency analysis (TFA) has been identified as a crucial method for examining non-stationary signals; however, many existing techniques face limitations in resolution, particularly in the context of chirp signals, as dictated by the Heisenberg uncertainty principle. To address these challenges, the superlet synchroextracting transform (SLSET) is introduced as an innovative TFA approach that combines the strengths of superlet (SL) and synchroextracting transforms, resulting in improved resolution. This research utilizes NWM alongside SLSET to detect boundary loosening in sandwich beams, demonstrating the method's effectiveness in identifying structural damage while maintaining robustness against noise. Results indicate that SLSET significantly enhances the damage index compared to traditional TFA methods. The high resolution achieved allows for the detection of sidebands in vibro-acoustic modulation (VAM) tests conducted at low pump frequencies. Furthermore, three machine learning (ML) models including support vector machine (SVM), Adaptive Boosting (AdaBoost), and Random Forest (RF) were trained. The stack ensemble method combined the outputs of these models, resulting in an overall accuracy of 99.2%. This approach effectively leveraged the strengths of individual models, enhancing generalization and robustness in detecting damage across complex data scenarios. The features extracted using SLSET for VAM data of faulty structure attains a classification accuracy of 98.9%. In contrast, features derived from conventional time-frequency methods fail to identify damage, even in noise-free conditions.}, archive = {J_EAAI}, author = {Naserodin Sepehry and Mohammad Ehsani and Hamdireza Amindavar and Weidong Zhu and Firooz Bakhtiari Nejad}, doi = {10.1016/j.engappai.2025.110341}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110341}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel enhanced superlet synchroextracting transform ensemble learning for structural health monitoring using nonlinear wave modulation}, volume = {147}, year = {2025}, }
• (2025). Not all samples are equal: Boosting action segmentation via selective incremental learning. EAAI, 147, 110334. (www)

  @article{ , abstract = {Temporal action segmentation (TAS) seeks to perform classification for each frame in a video. Existing methods tend to design diverse network architectures, while overlooking the intrinsic characteristics of training samples. Notably, two key issues arise: (1) Frames around action boundaries are more ambiguous and thus pose greater difficulties for training compared to other frames; and (2) beyond the commonly used categorical labels, the total number of action instances within a video may serve as an additional, potentially vital, supervision cue. To address these issues, this paper introduces a novel method that combines a model-agnostic training strategy with an instance number alignment loss, designed to enhance the performance of existing models. Specifically, a selective incremental learning (SIL) strategy is proposed to alleviate the impact of noisy samples by progressively training the model in an easy-to-difficult manner through a dynamic sample selection mechanism. Furthermore, an instance number alignment loss (INAL) is developed to capture both global and local features simultaneously by incorporating a multi-task learning module. Extensive evaluations are conducted on three benchmark datasets, namely 50Salads, Georgia Tech egocentric activities (GTEA), and Breakfast. The experimental results demonstrate that the proposed method achieves substantial performance improvements over state-of-the-art approaches.}, archive = {J_EAAI}, author = {Feng Huang and Xiao-Diao Chen and Wen Wu and Weiyin Ma}, doi = {10.1016/j.engappai.2025.110334}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110334}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Not all samples are equal: Boosting action segmentation via selective incremental learning}, volume = {147}, year = {2025}, }
• (2025). Application of artificial intelligence models for assessing the performance of closed vertical refrigerated display cabinet - A comparative study of different operating scenarios. EAAI, 147, 110332. (www)

  @article{ , abstract = {Display cabinets are widely utilized in retail stores/supermarkets to keep the food at its desired temperature with the least fluctuation possible. However, they account for a large proportion of supermarket consumption. Better energy management of display cabinets would reduce the load on the electric grid, paving a path toward sustainability. This research presents a multidisciplinary work by integrating a thermal system, i.e., a refrigeration equipment, with artificial intelligence models. An experimental investigation has been conducted to assess the dynamic behavior of a closed vertical refrigerated display cabinet (CVRDC) under three operating scenarios: hotter, colder, and reference setting temperature with door-closed and door-opening cases. The test product made of methylcellulose was loaded inside the cabinet during experimentation. The impact of varying air return and outlet temperatures on the display cabinet's product temperature and demand power has been investigated. Two artificial intelligence models, adaptive neuro-fuzzy inference system (ANFIS), and artificial neural network (ANN), have been developed and practically applied to a display cabinet for predicting the product temperature and power in the door-closed and door-opening cases for the three scenarios. The work aims to support sustainable development in supermarket refrigeration systems by applying artificial intelligence models to optimize the functioning of the display cabinet and find the optimal setting temperature. The hyperparameters optimizing the prediction capability of each model are identified, and their performance is validated against experimental data. The findings demonstrated that both models effectively predicted the product temperature and power variation of CVRDC under different operating scenarios, with ANN achieving higher accuracy due to its superior ability to capture transient behaviors. Furthermore, the analysis identified the operating scenario that minimizes energy consumption while maintaining the desired product temperature, providing actionable insights for energy-efficient operation. These findings can guide supermarkets in optimizing temperature settings for energy efficiency in real-world conditions. Also, results highlight the potential of artificial intelligence to optimize supermarket refrigeration systems, reducing energy consumption and supporting sustainability initiatives.}, archive = {J_EAAI}, author = {Gaurav Priyadarshi and Hong-Minh Hoang and Romuald Hunlede and Yvanne Paviet-Salomon and Anthony Delahaye and Bukke Kiran Naik}, doi = {10.1016/j.engappai.2025.110332}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110332}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Application of artificial intelligence models for assessing the performance of closed vertical refrigerated display cabinet - A comparative study of different operating scenarios}, volume = {147}, year = {2025}, }
• (2025). Geometry-sensitive semantic modeling in visual and visual-language domains for image captioning. EAAI, 147, 110330. (www)

  @article{ , abstract = {Transformer-based models with grid features as visual representations perform well in image captioning. However, the division and flattening operations increase the difficulty of capturing objects and their relationships via pure semantic modeling. Furthermore, the natural language generated by the current Transformer model still suffers from semantic overconcentration. In this paper, we aim to improve the attention modules in two ways to solve the above issues. We first propose a Geometry-Sensitive Self-Attention (GSSA) module, subdivide geometric signals in the visual domain into relative position and distance, and assist the semantic modeling process according to their unique characteristics. It compensates for the lack of objects and their relationships in the grid features. Then, we propose a Geometry-Sensitive Cross-Attention (GSCA) module, which perceives the source neighboring relationships between images and text in the visual-language domain from a geometric perspective and uses these relationships to adjust the semantic correspondences between the two dynamically. It spreads overly focused attention to surrounding grids to improve understanding of full image content during captioning. To prove our designs, we apply GSSA and GSCA to a standard Transformer to construct a novel Geometry-Sensitive Transformer Network (GSTNet), which conducts geometry-sensitive semantic modeling in visual and visual-language domains. Extensive experiments are conducted to verify the effectiveness of our proposal. The results show that our GSTNet achieves superior performance compared to many state-of-the-art image captioning models on the Microsoft Common Objects in Context (MSCOCO) dataset. Besides, the generalization of GSTNet is also verified on the Flickr30k dataset.}, archive = {J_EAAI}, author = {Wencai Zhu and Zetao Jiang and Yuting He}, doi = {10.1016/j.engappai.2025.110330}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110330}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Geometry-sensitive semantic modeling in visual and visual-language domains for image captioning}, volume = {147}, year = {2025}, }
• (2025). Dual-clustering-based two-population co-evolutionary algorithm for segmentation coding in flash memory. EAAI, 147, 110329. (www)

  @article{ , abstract = {To ensure the reliability of stored data, error checking and correction (ECC) module is widely used in flash memory-based storage devices. With the increase of raw bit error rate (RBER), the traditional error correction coding mode is not only hard to satisfy the requirements of storage devices, but also reduces the efficiency of data recovery when error data cannot be located. In this paper, a novel segmentation coding mode is proposed to improve the drawbacks of traditional coding mode. To our knowledge, this is an innovative study of the error correction coding mode in ECC. It can make the coding mode used in ECC to become more flexible and efficient. Since the performance of this coding mode is related to segmentation, we propose a dual-population co-evolutionary algorithm based on clustering algorithm to optimize the performance. The proposed algorithm adopts clustering algorithm to measure the diversity of the population and dynamic weight allocation strategy to regulate evolution indicators of the population. Some experiments are conducted on benchmark problems and the segmentation coding problem, respectively. Experimental results show that the proposed algorithm is superior to other state-of-the-art evolutionary algorithms.}, archive = {J_EAAI}, author = {Jianjun Luo and Menghao Chen and Boming Huang and Hailuan Liu and Lingyan Fan and Lijuan Gao and Guorui Feng}, doi = {10.1016/j.engappai.2025.110329}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110329}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-clustering-based two-population co-evolutionary algorithm for segmentation coding in flash memory}, volume = {147}, year = {2025}, }
• (2025). An integrated method of hotel site selection based on probabilistic linguistic multi-attribute group decision making. EAAI, 147, 110328. (www)

  @article{ , abstract = {The site selection acts a pivotal role in determining the success of a construction project. Since the site selection needs to gather the wisdom of a group of decision makers (DMs) and involves many factors, it can be regarded as a multi-attribute group decision making (MAGDM) problem in artificial intelligence. The assessments of alternatives on attributes are expressed by probabilistic linguistic term sets (PLTSs). A new two-stage normalization method is proposed for PLTSs considering the psychological states of decision makers. A new score function for PLTS is defined. The best and worst method is extended for fuzzy preference relation. The individual objective attribute weights are determined via information entropy. The individual subjective attribute weights are derived through the extended best and worst method. The individual comprehensive attribute weights are derived by minimum relative entropy principle. The weights of DMs are acquired through an optimization model. It minimizes the deviation between the opinions of all DMs and the deviation between the individual and collective comprehensive attribute weight vectors, simultaneously, which effectively overcomes the drawback of only minimizing single deviation. A new method is presented for MAGDM with PLTSs. A hotel site selection example is demonstrated and comparative analyses are executed to verify the validity and advantages of the proposed method. The test statistic Z values of Spearman's rank-correlation test are all smaller than 1.645, which shows that the ranking order obtained by the proposed method is statistically sharply distinct from that produced by other methods and thus further validates the proposed method.}, archive = {J_EAAI}, author = {Jiu-Ying Dong and Ying-Ying Yao and Shyi-Ming Chen and Shu-Ping Wan}, doi = {10.1016/j.engappai.2025.110328}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110328}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated method of hotel site selection based on probabilistic linguistic multi-attribute group decision making}, volume = {147}, year = {2025}, }
• (2025). Automated acquisition and analysis of illegal fund accounts in gambling websites. EAAI, 147, 110326. (www)

  @article{ , abstract = {While the rapid growth of the internet has made life more convenient, it has also made it easier for people to access online gambling sites. This has led to an increase in fraud, resulting in significant financial losses and serious societal issues. However, the abundance of gambling sites, diverse development frameworks, and complex identity verification methods pose significant challenges to gaining comprehensive access to their illegal funds. To address online gambling fraud and assist law enforcement in cutting off their funding chains, we investigate gambling websites to uncover information about the physical funding accounts used by criminal groups. Given the extensive scale of these websites, we propose an auto-registration framework based on You Only Look Once version 4 (YOLOv4) to automate account registration and retrieve illegal fund account details. Additionally, we construct a dataset of user interface (UI) elements from 17 types of gambling websites for model training. The YOLOv4 model achieves an F1-score of 0.84 and a mean Average Precision (mAP) of 83.96%. The proposed framework achieves a registration success rate of 60.58%. After extracting data from numerous gambling websites in seven batches, we identify 7496 entity account details and 47 payment methods, analyze the wealth of entity information, and highlight six new payment methods. Finally, by identifying recurring illegal fund accounts across multiple domains, we confirm 23 criminal gangs, providing substantial support to law enforcement agencies in combating online gambling-related crimes.}, archive = {J_EAAI}, author = {Shenao Zheng and Yanan Cheng and Guoying Sun and Zhaoxin Zhang}, doi = {10.1016/j.engappai.2025.110326}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110326}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated acquisition and analysis of illegal fund accounts in gambling websites}, volume = {147}, year = {2025}, }
• (2025). Joint prediction of multi-aircraft trajectories in terminal airspace: A flight pattern-guided social long-short term memory network. EAAI, 147, 110325. (www)

  @article{ , abstract = {Aircraft trajectory prediction is a critical foundation for tasks such as conflict detection and resolution, and monitoring of abnormal aircraft behavior, thus making it a key technology for the next generation of air traffic systems. Airport terminal areas, serving as the convergence points of the entire air transportation network, accommodate the highest density of aircraft within the aviation system. Among these densely packed aircraft, there are inherent and potential interactions that are invisible yet influential, posing a significant challenge for trajectory prediction in the terminal area. To address this issue, we propose a Flight Pattern-Guided Social Long-Short Term Memory (FPG-SLSTM) network to jointly predict multiple aircraft trajectories. Firstly, it introduces a flight pattern matching method based on classification concepts to assign a flight pattern to each trajectory. Following this, the flight intent trajectory is generated based on that flight pattern. Then, the loss function is improved by incorporating the flight intent trajectory as prior physical information, and the Social Long-Short Term Memory (Social LSTM) neural network is employed to model the trajectory prediction problem for multiple aircraft, with social pooling operations to integrate the mutual influences among aircraft. A real-world dataset was constructed to validate the proposed approach. Experimental results show that the method achieved a 14.5% improvement in horizontal prediction accuracy and a 29.5% improvement in height prediction accuracy on average for a 6-minute horizon, compared to Binary Encoding Representation for Flight Trajectory Prediction (FlightBERT). These findings highlight that the proposed framework outperforms other baseline models in terms of prediction accuracy, particularly in complex traffic environments during busy periods, demonstrating the model’s ability to further enhance both the precision and robustness of trajectory prediction.}, archive = {J_EAAI}, author = {Xiao Chu and Weili Zeng and Lingxiao Wu}, doi = {10.1016/j.engappai.2025.110325}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110325}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Joint prediction of multi-aircraft trajectories in terminal airspace: A flight pattern-guided social long-short term memory network}, volume = {147}, year = {2025}, }
• (2025). Evolution of building energy management systems for greater sustainability through explainable artificial intelligence models. EAAI, 147, 110324. (www)

  @article{ , abstract = {Predicting energy consumption is a task that allows energy supply companies to adapt to certain behaviours. The activities that companies can undertake include learning about the behaviour of their customers in order to adapt their tariffs to consumption or identifying the intervals in which there will be a higher demand for energy and to plan for the adaptation of supply chains. While predicting energy consumption is no longer a major challenge, and models with high accuracy rates have been developed, an clear understanding of energy consumption among users is still obscure. If the problem of explainability is resolved, companies will be able to better adapt their services by generating the exact amount of energy to be sold, which will also reduce its cost for customers. There is no single explanatory approach to learning models that works best. There are multiple paths to achieving explainability: data vs. model, directly interpretable vs. post hoc explanation, local vs. global, etc. This article reviews which explainable artificial intelligence algorithms are the most appropriate for a given use case, as multiple forms of explanation can lead to confusion in figuring out which algorithms are the most appropriate for a given use case. In our case study, a specific dataset, extracted from a two-year period in a shoe store, is used to review some of the main explainable artificial intelligence algorithms on machine learning models, capable of predicting energy consumption and subsequently providing explainability to the process.}, archive = {J_EAAI}, author = {Alfonso González-Briones and Javier Palomino-Sánchez and Zita Vale and Carlos Ramos and Juan M. Corchado}, doi = {10.1016/j.engappai.2025.110324}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110324}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evolution of building energy management systems for greater sustainability through explainable artificial intelligence models}, volume = {147}, year = {2025}, }
• (2025). Optimizing artificial intelligence-aided breast cancer models: An empirical analysis of binary classifiers and regression-based feature selectors. EAAI, 147, 110318. (www)

  @article{ , abstract = {Artificial Intelligence (AI)-aided solutions for the early breast cancer (BCa) detection have a high potential to enhance survival rates and improve the quality of life for millions of women. Despite numerous research in the domain, the persistent challenges (over- & under-fitting) within existing binary classifiers continues to impede the accuracy and efficiency of current AI solutions. One key contributing factor to these issues is the practice of training AI models with full or irrelevant feature sets of datasets. Hence, determining important (relevant) feature subsets, and utilizing the subsets in training phase, are critical. In this study, we conduct a comprehensive empirical analysis involving five binary classification and four regression-based feature selection techniques using the BCa Wisconsin and Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSIM) datasets as our testbeds. Our objective was to identify the most accurate (efficient) combinations of five classifiers and four regression-based feature selectors to overcome under- & over-fitting issues. In the first dataset, promising results were achieved with accuracy exceeding 90%. However, only an average of 2% improvement was observed in addressing these issues. In contrast, while slightly lower classification results (less than 60%) were achieved with the second dataset, there was a more significant improvement of over 10% in overcoming both under- & over-fitting cases.}, archive = {J_EAAI}, author = {Fakhriddin Madolimov and Asilbek Medatov and Elmira Nazirova and Hakimjon Zaynidinov and Uktam Azimov and Shakhnoza Turakhonova and Jahongir Azimjonov}, doi = {10.1016/j.engappai.2025.110318}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110318}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing artificial intelligence-aided breast cancer models: An empirical analysis of binary classifiers and regression-based feature selectors}, volume = {147}, year = {2025}, }
• (2025). Speech emotion recognition based on spiking neural network and convolutional neural network. EAAI, 147, 110314. (www)

  @article{ , abstract = {There is an urgent need to determine emotions automatically through speech signals to promote the progress of intelligent technology. However, the low accuracy problem isn't solved so far as, this hinders potential applications of Speech Emotion Recognition (SER). One of the most critical reasons for this low accuracy is that subjective emotions are random and generate weak pulse signals; moreover, they are often hidden in audio, video, and text feature which are extracted from speech. Hence, the features may not be discriminative enough to depict subjective emotions. Therefore, a dual-path SER framework is designed in this paper. Added to the traditional Convolutional Neural Network (CNN)-based SER scheme to handle speech emotion features, the Spiking Neural Network (SNN) framework is added to identify the dynamic pulse emotion features and improve the accuracy of SER. At the same time, a Perceptual Neuron Encoding Layer (PNEL) is proposed to enhance the ability to process speech signals. Overall, the experimental results on the interactive emotional dyadic motion capture database (IEMOCAP) databases show that the proposed approach can achieve 65.3% accuracy and excellent performance in solving the SER issues compared to other existing approaches.}, archive = {J_EAAI}, author = {Chengyan Du and Fu Liu and Bing Kang and Tao Hou}, doi = {10.1016/j.engappai.2025.110314}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110314}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Speech emotion recognition based on spiking neural network and convolutional neural network}, volume = {147}, year = {2025}, }
• (2025). A global information-guided denoising diffusion probabilistic model for fault diagnosis with imbalanced data. EAAI, 147, 110312. (www)

  @article{ , abstract = {Data imbalance is a significant challenge in intelligent fault diagnosis. Diffusion models offer potential solutions to data imbalance by generating samples from limited data. However, their limited global inference capability can create discrepancies between the generated and real data. To address this issue, this paper introduces a global information-guided denoising diffusion probabilistic model (GI-DDPM) for intelligent fault diagnosis. GI-DDPM employs a pretrained Vision Transformer (ViT) as a guide for global information. It extracts the global information from the data through ViT and aligns it with features extracted by the diffusion model. The cosine similarity between the two sets of features is calculated and used as a regularization term to guide the diffusion model to pay more attention to the comprehensive features in the data. This approach encourages the model to generate fault data with more comprehensive representations. The proposed method is tested on two datasets with different balance ratios for imbalanced fault diagnosis experiments, achieving average accuracy rates of 96.95%, 97.18% and 96.06%. These results outperform other intelligent methods, demonstrating significant potential in handling imbalanced fault diagnosis.}, archive = {J_EAAI}, author = {Han Yang and Yan Song and Daichao Wang and Jinghao Xing and Yibin Li}, doi = {10.1016/j.engappai.2025.110312}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110312}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A global information-guided denoising diffusion probabilistic model for fault diagnosis with imbalanced data}, volume = {147}, year = {2025}, }
• (2025). Rule training by score-based supervised contrastive learning for sketch explanation. EAAI, 147, 110310. (www)

  @article{ , abstract = {This paper presents a novel approach to explain scoring results of infant visual-motor integration sketches utilized in developmental tests by training predefined rules for each test item. To address the performance issues caused by limited data, we employ a pre-trained model that uses supervised contrastive learning based on item scores. To ensure effective training, a memory bank structure is proposed to accumulate diverse embeddings over multiple iterations and prevent the encoder that processes item information from being trained to prevent collapsing in the Siamese network. Experiments demonstrate that the proposed method improves performance in both score and rule inferences, achieving an accuracy of approximately 75.95% in rule inference. In addition, an ablation study validates the effectiveness of the proposed approach in enhancing performance, confirming its potential as a reliable tool for early developmental screenings and clinical assessments. As such, the proposed approach could enhance clinical decision-making by providing essential interpretability for developmental tests.}, archive = {J_EAAI}, author = {Tae-Gyun Lee and Jang-Hee Yoo}, doi = {10.1016/j.engappai.2025.110310}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110310}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rule training by score-based supervised contrastive learning for sketch explanation}, volume = {147}, year = {2025}, }
• (2025). Regression loss-assisted conditional style generative adversarial network for virtual sample generation with small data in soft sensing. EAAI, 147, 110306. (www)

  @article{ , abstract = {Existing methods that extend virtual sample pools to address small sample problem caused by sample atypicality and uneven distribution often overlook data sparsity and inverse sample generation challenges, which limits the accuracy of subsequent modeling. To address above problem, we propose a novel regression-assisted conditional style generative adversarial network (RAC-StyleGAN). The proposed method leverages the strengths of StyleGAN in latent space mapping to enhance data diversity and granularity, while incorporating regression-assisted conditions to improve modeling performance. Specifically, RAC-StyleGAN utilizes kernel density estimation and radial basis function interpolation to ensure that the generated output variables are uniformly distributed. Based on the principle of inverse transformation, the interpolated output variables are then used as conditional inputs for the StyleGAN model, generating virtual input variables that faithfully reflect the marginal distribution of the original data. Furthermore, to preserve the complex nonlinear relationships between input and output variables, RAC-StyleGAN integrates a regression loss strategy based on empirical risk minimization into the StyleGAN framework. By fine-tuning the generation process, the soft-sensing model effectively captures the nonlinear mapping between inputs and outputs. Experimental validations on synthetic nonlinear functions, University of California Irvine machine learning (UCI) datasets, and a real-world purified terephthalic acid (PTA) solvent system demonstrate that RAC-StyleGAN effectively generates high-quality virtual samples, significantly enhancing the modeling performance.}, archive = {J_EAAI}, author = {Xue-Yu Zhang and Qun-Xiong Zhu and Wei Ke and Yan-Lin He and Ming-Qing Zhang and Yuan Xu}, doi = {10.1016/j.engappai.2025.110306}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110306}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Regression loss-assisted conditional style generative adversarial network for virtual sample generation with small data in soft sensing}, volume = {147}, year = {2025}, }
• (2025). Graph neural networks for multivariate time-series forecasting via learning hierarchical spatiotemporal dependencies. EAAI, 147, 110304. (www)

  @article{ , abstract = {Multivariate time-series forecasting is one of the essential tasks to draw insights from sequential data. Spatiotemporal Graph Neural Networks (STGNN) have attracted much attention in this field due to their capability to capture the underlying spatiotemporal dependencies. However, current STGNN solutions succumb to a higher degree of error in their predictions due to insufficient modelling of the dependencies and dynamics at different levels. In this paper, a Graph Neural Networks-based model is proposed for multivariate time-series forecasting via learning hierarchical spatiotemporal dependencies (HSDGNN). Specifically, variables are organised as nodes in a graph while each node serves as a subgraph consisting of the attributes of variables. Then two-level convolutions are designed on the hierarchical graph to model the spatial dependencies with different granularities. The changes in graph topologies are also encoded for strengthening dependency modelling across time and spatial dimensions. The proposed model is tested using real-world datasets from different domains, including transportation, electricity, and meteorology. The experimental results demonstrate that HSDGNN can outperform state-of-the-art baselines by up to 15.3% in terms of prediction accuracy, without compromising model scalability.}, archive = {J_EAAI}, author = {Zhou Zhou and Ronisha Basker and Dit-Yan Yeung}, doi = {10.1016/j.engappai.2025.110304}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110304}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph neural networks for multivariate time-series forecasting via learning hierarchical spatiotemporal dependencies}, volume = {147}, year = {2025}, }
• (2025). Post-earthquake structural damage detection with tunable semi-synthetic image generation. EAAI, 147, 110302. (www)

  @article{ , abstract = {In the aftermath of an earthquake, conducting rapid structural safety assessments is essential. A Deep Learning-based damage detector capable of automatically analyzing videos from Unmanned Aircraft Systems (UAS) surveys would be highly beneficial for this purpose. Despite significant advancements in object detection using Deep Convolutional Neural Networks (DCNNs), developing an effective post-earthquake damage detector remains challenging due to the scarcity of large, annotated image datasets. In this work, we present a system to create a large number of images where artificial damage instances are applied to real-world three-dimensional (3D) models of buildings and bridges. We defined such images as semi-synthetic. The proposed method relies on the definition, made by human experts, of meta-annotations from which a variety of damage instances can be generated in a controlled way. Semi-synthetic images are designed to augment real-world datasets, enhancing the training process of a DCNN-based damage detector. This semi-synthetic image augmentation can be iteratively refined to target the most critical cases. Experiments conducted on the ‘Image Database for Earthquake damage Annotation’ (IDEA) dataset shown that a detector trained on a combination of real and semi-synthetic images performs better than one trained on real images alone. A damage detector trained using the proposed strategy was then incorporated into a system that analyzes and tracks multiple damage instances in UAS-acquired videos, generating concise summaries of the findings. The effectiveness of the system was validated by the analysis of post-earthquake UAS videos and the production of reports that were reviewed by structural engineering experts.}, archive = {J_EAAI}, author = {Piercarlo Dondi and Alessio Gullotti and Michele Inchingolo and Ilaria Senaldi and Chiara Casarotti and Luca Lombardi and Marco Piastra}, doi = {10.1016/j.engappai.2025.110302}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110302}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Post-earthquake structural damage detection with tunable semi-synthetic image generation}, volume = {147}, year = {2025}, }
• (2025). Predicting potential microbe-disease associations based on heterogeneous graph attention network and deep sparse autoencoder. EAAI, 147, 110301. (www)

  @article{ , abstract = {Identifying potential associations between microbes and diseases is crucial for explaining disease pathogenesis and designing targeted therapeutic strategies. Basic biological experiments for microbe-disease association (MDA) prediction are costly, time-consuming, and labor-intensive, whereas computational methods can effectively complement traditional biological experiments. We propose a computational framework called graph attention convolutional deep sparse autoencoder microbe-disease association (GCDSAEMDA) to predict unknown MDAs. First, we calculate the semantic similarity and Gaussian interaction profile (GIP) similarity of diseases, as well as the functional similarity and GIP similarity of microbes, and integrate these similarity matrices to construct a heterogeneous graph. Next, a multi-head dynamic graph attention mechanism is employed to extract low-order features of microbe and disease nodes in the heterogeneous graph, while multiple convolutional neural networks with different kernels aggregate and concatenate these low-order features to form new high-order representations. Third, we apply a cosine distance-based k-means clustering to select reliable negative samples and use a deep sparse autoencoder to extract high-order features of microbe-disease pairs. Finally, an ensemble Light Gradient Boosting Machine (LightGBM) algorithm is used to predict potential MDAs. GCDSAEMDA was compared to four state-of-the-art MDA models on the Human Microbe-Disease Association Database (HMDAD) and Disbiome databases and validated through five-fold cross-validation on diseases, microbes, and microbe-disease pairs. Results indicate that GCDSAEMDA outperforms the other four models in MDA prediction. Additionally, case studies demonstrate the robust predictive capability of GCDSAEMDA. The source code and datasets for GCDSAEMDA are available at https://github.com/chenyunmolu/GCDSAEMDA .}, archive = {J_EAAI}, author = {Bo Wang and Wenlong Zhao and Xiaoxin Du and Jianfei Zhang and Chunyu Zhang and Liping Wang and Yang He}, doi = {10.1016/j.engappai.2025.110301}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110301}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predicting potential microbe-disease associations based on heterogeneous graph attention network and deep sparse autoencoder}, volume = {147}, year = {2025}, }
• (2025). Intelligent application of interactive scale transformer for fine grained feature extraction in sheep. EAAI, 147, 110300. (www)

  @article{ , abstract = {In modern animal husbandry, artificial intelligence helps accurately manage individual sheep. However, it is difficult to recognize the sheep’s facial features and capture the nuances. It is not easy to extract the fine-grained features of a sheep’s face because the traditional vision transformer cannot realize the effective embedding of the interaction scale. To address this problem, we propose a novel sheep Transformer tool called SheepFormer . This model comprises components such as the Interactive Scale Embedded Images Block ( ISEI ), Patch Short Long Distance Attention Module ( PSLDA ), Dynamic Relative Position Offset ( DRPO ), and Transformer Neck and Head ( TNH ). These components are designed to embed features at multiple scales, fuse long and short-distance self-attention, adaptively handle relative position offsets for various group sizes, and introduce a prediction head to detect fine-grained facial targets in sheep at different scales. SheepFormer integrates Residual Attention to seek dense facial features in sheep and utilizes a Transformer Head to replace the traditional head, exploring the predictive potential of self-attention mechanisms in sheep faces. Experimental results demonstrate a 7.1% improvement in average precision (AP) for sheep face detection compared to Collaborative DEtection TRansformer (CO-DETR) and an 11.12% enhancement in accurate classification of sheep identity document (ID) compared to Shifted Windows Transformer (Swin Transformer). This study demonstrates that SheepFormer can extract fine-grained facial features of sheep, which promotes the advancement of high-precision sheep individual recognition and provides a guide for recognizing kinship.}, archive = {J_EAAI}, author = {Xingran Guo and Haizheng Yu and Hong Bian and Wenrong Li and Xueying Liao and Yongqi Zhu}, doi = {10.1016/j.engappai.2025.110300}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110300}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent application of interactive scale transformer for fine grained feature extraction in sheep}, volume = {147}, year = {2025}, }
• (2025). Hybrid deep-learning prediction model based on kernel multi-granularity fuzzy rough sets and its application in the diagnosis and treatment of chronic kidney disease. EAAI, 147, 110297. (www)

  @article{ , abstract = {In uncertain decision-making scenarios, quantitative scientific prediction models and methods can provide valuable support for making scientific decisions. However, the characteristics of hybrid attribute information may lead to a series of issues. These include difficulties in comparing and comprehensively evaluating different types of attributes, nonlinear relationships between attributes, and a lack of effective decision-support methods. To overcome these issues, this study introduces a kernel function to abstract the similarity of different attribute types and proposes a model called kernel multi-granularity fuzzy rough sets (KMGFRS). The KMGFRS model facilitates a thorough exploration and analysis of the uncertainties associated with decision objects. Additionally, an attribute reduction method based on KMGFRS is discussed to address redundant attributes in hybrid information systems. This method eliminates attributes that have a minimal influence on the decision results, simplifies the decision process, and enhances its effectiveness. This study integrates the KMGFRS and hybrid deep learning concepts to propose a novel prediction method aimed at enhancing accuracy and robustness. From the perspective of hybrid attribute information, this method can more accurately predict the unknown attributes of decision objects, thereby providing robust support for disease prediction in medical diagnostics and therapeutic decision-making. The experimental results indicated that the constructed model effectively handled uncertain decision-making scenarios involving hybrid attributes and fuzzy decision objects. It provides accurate and reliable decision support for chronic kidney disease (CKD), significantly enhancing the predictive accuracy of CKD types.}, archive = {J_EAAI}, author = {Jiqian Liu and Bingzhen Sun and Jin Ye and Xixuan Zhao and Xiaoli Chu}, doi = {10.1016/j.engappai.2025.110297}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110297}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid deep-learning prediction model based on kernel multi-granularity fuzzy rough sets and its application in the diagnosis and treatment of chronic kidney disease}, volume = {147}, year = {2025}, }
• (2025). Enhancing cross-lingual hate speech detection through contrastive and adversarial learning. EAAI, 147, 110296. (www)

  @article{ , abstract = {The rise of hate speech on social media platforms, particularly in low-resource languages, necessitates innovative solutions. In response, we introduce a zero and few-shot model combining supervised contrastive learning and adversarial training. To address the scarcity of labeled data in diverse languages, our approach adapts features from well-resourced languages to efficiently detect hate speech in low-resource contexts. The proposed framework first leverages supervised contrastive learning, maximizing the utility of limited labeled data by transferring knowledge from source languages. This adaptation empowers the accurate detection of hate speech in underrepresented languages, optimizing available resources. We then introduce contrastive adversarial training, refining hate speech representations in low-resource languages. This approach ensures a nuanced understanding of hate speech across linguistic boundaries, significantly enhancing the model’s adaptability and accuracy. To validate our approach, we conducted zero-shot and few-shot cross-lingual evaluations in three languages. Our results demonstrate the superiority of the proposed contrastive learning-based models. To ensure reproducibility, the code associated with this paper is available on GitHub ( Almahdi, 2024 ). .}, archive = {J_EAAI}, author = {Asseel Jabbar Almahdi and Ali Mohades and Mohammad Akbari and Soroush Heidary}, doi = {10.1016/j.engappai.2025.110296}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110296}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing cross-lingual hate speech detection through contrastive and adversarial learning}, volume = {147}, year = {2025}, }
• (2025). Meta-machine learning framework for robust short-term solar power prediction across different climatic zones. EAAI, 147, 110295. (www)

  @article{ , abstract = {The global energy landscape is increasingly dominated by solar power installations, driven by the sun’s position as Earth’s most abundant and sustainable energy resource. However, the intermittent nature of solar radiation, influenced by both astronomical cycles and meteorological conditions, creates significant challenges for reliable power generation and grid integration. To address the issue of uncertainty, this study proposes a robust and improved capacity machine learning framework with enhanced hypothesis functional space. The proposed model improves the capacity of an individual model by combining the hypothesis functions of individual machine learning models, increasing the representational capacity and hence the model’s generalization. Moreover, a non-linear second stage is stacked to increase the depth of the proposed model, which utilizes meta-data of first stage to further improve the forecasting accuracy. Furthermore, the proposed model is validated on four different climatic zones of the world for solar power forecasting. The proposed model achieves an average improvement of 66.7% in mean absolute error across all locations compared to the next best performing algorithm, with particularly strong performance in arid zones. Statistical validation through Cook’s distance analysis also confirms the model’s reliability with an average of 8.64% influential points across all locations.}, archive = {J_EAAI}, author = {Amit Rai and Ashish Shrivastava and Kartick C. Jana and Jay Liu and Kulwant Singh and N.S. Jayalakshmi and Amit Agrawal}, doi = {10.1016/j.engappai.2025.110295}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110295}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Meta-machine learning framework for robust short-term solar power prediction across different climatic zones}, volume = {147}, year = {2025}, }
• (2025). Pythagorean fuzzy quasi coincidence: Analysis and applications. EAAI, 147, 110291. (www)

  @article{ , abstract = {This article introduces an improved definition of Pythagorean fuzzy quasi-coincidence between the Pythagorean fuzzy sets enhancing the theoretical foundation of previous approaches. We investigate the theoretical aspects of the introduced Pythagorean fuzzy quasi-coincidence and corresponding Pythagorean fuzzy quasi-coincident set. The Pythagorean fuzzy quasi-coincident set is proven to be a Pythagorean fuzzy t-norm and the corresponding t-conorm with respect to the standard negation operator has been obtained. The introduced concept can reveal interrelationships among Pythagorean fuzzy sets, overcoming limitations of the fuzzy version of quasi-coincidence in scenarios involving uncertainty and hesitancy. Additionally, it enables range divisions, a feature previously unattainable in fuzzy quasi-coincidence methods. A significant novelty lies in the development of a Pythagorean fuzzy generator, the first of its kind, to generate Pythagorean fuzzy data from conventional fuzzy or Intuitionistic fuzzy information. This generator, accompanied by a generator sequence, allows customizable non-membership values based on user requirements. These advancements facilitate applications such as identifying high-risk zones during pandemics or natural disasters. Building on this foundation, we present practical applications to identify high-risk areas within regions affected by outbreaks of pandemics. We use he quasi-coincidence property, and group the affected area into distinct categories, such as red and yellow zones. Further, the application of the proposed theories is also depicted in medical diagnosis and we show that the method can also be used to re-frame traditional multi-criteria decision-making processes.}, archive = {J_EAAI}, author = {Subhankar Jana and Anjali Patel and Juthika Mahanta}, doi = {10.1016/j.engappai.2025.110291}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110291}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pythagorean fuzzy quasi coincidence: Analysis and applications}, volume = {147}, year = {2025}, }
• (2025). Global polynomial synchronization of proportional delay memristive neural networks with uncertain parameters and its application to image encryption. EAAI, 147, 110290. (www)

  @article{ , abstract = {This article explores the global polynomial synchronization (GPS) for a type of proportional delay memristive neural networks (PDMNNs), uncertain parameters are considered. First, the theory of differential inclusion is utilized, and then the error system is obtained. Secondly, combining the principles of sliding mode control (SMC) and adaptive control, two different controllers are designed to achieve GPS between the obtained drive–response system. Then, two GPS criteria are obtained through the application of Lyapunov stability theory and inequality analysis techniques. Ultimately, we offer three numerical exemplifications to corroborate the efficacy of the obtained results, along with a demonstration of an application pertaining to image encryption.}, archive = {J_EAAI}, author = {Yan Wan and Liqun Zhou and Jiapeng Han}, doi = {10.1016/j.engappai.2025.110290}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110290}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Global polynomial synchronization of proportional delay memristive neural networks with uncertain parameters and its application to image encryption}, volume = {147}, year = {2025}, }
• (2025). One-dimensional input space modelling of a simplified general type-2 mamdani and Takagi–Sugeno fuzzy proportional integral derivative controller. EAAI, 147, 110289. (www)

  @article{ , abstract = {A wide variety of fuzzy controllers have evolved over several decades. Several mathematical models of type-1 and interval type-2 fuzzy controllers have been explored. Most of these modelling approaches involved two-/three- dimensional input space. In this work, we have presented a simplified modelling approach for a General Type-2 (GT2) Mamdani and Takagi–Sugeno (TS) fuzzy Proportional Integral Derivative (PID) controllers involving input space of one-dimension. The fuzzy PID controller’s structure is the parallel combination of the Fuzzy Proportional (FP) plus the Fuzzy Integral (FI) plus the Fuzzy Derivative (FD) control actions. Having a parallel PID control structure simplifies the fuzzy ‘ I F − T H E N ’ rules by eliminating the role of ‘ A N D ’ (triangular norms) and ‘ O R ’ (triangular co-norms) operators. This decoupled rule base aids in decreasing the computing complexity of the GT2 fuzzy PID controller. Owing to the one-dimensional input space, the number of tuneable parameters for fuzzy controllers reduces significantly when compared to two- or three-dimensional input spaces. It is also demonstrated that the type-1 (T1) and interval type-2 (IT2) fuzzy controllers are variations of the GT2 fuzzy controller. In order to assess the controller models, we simulate two systems: the unstable first-order system with dead time and the Continuously Stirred Tank Reactor (CSTR). These models, nevertheless, may also be applied to other dynamic processes and systems.}, archive = {J_EAAI}, author = {Ritu Raj}, doi = {10.1016/j.engappai.2025.110289}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110289}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {One-dimensional input space modelling of a simplified general type-2 mamdani and Takagi–Sugeno fuzzy proportional integral derivative controller}, volume = {147}, year = {2025}, }
• (2025). Single-stage overlapping entity and relation extraction based on relation-specific heterogeneous graph neural network. EAAI, 147, 110288. (www)

  @article{ , abstract = {Entity and relation extraction, a crucial aspect of information extraction, aims to distill structured knowledge from unstructured text, vital for downstream applications such as knowledge graph construction, intelligent Q&A systems, and machine translation. A significant challenge lies in the fact that sentences often contain multiple overlapping entity pairs, significantly complicating the extraction process. Our objective is to efficiently extract these overlapping entities and their corresponding relations. Current approaches primarily focus on learning entity information while neglecting the associated relations. Furthermore, these methods often decompose the task into multiple processing steps, ignoring the strong interdependencies among the three elements of a triple. This results in issues such as cascading errors, redundant information, and exposure bias. Additionally, matrix-based labeling methods have significantly hindered operational efficiency. To address these issues, we introduce the Relation-Specific Heterogeneous Graph Neural Network (RSHGNN) model. RSHGNN directly obtains triples in a single-stage manner. It leverages a heterogeneous graph to integrate information about entities and relations. This approach eliminates cascading errors and exposure bias. Furthermore, the model employs a relation-specific labeling strategy and a negative sampling training method to enhance memory utilization. Extensive experiments demonstrate that RSHGNN exhibits superior performance in handling overlapping situations and achieves state-of-the-art results on two public datasets: NYT and WebNLG.}, archive = {J_EAAI}, author = {Dijing Pan and Runhe Qiu and Xueqin Jiang}, doi = {10.1016/j.engappai.2025.110288}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110288}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Single-stage overlapping entity and relation extraction based on relation-specific heterogeneous graph neural network}, volume = {147}, year = {2025}, }
• (2025). Lattice-based sensor data acquisition strategy to solve sensor position drift in human gait phase recognition system with a single inertia measurement unit. EAAI, 147, 110286. (www)

  @article{ , abstract = {The inertia measurement unit (IMU) sensor units have garnered considerable utilization in gait phase recognition systems owing to their inherent data stability and compatibility with low-usage conditions. The extant scholarship concerning gait phase recognition predicated upon IMU sensors manifests a concentrated endeavor to augment the accuracy of recognition through the employment of diverse recognition algorithms and the fusion of multiple sensors. However, the impact of the drift of the IMU sensor position on the accuracy of the gait phase recognition algorithm during training and use is ignored, which is especially important for a single IMU gait recognition system. Therefore, taking convolutional neural network, deep neural network, support vector machine, and random forest algorithms as examples, this paper studies the impact of IMU sensor position drift on the accuracy of gait phase recognition algorithms. To address the quandary of position drift, this study proposes an innovative lattice-based data acquisition strategy for a single IMU sensor by selecting 9 uniformly distributed points on the posterior region of the calf. A treadmill walking test wearing an embedded IMU data acquisition system was organized to verify the practical performance of the lattice-based data acquisition strategy. By comparing the performance of different lattice combinations, a 5-point sensor acquisition strategy is proposed, which can effectively increase the accuracy of gait phase recognition with IMU sensor position drift by more than 10.29%.}, archive = {J_EAAI}, author = {Dianbiao Dong and Nannan Zhu and Jiehong Wang and Yuzhu Li}, doi = {10.1016/j.engappai.2025.110286}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110286}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lattice-based sensor data acquisition strategy to solve sensor position drift in human gait phase recognition system with a single inertia measurement unit}, volume = {147}, year = {2025}, }
• (2025). A dual-method approach using autoencoders and transductive learning for remaining useful life estimation. EAAI, 147, 110285. (www)

  @article{ , abstract = {Estimating the remaining useful life (RUL) of lithium-ion batteries presents a critical challenge, as it necessitates predicting their future performance and lifespan under diverse operational conditions. Addressing this issue is crucial for enhancing battery maintenance, improving reliability, and safeguarding devices that depend on lithium-ion technology. In this article, we propose a dual-method approach for RUL estimation. Firstly, an autoencoder (AE) extracts pivotal features from the input. Key measurable parameters, such as voltage, current, and temperature from charging profiles, are derived from the battery management system, providing robust data for the AE. The core of the AE is constructed using a spatial attention-based transductive long short-term memory (TLSTM) model, which is trained with an advanced generative adversarial network (GAN). The TLSTM model employs transductive learning, emphasizing samples near the test point to refine the fitting process and surpassing conventional LSTM models in performance. Following the AE training phase, the input's latent representation is inputted into a multilayer perceptron (MLP) designed for RUL prediction. We conduct thorough evaluations using National Aeronautics and Space Administration (NASA) datasets. Additionally, experiments from the Center for Advanced Life Cycle Engineering (CALCE) at the University of Maryland are underway to examine the influence of transfer learning (TL) on our model. The TLSTM model performs better than other deep learning models, achieving an impressive mean absolute percentage error (MAPE) ranging between 0.0053 and 0.0095. This highlights the efficacy and superiority of our approach in accurately predicting RUL, offering significant potential benefits for industries reliant on energy storage systems.}, archive = {J_EAAI}, author = {Jing Yang and Nika Anoosha Boroojeni and Mehran Kazemi Chahardeh and Lip Yee Por and Roohallah Alizadehsani and U. Rajendra Acharya}, doi = {10.1016/j.engappai.2025.110285}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110285}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual-method approach using autoencoders and transductive learning for remaining useful life estimation}, volume = {147}, year = {2025}, }
• (2025). Validation and extraction of reliable information through automated scraping and natural language inference. EAAI, 147, 110284. (www)

  @article{ , abstract = {In today’s day and age where information is spread rapidly through online platforms, the rise of fake news poses an alarming threat to the integrity of public discourse, societal trust, and reputed news sources. Classical machine learning and transformer-based models have been studied extensively for claim verification, however, they are hampered by their reliance on static training data and cannot generalize on unseen headlines. To address these challenges, we propose our explainable solution, which leverages web-information retrieval techniques and Natural Language Inference models to verify the veracity of a news headline. We evaluate our solution on a diverse self-curated evaluation dataset spanning multiple news channels and domains. Our best-performing pipeline achieves an accuracy of 84.3% surpassing the best classical Machine Learning model by 33.3% and Bidirectional Encoder Representations from Transformers by 31.0% . Utilizing hardware accelerators our pipelines achieve end-to-end fact verification inference times ranging from 2.92-6.97 seconds. Our approach highlights the efficacy of combining dynamic information retrieval with Natural Language Inference to find support for a claimed headline in the corresponding externally retrieved knowledge. The respective code, dataset, and results of this study are available in our artifact: https://github.com/Arjun-254/VERITAS-NLI .}, archive = {J_EAAI}, author = {Arjun Shah and Hetansh Shah and Vedica Bafna and Charmi Khandor and Sindhu Nair}, doi = {10.1016/j.engappai.2025.110284}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110284}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Validation and extraction of reliable information through automated scraping and natural language inference}, volume = {147}, year = {2025}, }
• (2025). Recent advances in flotation froth image analysis via deep learning. EAAI, 147, 110283. (www)

  @article{ , abstract = {Flotation froth image analysis with computer vision systems has witnessed a transformative evolution through the integration of deep learning. Deep learning outperforms traditional feature design by effectively learning intricate feature representations, thus enhancing the assessment of froth flotation processes' operational performance. Flotation froth image analysis via deep learning facilitates real-time monitoring of dynamic flotation processes, guiding the adjustment of operational variables through predicting performance indicators, recognizing froth states and segmenting foam edges, which promotes resource efficiency and supports the sustainable development of beneficiation. Despite the vast potential of deep learning for time-series forecasting within the multistage flotation cycle, its capabilities remain underexplored. To fill this gap, based on recent research, we discuss the application of temporal and multistage information in flotation cycle. We introduce the development trends of deep learning in various processes of flotation froth image analysis, including data collection, dataset preprocessing, feature extraction, and modeling. We particularly discuss advanced techniques for extracting time-series features, and developing multistage models and innovative data collection methods, so as to emphasize the importance of using temporal information. Eventually, the review explores several trends and challenges for future research. This review is expected to leave readers with deeper thoughts about algorithm design and data collection in the flotation domain, thereby promoting further research and development in beneficiation automation.}, archive = {J_EAAI}, author = {Xin Chen and Dan Liu and Longzhou Yu and Ping Shao and Mingyan An and Shuming Wen}, doi = {10.1016/j.engappai.2025.110283}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110283}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Recent advances in flotation froth image analysis via deep learning}, volume = {147}, year = {2025}, }
• (2025). Explainable differential privacy-hyperdimensional computing for balancing privacy and transparency in additive manufacturing monitoring. EAAI, 147, 110282. (www)

  @article{ , abstract = {Machine Learning (ML) models integrated with in-situ sensing offer transformative solutions for defect detection in Additive Manufacturing (AM), but this integration brings critical challenges in safeguarding sensitive data, such as part designs and material compositions. Differential Privacy (DP), which introduces mathematically controlled noise, provides a balance between data utility and privacy. However, black-box Artificial Intelligence (AI) models often obscure how this noise impacts model accuracy, complicating the optimization of privacy–accuracy trade-offs. This study introduces the Differential Privacy-Hyperdimensional Computing (DP-HD) framework, a novel approach combining Explainable AI (XAI) and vector symbolic paradigms to quantify and predict noise effects on accuracy using a Signal-to-Noise Ratio (SNR) metric. DP-HD enables precise tuning of DP noise levels, ensuring an optimal balance between privacy and performance. The framework has been validated using real-world AM data, demonstrating its applicability to industrial environments. Experimental results demonstrate DP-HD’s capability to achieve state-of-the-art accuracy (94.43%) with robust privacy protections in anomaly detection for AM, even under significant noise conditions. Beyond AM, DP-HD holds substantial promise for broader applications in privacy-sensitive domains such as healthcare, financial services, and government data management, where securing sensitive data while maintaining high ML performance is paramount.}, archive = {J_EAAI}, author = {Fardin Jalil Piran and Prathyush P. Poduval and Hamza Errahmouni Barkam and Mohsen Imani and Farhad Imani}, doi = {10.1016/j.engappai.2025.110282}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110282}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable differential privacy-hyperdimensional computing for balancing privacy and transparency in additive manufacturing monitoring}, volume = {147}, year = {2025}, }
• (2025). Confidence-driven under-sampling decision forest for imbalanced credit scoring. EAAI, 147, 110278. (www)

  @article{ , abstract = {Credit scoring is vital for assessing financial trustworthiness. Ensemble frameworks and resampling-based strategies have great potential in handling imbalanced credit scoring. However, resampling-based ensemble learning can potentially distort the original credit dataset distribution, thereby posing challenges in optimizing each base predictor. Additionally, achieving balance in the entire sample space with one-step resampling is challenging and can lead to information loss. To tackle these issues, a confidence-driven under-sampling decision forest (CUDF) is proposed. CUDF is established as an ensemble approach, harnessing the collective power of multiple decision trees; in addition, a confidence-driven under-sampling (CDUS) mechanism is incorporated into CUDF, enabling each base learner within CUDF to become an imbalance-aware predictor; besides, a weighted aggregation scheme is designed for CUDF enable precise fine-tuning of weight allocation for each base classifier, facilitating tailored adjustments to maximize their contributions for imbalanced credit scoring. Empirical findings substantiate that the CUDF surpasses the performance of both traditional credit scoring models and imbalanced credit scoring models across four distinct credit scoring datasets. The algorithm not only amplifies the precision in categorizing risky borrower classes but also demonstrates notable efficiency in managing extensive imbalanced credit scoring datasets. Consequently, CUDF exhibits promising potential for applications in addressing imbalanced data challenges within the domain of credit scoring.}, archive = {J_EAAI}, author = {Meng Xia and Zhijie Wang and Xingyu Lan and Wanan Liu and Jiawei Wu}, doi = {10.1016/j.engappai.2025.110278}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110278}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Confidence-driven under-sampling decision forest for imbalanced credit scoring}, volume = {147}, year = {2025}, }
• (2025). A dual residual dense network for image denoising. EAAI, 147, 110275. (www)

  @article{ , abstract = {Digital images have applications in almost every field, such as medicine, autonomous vehicles, astronomy, etc., for various purposes. The performance of all these applications depends on the quality of the images; that is, digital images should be noise-free for better results. Due to various factors, images capture noise at the acquisition time, which needs to be removed before subsequent applications. Therefore, this paper presents a dual residual dense network for image denoising. The network is designed using three types of blocks: feature extraction blocks, combination blocks, and residual blocks. These blocks exploit residual learning, dense connectivity, and batch renormalization to remove image noise. The network is also designed to be wide rather than deep, which makes it computationally and time-efficient. The network was trained on the DIV2K dataset and tested on the Kodak24 and Berkeley Segmentation (BSDS300) datasets. The results show that the proposed network outperforms existing state-of-the-art architectures in image denoising, achieving peak signal-to-noise ratio (PSNR) scores of 35.03 for the Kodak dataset and 34.64 for the BSDS300 dataset, as well as structural similarity index measure (SSIM) scores of 0.99 for both datasets. The code of the proposed method is publicly available to allow others to reproduce the findings and validate the results.}, archive = {J_EAAI}, author = {Isma Batool and Muhammad Imran}, doi = {10.1016/j.engappai.2025.110275}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110275}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual residual dense network for image denoising}, volume = {147}, year = {2025}, }
• (2025). Hybrid long short-term memory and bidirectional multichannel network cascaded with split convolution for short-term load forecasting. EAAI, 147, 110268. (www)

  @article{ , abstract = {Accurate multi-horizon Short-Term Load Forecasting (STLF) is essential for load scheduling, effective energy trading, unit commitment, and intelligent demand response. However, due to the integration of highly intermittent distributed renewable generation sources and the dynamic load behavior of prosumers, an accurate load forecasting with already existing methods is challenging. To overcome this challenge, a novel hybrid multi-channel parallel LSTM–BLSTM sub-network cascaded in series with a modified split convolution (SC) framework is proposed for single-step and multi-step STLF. The multi-channel parallel LSTM–BLSTM sub-network extracts the sequence-dependent features and modified SC extracts multi-scale hierarchical spatial features. The power consumption data is also modified for multi-channel sub-network. The historical load data is applied to BLSTM for extracting patterns in both forward and backward directions. On the other hand, load data concatenated with highly correlated calendric features is applied to the LSTM module. The proposed framework is evaluated on American Electric Power (AEP) dataset. For generalization capability, the performance of the model is tested on five publicly available datasets: AEP, ComEd, Malaysia, ISONE, and Turkey. The evaluation parameters such as MAE, RMSE, and MAPE of the proposed framework are 474.2, 668.6, and 3.16 respectively for 24 h ahead, 358.5, 512.5, and 2.39 for 12 h ahead, and 95.4, 126.8 and 0.52 for a single step ahead respectively. The results are compared with the different existing state-of-the-art on AEP and four other publicly available datasets. The result shows that the proposed method has less forecasting error, strong generalization capability, and satisfactory performance on multi-horizon.}, archive = {J_EAAI}, author = {Syed Muhammad Hasanat and Irshad Ullah and Khursheed Aurangzeb and Muhammad Rizwan and Musaed Alhussein and Muhammad Shahid Anwar}, doi = {10.1016/j.engappai.2025.110268}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110268}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid long short-term memory and bidirectional multichannel network cascaded with split convolution for short-term load forecasting}, volume = {147}, year = {2025}, }
• (2025). A residual learning-based grey system model and its applications in electricity transformer’s seasonal oil temperature forecasting. EAAI, 147, 110260. (www)

  @article{ , abstract = {Accurately predicting cross-regional electricity demand is crucial for efficient distribution management, but it remains challenging due to its complexity. Transformer oil temperature is a key indicator of operational status, and analyzing its seasonal variation is vital for addressing distribution issues. Grey models based on neural networks are effective for predicting nonlinear and small-scale datasets but are prone to overfitting. While residual networks help mitigate overfitting, their application to small-scale time series forecasting is still limited. To improve prediction accuracy for nonlinear and small-scale data, this study introduces residual learning into grey models, proposing a hybrid model. This model combines the feature-capturing ability of residual learning networks with the robustness of grey models, helping to reduce overfitting. The model is trained using the Adam algorithm, with parameters optimized by the Gridsearch algorithm. Performance is demonstrated using four seasonal datasets of transformer oil temperature. A comparison with 13 grey system models and 9 machine learning models shows that the proposed method outperforms the others. By calculating the percentage improvements of various metrics, the model demonstrates consistent performance gains. Sensitivity analysis reveals that the model’s performance is sensitive to the number of neurons and network depth, with higher values significantly improving accuracy and robustness. The results confirm the model’s effectiveness. This study fills the gap between neural grey models and residual networks, successfully applying the model to forecast the seasonal temperature trends of power transformers and providing a theoretical basis for addressing power distribution challenges.}, archive = {J_EAAI}, author = {Yiwu Hao and Xin Ma and Lili Song and Yushu Xiang}, doi = {10.1016/j.engappai.2025.110260}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110260}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A residual learning-based grey system model and its applications in electricity transformer’s seasonal oil temperature forecasting}, volume = {147}, year = {2025}, }
• (2025). State-of-the-art review and benchmarking of barcode localization methods. EAAI, 147, 110259. (www)

  @article{ , abstract = {Barcodes, despite their long history, remain an essential technology in supply chain management. In addition, barcodes have found extensive use in industrial engineering, particularly in warehouse automation, component tracking, and robot guidance. To detect a barcode in an image, multiple algorithms have been proposed in the literature, with a significant increase of interest in the topic since the rise of deep learning. However, research in the field suffers from many limitations, including the scarcity of public datasets and code implementations which hinders the reproducibility and reliability of published results. For this reason, we developed “BarBeR” (Barcode Benchmark Repository), a benchmark designed for testing and comparing barcode detection algorithms. This benchmark includes the code implementation of various detection algorithms for barcodes, along with a suite of useful metrics. Among the supported localization methods, there are multiple deep-learning detection models, that will be used to assess the recent contributions of Artificial Intelligence to this field. In addition, we provide a large, annotated dataset of 8 748 barcode images, combining multiple public barcode datasets with standardized annotation formats for both detection and segmentation tasks. Finally, we provide a thorough summary of the history and literature on barcode localization and share the results obtained from running the benchmark on our dataset, offering valuable insights into the performance of different algorithms when applied to real-world problems.}, archive = {J_EAAI}, author = {Enrico Vezzali and Federico Bolelli and Stefano Santi and Costantino Grana}, doi = {10.1016/j.engappai.2025.110259}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110259}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {State-of-the-art review and benchmarking of barcode localization methods}, volume = {147}, year = {2025}, }
• (2025). The two-tier multi-depot vehicle routing problem with robot stations and time windows. EAAI, 147, 110258. (www)

  @article{ , abstract = {The deployment of sidewalk robots or automated guided vehicles (AGVs) offers significant potential to reduce traffic congestion and operational costs in last-mile logistics. Delivery systems using a two-layer structure, with trucks in the first layer and sidewalk robots at stations in the second, are modeled as Two-Tier Vehicle Routing Problems with Robot Stations (2T-VRP-RS). In this paper, we extend the 2T-VRP-RS to introduce the Two-Tier Multi-Depot Vehicle Routing Problem with Robot Stations and Time Windows (2T-MDVRP-RS-TW). We propose a Mixed-Integer Linear Programming (MILP) formulation and a Multi-Start Iterated Local Search metaheuristic with Crossover and Repair (MS-ILS-CR) operators. The MS-ILS-CR combines three strategies: a Multi-Start mechanism, a Crossover operator, and a Repair operator adapted from Variable Neighborhood Search, Genetic Algorithms, and Adaptive Large Neighborhood Search, respectively. We introduce new instance sets to study the 2T-MDVRP-RS-TW. Results showed that the MILP formulation is competitive for small instances of 10 customers. However, the metaheuristic consistently achieved better solution quality and lower computational times, with improvements of over 20% in the average objective values for instances of up to 30 customers. The MS-ILS-CR outperformed similar metaheuristic methods, improving its initial solutions by 66%. Furthermore, a sensitivity analysis showed a limited impact of time-window variations on the objective values. Nevertheless, a significant sensitivity was found to the depot configurations. Hence, increasing the number of depots from 1 to 3 improved the average objective values by 56%. Additionally, the MS-ILS-CR performance was further enhanced with more crossover iterations, but at the cost of higher computational times.}, archive = {J_EAAI}, author = {Giovanni Campuzano and Eduardo Lalla-Ruiz and Martijn Mes}, doi = {10.1016/j.engappai.2025.110258}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110258}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The two-tier multi-depot vehicle routing problem with robot stations and time windows}, volume = {147}, year = {2025}, }
• (2025). A three layer stacked multimodel transfer learning approach for deep feature extraction from chest radiographic images for the classification of COVID-19. EAAI, 147, 110241. (www)

  @article{ , abstract = {COVID-19 has had a profound impact on global health, targeting the human respiratory system and causing significant disruptions to human life. To aid in effective diagnosis and classification, numerous machine learning and deep learning models have been developed to analyse limited Chest Radiographic and Computed Tomography images. In this study, we propose a three-layer stacked multimodal approach for deep feature extraction from a high volume of COVID-19 Chest Radiographic images. The proposed model utilizes eight transfer learning models pre-trained on the ImageNet dataset, evaluated based on key performance metrics such as accuracy, precision, and recall. The unique stacking model integrates the outputs of these transfer learning models, extracting deep features through a three-layer architecture. These features are flattened, concatenated, and passed through seven deep dense layers with varying kernel and bias dimensions to achieve optimal classification performance. The approach was applied to a large COVID-19 Chest Radiographic images dataset, achieving the highest accuracy (95.79%), precision (95.44%), and recall (96.65%) when compared to state-of-the-art models. This study demonstrates the effectiveness of leveraging a stacked transfer learning multimodal framework for COVID-19 diagnosis. The proposed method not only ensures high accuracy but also provides a computationally efficient solution for analysing large-scale radiographic datasets, positioning it as a robust tool for aiding in the early detection and classification of COVID-19.}, archive = {J_EAAI}, author = {Baijnath Kaushik and Akshma Chadha and Abhigya Mahajan and Malvika Ashok}, doi = {10.1016/j.engappai.2025.110241}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110241}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A three layer stacked multimodel transfer learning approach for deep feature extraction from chest radiographic images for the classification of COVID-19}, volume = {147}, year = {2025}, }
• (2025). Drug–target affinity prediction using rotary encoding and information retention mechanisms. EAAI, 147, 110239. (www)

  @article{ , abstract = {Drug–target affinity (DTA) prediction has been widely used in pharmaceutical research as a novel and effective method to explore the interaction strength between drugs and targets. However, existing DTA prediction models mainly rely on graphical representations of drug molecules, overlooking the intricate interactions between individual substructures. This limitation impacts both the predictive accuracy and the informational richness within the model nodes. To address these challenges, this paper proposes the Rotary Retention Graph Drug–Target Affinity (RRGDTA) network with rotation and retention mechanisms. The RRGDTA integrates an information interaction module into the extraction of drug and target features across multiple scale levels. This approach enhances the correlation within the graph representation, leading to an optimal feature representation. Furthermore, to tackle the issue of limited relationship between molecular structure and context, a Multi-Scale Interaction module (MSI) is proposed to enhance important features related to both. Additionally, to address inaccuracies in the structural features of drugs and targets, a Rotary Encoding Module (ROE) is proposed, which focuses on nearby contextual information and effectively captures the correlation between them. In order to solve the problem of insufficient information representation, an Association Prediction Module (APM) and an Intra-Mask Retention Module (IMR) are proposed to maximize the retention of drug–target information. The efficacy of the proposed RRGDTA in DTA prediction was validated on the Davis, kinase inhibitors biochemical assays (KIBA) and binding database (BindingDB) datasets. Compared with current baseline models, the proposed model achieved better results across various metrics, demonstrating its superior performance in accurate DTA prediction.}, archive = {J_EAAI}, author = {Zhiqin Zhu and Yan Ding and Guanqiu Qi and Baisen Cong and Yuanyuan Li and Litao Bai and Xinbo Gao}, doi = {10.1016/j.engappai.2025.110239}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110239}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Drug–target affinity prediction using rotary encoding and information retention mechanisms}, volume = {147}, year = {2025}, }
• (2025). Credit risk assessment of automobile loans using machine learning-based SHapley additive exPlanations approach. EAAI, 147, 110236. (www)

  @article{ , abstract = {In recent years, artificial intelligence, especially machine learning methods, have been increasingly applied to the accurate prediction of various loan risks. However, the integration of these models into legal frameworks and technology deployment is hindered due to their so-called “black-box” nature. This paper aims to explore the credit risk assessment of automobile loans using machine learning-based SHapley Additive exPlanations approach. It examines the automobile loan data from a specific Chinese automobile finance company, applying a logistic regression model as the foundation, and integrates the eXtreme Gradient Boosting with SHapley Additive exPlanations approach to interpret the predictive outcomes. The SHapley Additive exPlanations approach effectively quantifies the contributions of individual features to the predictions, highlighting key features such as credit level, credit score, and disbursed_amount (the amount of loans disbursed) that significantly influence automobile loan risk. The contribution values of these key features to the model prediction results are found to be 0.67, 0.337, and 0.34, respectively. Moreover, the SHapley Additive exPlanations approach not only enhances the interpretability of the model, but also optimizes the accuracy and efficiency of automobile loan risk assessment. This study provides a robust tool for financial institutions, aiding in the reduction of loan risks and fostering the healthy growth of the automobile loan market. The findings are particularly significant in addressing the high delinquency rates of automobile loans in China and offer new perspectives and methodologies for future risk assessment research.}, archive = {J_EAAI}, author = {Shuoyan Lin and Dandan Song and Boyi Cao and Xin Gu and Jiazhan Li}, doi = {10.1016/j.engappai.2025.110236}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110236}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Credit risk assessment of automobile loans using machine learning-based SHapley additive exPlanations approach}, volume = {147}, year = {2025}, }
• (2025). Task recognition integrating worker actions and machine operations: A video-based sensing approach without physical sensors. EAAI, 147, 110232. (www)

  @article{ , abstract = {Automating work process analysis is crucial in manufacturing to improve efficiency and productivity. However, traditional deep learning methods often fail to capture subtle temporal changes in machine operations, such as varying speeds. We propose a cost-effective approach called pseudo-sensing, which simulates sensor data by measuring machine speeds directly from video using wavelet transformation, a mathematical tool for time-frequency analysis. This approach eliminates the need for physical sensors. We evaluated pseudo-sensing by integrating it into two task classification models. The first is a convolutional neural network-long short-term memory (CNN-LSTM) model, which extracts spatial features via a CNN and learns temporal patterns using an LSTM. The second is a three-dimensional residual network (3D ResNet, R3D), designed to process spatiotemporal data simultaneously. With pseudo-sensing, the CNN-LSTM’s micro-F1 score—an accuracy metric averaging precision and recall across all classes—improved from 0.712 to 0.736 (+2.4 points), while R3D’s score rose from 0.675 to 0.701 (+2.7 points). To assess general applicability, we tested pseudo-sensing on another dataset featuring diverse machine motions: unidirectional movements (e.g., conveyor belts), oscillatory movements (e.g., pendulum-like motions), rotational movements (e.g., rotary presses), and intermittent movements (e.g., blinking or toggling mechanisms). The method achieved an 83% success rate in identifying machine dynamics. By leveraging deep learning, this method integrates video-based machine operation sensing with task recognition, considering both human actions and machine states. Eliminating additional sensors while enhancing accuracy and efficiency, pseudo-sensing offers broad potential for advancing manufacturing process analysis.}, archive = {J_EAAI}, author = {Shotaro Kataoka and Masashi Oba and Hirofumi Nonaka}, doi = {10.1016/j.engappai.2025.110232}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110232}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Task recognition integrating worker actions and machine operations: A video-based sensing approach without physical sensors}, volume = {147}, year = {2025}, }
• (2025). MiFDeU: Multi-information fusion network based on dual-encoder for pelvic bones segmentation. EAAI, 147, 110230. (www)

  @article{ , abstract = {The segmentation of bone fragments is crucial for preoperative planning and intraoperative navigation in reduction surgery. Recent advances in medical segmentation have predominantly focused on U-shaped frameworks that employ convolutional neural networks or transformer variants as the backbone. However, these frameworks, which rely on a single encoder, often struggle with integrating information from diverse features and processing irregular shapes in visual objects. Such limitations can reduce segmentation accuracy and impair generalization performance across different datasets. To address these issues, we introduce a multi-information fusion network based on dual-encoder for pelvic bones segmentation. In order to capture global contextual information and local features simultaneously, our model takes a light resnet and a graph neural network with swin-pool module as dual-encoder for effectively representing the global and local topologies. We construct a high-low multi-dimensional paired attention in the bottleneck for fusing spatial and channel information from different dimensions. Instead of using the traditional dice loss in the unet-like architecture, our model employs both topological loss and boundary loss to enhance the goal optimization. In the experiments, our model achieves a substantially lower dice similarity coefficient and comparable 95 % Hausdorff distance compared to other state-of-the-art. The experiments on across datasets verify the superiority and generalization of the proposed model.}, archive = {J_EAAI}, author = {Fujiao Ju and Yichu Wu and Mingjie Dong and Jingxin Zhao}, doi = {10.1016/j.engappai.2025.110230}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110230}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MiFDeU: Multi-information fusion network based on dual-encoder for pelvic bones segmentation}, volume = {147}, year = {2025}, }
• (2025). Information guided attention network for bearing remaining useful life prediction adaptive to working conditions and fault modes. EAAI, 147, 110197. (www)

  @article{ , abstract = {Bearing remaining useful life (RUL) prediction is a major concern in prognostics and health management for industrial system. Recently, deep learning models have significantly advanced the development of RUL prediction technologies. However, the accuracy of deep learning-based RUL prediction methods is easily affected by variable working conditions and multiple fault modes. In this paper, an information guided attention network (IGAN) is developed for bearing RUL prediction adaptive to working conditions and fault modes. First, the proposed IGAN builds the multiscale convolutional layers, which are multiple dilated convolutions with different dilation rates, to fully learn multiscale representations. This ensures that no degradation information of bearings under different working conditions and different fault modes is missed. Second, a novel plug-and-play attention block called information guided attention mechanism (IGAM) is designed to adaptively highlight the informative convolutional channels informed by working conditions and fault modes. Finally, a temporal attention mechanism is integrated into the IGAN to adaptively emphasize the degradation features at different temporal locations to further enhance the feature representation. Two case studies on a bearing dataset across different working conditions and a bearing dataset under time-varying working conditions are conducted to validate the effectiveness and superiority of the proposed method.}, archive = {J_EAAI}, author = {Lei Wang and Hongrui Cao and Xuefeng Chen}, doi = {10.1016/j.engappai.2025.110197}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110197}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Information guided attention network for bearing remaining useful life prediction adaptive to working conditions and fault modes}, volume = {147}, year = {2025}, }
• (2025). A neural network guided dual-space search evolutionary algorithm for large scale multi-objective optimization. EAAI, 147, 110089. (www)

  @article{ , abstract = {The curse of dimensionality caused by the increase of decision variables in large-scale multi-objective problems (LSMOPs) is still the current challenge. Although existing algorithms can simple large-scale multi-objective optimization problems. Nevertheless, a single search strategy might impact the solution of large-scale multi-objective optimization problems. To solve this problem, a dual-space search evolutionary algorithm for large-scale multi-objective optimization is proposed. Firstly, in the decision space, a neural network assisted operator with adaptive strategy is introduced. Specifically, when the number of non-dominated solutions is decreasing, the neural network is adopted to optimize the solutions with poor fitness for breaking away from local optimality. After that, the objective space of population is divided into several sub-regions by k-means clustering strategy. The solutions in these subregions are mapped onto the decision space through the inverse model, so that population can obtain as many non-dominated solutions as possible. Finally, the proposed algorithm is tested on a real-life problem which is Time-varying Ratio Error Estimation (TREE) and two benchmark suites which are large-scale multi-objective optimization problem (LSMOP) and unconstrained front (UF). The results show that the proposed algorithm exhibits competitive performance compared to other state-of-the-art algorithms on Inverted Generational Distance (IGD) Indicator and Hyper-volume (HV) Indicator.}, archive = {J_EAAI}, author = {Jie Cao and Chengzhi Liu and Zuohan Chen and Jianlin Zhang and Peng Zhao}, doi = {10.1016/j.engappai.2025.110089}, journal = {Engineering Applications of Artificial Intelligence}, month = {5}, pages = {110089}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A neural network guided dual-space search evolutionary algorithm for large scale multi-objective optimization}, volume = {147}, year = {2025}, }
• (2025). Reducing overestimation with attentional multi-agent twin delayed deep deterministic policy gradient. EAAI, 146, 110352. (www)

  @article{ , abstract = {In multi-agent reinforcement learning, establishing effective communication protocols is crucial for enhancing agent collaboration. However, traditional communication methods face challenges in scalability and efficiency as the number of agents increases, due to the expansion in the dimensions of observation and action spaces. This leads to heightened resource consumption and degrades performance in large multi-agent scenarios. To address these issues, we introduce a novel Attentional Multi-agent Twin Delayed Deep Deterministic Policy Gradient (AMATD3) algorithm that incorporates an attentional communication policy gradient approach. This approach selectively initiates communications through an attention unit that assesses the necessity of information exchange among agents, combined with a communication module that effectively integrates essential information. By implementing a double-Q function, AMATD3 further addresses issues of overestimation and suboptimal policy choices in existing methods, enhancing the algorithm's accuracy and reducing communication overheads. Specifically, our algorithm demonstrates superior performance in the StarCraft II environment by achieving higher cumulative rewards and enhancing task success rates compared to existing algorithms. For example, AMATD3 yields reward values of 16.908 and 6.858 for the 8m and 25m scenarios, respectively, which is more than double the reward achieved by other methods. This confirms the algorithm's enhanced efficiency and effectiveness in complex multi-agent settings, contributing to the ongoing development of scalable and efficient communication protocols in artificial intelligence.}, archive = {J_EAAI}, author = {Yizhi Cao and Zijian Tian and Zhaoran Liu and Naizheng Jia and Xinggao Liu}, doi = {10.1016/j.engappai.2025.110352}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110352}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reducing overestimation with attentional multi-agent twin delayed deep deterministic policy gradient}, volume = {146}, year = {2025}, }
• (2025). On-board detection of rail corrugation using improved convolutional block attention mechanism. EAAI, 146, 110349. (www)

  @article{ , abstract = {Leveraging acceleration sensors affixed to the train body enables continuous surveillance of rail corrugation, delivering cost-effectiveness, operational efficiency, and portability. Establishing the correlation between vertical body acceleration and rail corrugation poses a substantial challenge. To ensure uninterrupted monitoring of rail corrugation, an initial development involved constructing a train-track integrated simulation model that accounted for the dynamics of flexible wheelsets and tracks, thereby generating a simulated dataset of vertical body acceleration. Subsequent improvements were made to the conventional Convolutional Block Attention Module (CBAM) architecture, culminating in the proposal of a deep one-dimensional convolutional residual network model named Train Body Vertical Acceleration Network (TBVA-Net), founded on an improved CBAM framework. Training was conducted using the simulated dataset, showcasing the reduced model complexity and total parameter count of the improved CBAM architecture, which notably amplified classification accuracy. The TBVA-Net, employing the refined CBAM, consistently achieved test accuracies exceeding 95%, averaging at 98.6% on the simulated dataset. Validation through field-measured data corroborated the rationale behind the proposed TBVA-Net architecture. Fine-tuning with a limited subset of labeled field data led to a transfer accuracy of 98.5%. This paper presents an innovative approach for detecting rail corrugation through vertical acceleration signals obtained from operational vehicles.}, archive = {J_EAAI}, author = {Yang Wang and Hong Xiao and Chaozhi Ma and Zhihai Zhang and Xuhao Cui and Aimin Xu}, doi = {10.1016/j.engappai.2025.110349}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110349}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {On-board detection of rail corrugation using improved convolutional block attention mechanism}, volume = {146}, year = {2025}, }
• (2025). Adaptive feature selection with flexible mapping for diagnosis and prediction of parkinson's disease. EAAI, 146, 110342. (www)

  @article{ , abstract = {Parkinson's disease (PD) is a neurodegenerative disorder in which symptoms gradually worsen over time, with no cure currently available. Therefore, its early-stage treatment is crucial. However, medical neuroimaging data often contain redundant features and high dimensionality, which can negatively impact algorithm accuracy and require greater computational resources. To address this challenge, a supervised feature selection algorithm is proposed for the early diagnosis of PD. Specifically, the proposed method incorporates adaptive learning during iterations, which enables adaptive updating of the similarity matrix and selection of informative features. Meanwhile, flexible mapping is introduced to address the limitations of strict linear mapping. To address the optimization challenge of the model, we propose an alternative iterative algorithm and provide theoretical proof of its strict convergence. To measure the effectiveness of the algorithm, we conducted experiments on the Parkinson's Progression Markers Initiative (PPMI) public dataset, involving three groups: PD vs. normal controls (NC), scans without evidence of dopaminergic deficit (SWEDD) vs. NC, and PD vs. SWEDD. Based on baseline data, the accuracies for these three groups were 83.29%, 87.67%, and 85.16%, respectively. Using the 12-month data, the accuracies improved to 79.70%, 96.79%, and 96.95%, while the 24-month data resulted in accuracies of 91.33%, 70.10%, and 93.36%. The experimental results demonstrate that the proposed method outperforms existing feature selection methods in overall performance.}, archive = {J_EAAI}, author = {Zhongwei Huang and Jianqiang Li and Jiatao Yang and Jun Wan and Jianxia Chen and Zhi Yang and Ming Shi and Ran Zhou and Haitao Gan}, doi = {10.1016/j.engappai.2025.110342}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110342}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive feature selection with flexible mapping for diagnosis and prediction of parkinson's disease}, volume = {146}, year = {2025}, }
• (2025). One-stage keypoint detection network for end-to-end cow body measurement. EAAI, 146, 110333. (www)

  @article{ , abstract = {Body size measurement plays a crucial role in dairy cow breed selection and milk production. Employing intelligent systems for periodic assessments of body size empowers farmers to gauge the nutritional status of cows. The study introduces an end-to-end intelligent approach for the automatic measurement of cow body size via keypoint detection. Introducing Cow Keypoint-Net (CowK-Net), a one-stage dairy cow keypoint detection network. To improve the interaction of cow features at the channel level, we created the Keypoint Refine Machine (KPRM), designed to balance channel and spatial information through separate pathways effectively. Moreover, we devised an efficient hybrid encoder to interact the information across different scales. This encoder combines Convolutional Neural Network (CNN) based cross-scale fusion with Transformer-based intra-scale interaction, thereby optimizing the keypoint processing and integration. Customizing the loss function to the specific characteristics of the cow dataset ensures effective supervision of the keypoint prediction process. Additionally, we transformed the pixel coordinates of keypoints into three dimensions (3D) space, enabling automated measurement of body size. Field testing on a production farm revealed CowK-Net's accuracy, achieving an impressive 92.8%, surpassing existing keypoint detection methods. Notably, the hybrid encoder matched the accuracy of a Transformer-based encoder while reducing the number of parameters by 18%. Compared to manual measurements, our method demonstrated mean relative errors of 2.8%, 6.7%, 4.1%, and 4.4% for oblique body length, body height, hip height, and chest depth, respectively. The CowK-Net demonstrates its efficacy in measuring cow body size, laying solid foundation for the development of body measurement devices.}, archive = {J_EAAI}, author = {Guangyuan Yang and Yongliang Qiao and Hongxing Deng and Javen Qinfeng Shi and Huaibo Song}, doi = {10.1016/j.engappai.2025.110333}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110333}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {One-stage keypoint detection network for end-to-end cow body measurement}, volume = {146}, year = {2025}, }
• (2025). A transformer-based self-supervised learning model for fault diagnosis of air-conditioning systems with limited labeled data. EAAI, 146, 110331. (www)

  @article{ , abstract = {Despite the great successes of supervised learning-based fault diagnosis techniques for heating, ventilation and air-conditioning (HVAC) systems, their applications are severely limited due to insufficient labeled data accompanied with massive unlabeled data. To address this drawback, a Transformer-based self-supervised representation learning model (TSSRL) is proposed in this study for HVAC fault diagnosis with limited labeled data. Specifically, a customized Transformer model is developed as the feature encoder by embedding a context-attention module on the self-attention module, which enables TSSRL to mine the contextual representations among input data. In addition, a joint data augmentation strategy is designed to improve the diversity of inputs, promoting the pretext tasks to learn more extensive representations from unlabeled data. Meanwhile, two cooperative pretext tasks, namely contrastive similarity matching and data reconstruction, are formulated to extract discriminative representations from unlabeled data. The diagnosis-beneficial representations learned from unlabeled data are used for downstream classification modeling tasks with limited labeled data. Experiments on two benchmark HVAC fault datasets demonstrate the superiority of the proposed TSSRL model over other state-of-the-art HVAC fault diagnosis methods.}, archive = {J_EAAI}, author = {Mei Hua and Ke Yan and Xin Li}, doi = {10.1016/j.engappai.2025.110331}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110331}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A transformer-based self-supervised learning model for fault diagnosis of air-conditioning systems with limited labeled data}, volume = {146}, year = {2025}, }
• (2025). Physics informed neural network with fourier feature for natural convection problems. EAAI, 146, 110327. (www)

  @article{ , abstract = {In this paper, we investigate the application of deep neural networks to solve the Navier-Stokes and heat equations within the framework of modeling the natural convection phenomenon. The main objective is to reconstruct the velocity and temperature fields in a differentially heated rectangular cavity while adhering to the imposed boundary conditions. Two main architectures are compared: the Fully Connected Neural Network and the Fourier Features Neural Network. A hyper-parameter tuning process was carried out to optimize the network performances. This tuning led to a final architecture composed of 6 layers, each with 128 neurons, and 64 Fourier frequencies, with the Mish activation function selected after testing several alternatives. Both architectures were trained on four cases, where the Rayleigh number ranges from 10 4 to 10 7 , with quasi-randomly sampled points. The network predictions were then compared to the results obtained from numerical simulations of the Navier-Stokes and heat equations. The results show that for low Rayleigh numbers (10 4 and 10 5 ), both architectures converge quickly, producing smooth profiles dominated by low frequencies. However, for higher Rayleigh numbers (10 6 ), the Fourier Features Neural Network outperforms the Fully Connected Neural Network by better capturing the complex and localized variations, thanks to its explicit integration of periodic components, which makes it particularly well-suited for multi-scale problems. This study highlights the potential of deep neural networks to solve partial differential equations in complex configurations, offering a promising alternative to traditional methods. It also emphasizes the importance of choosing an architecture that fits the specific characteristics of the problem at hand, especially in cases where the solutions exhibit multi-scale variations or high frequencies.}, archive = {J_EAAI}, author = {Younes Bounnah and Mustapha Kamel Mihoubi and Salah Larbi}, doi = {10.1016/j.engappai.2025.110327}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110327}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics informed neural network with fourier feature for natural convection problems}, volume = {146}, year = {2025}, }
• (2025). Integrating case learning and consensus reaching in multi-criteria group sorting under hybrid assessment information. EAAI, 146, 110323. (www)

  @article{ , abstract = {The current multi-criteria group sorting methods face challenges in handling multi-criteria group sorting issues with hybrid assessment information, where the hybrid assessment information is provided by multiple decision-makers in evaluating alternatives. The main challenges lie in: (1) extracting preference information regarding sorting parameters from the hybrid assessment information provided by decision-makers, and (2) reconciling conflicts among different types of hybrid assessment information provided by decision-makers in the multi-criteria group sorting context. To address these challenges, we introduce a cohesive methodology that integrates case learning with consensus reaching for multi-criteria group sorting employing hybrid assessment information. Initially, a 0–1 mixed integer programming model is formulated to identify the group consistency case set with the utmost significance. Within this framework, the consensus degree of decision-makers regarding alternatives serves as a measure of the cases' importance. Subsequently, the group consistency case set is employed to construct a bi-objective programming model, which ascertains the sorting parameters and consensus preferences through the integration of the group consensus level and the deviation degree from the class threshold. Using these parameters and preferences, the consensual sorting results of alternatives are derived. Ultimately, the practicality of the proposed methodology is demonstrated through a case study focusing on the sorting of internet public opinion emergencies. Simulation and comparative analyses underscore the practical applicability and superior efficacy of this approach.}, archive = {J_EAAI}, author = {Shitao Zhang and Shiji Zhang and Fengli Zhu and Muhammet Deveci and Xiaodi Liu}, doi = {10.1016/j.engappai.2025.110323}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110323}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating case learning and consensus reaching in multi-criteria group sorting under hybrid assessment information}, volume = {146}, year = {2025}, }
• (2025). Enhanced forecasting method for realized volatility of energy futures prices: A secondary decomposition-based deep learning model. EAAI, 146, 110321. (www)

  @article{ , abstract = {To accurately measure and predict the volatility of energy futures prices, this study employs 5-min high-frequency price data to construct the realized volatility (RV) of crude oil futures and natural gas futures. On this basis, a deep learning model based on secondary decomposition with multi-feature input and single-target output is proposed for multi-step ahead forecasting of RV of energy futures prices. First, the original RV sequence of energy futures is initially decomposed using successive variational mode decomposition (SVMD) to obtain several subsequences. Then, a secondary decomposition of the residual series that cannot be handled by SVMD is performed using improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) to further extract finer features and modes. Afterward, by using multi-feature input and single-feature output, all the subsequences obtained from the two decompositions are input into the deep learning model consisting of spatial transformer network (STN) and convolutional long short-term memory (convLSTM) to obtain the prediction results of each subsequence individually. Finally, the prediction results of all subsequences are ensembled to obtain the final multi-step ahead prediction results of energy futures RV. By introducing the multi-horizon model confidence set (multi-horizon MCS) test, it is statistically confirmed that the proposed multi-feature input and single-target output SVMD-ICEEMDAN-STN-convLSTM model possesses the most superior joint prediction performance in multi-step ahead forecasting of the RV of energy futures prices. Overall, the application of this model holds substantial value and significance, promising to significantly contribute to the sound operation and development of the energy futures market.}, archive = {J_EAAI}, author = {Hao Gong and Haiyang Xing and Qianwen Wang}, doi = {10.1016/j.engappai.2025.110321}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110321}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced forecasting method for realized volatility of energy futures prices: A secondary decomposition-based deep learning model}, volume = {146}, year = {2025}, }
• (2025). Graph-aware pre-trained language model for political sentiment analysis in filipino social media. EAAI, 146, 110317. (www)

  @article{ , abstract = {Elections are emotionally and sentimentally charged events that offer unique opportunities for analysis of sentiments not typically observed during non-election periods. Unlike recurring phenomena, elections are inherently singular events, with each election shaped by distinct political, social, and cultural contexts. In the digital age, social media has become a direct channel for politicians and political parties to engage with voters, making it a critical platform for sentiment analysis. However, challenges such as imbalanced datasets, the prevalence of noisy non-text elements (e.g., emojis, hashtags, user mentions), and the need for effective integration of graph-based learning remain significant hurdles in sentiment prediction. To address these challenges, we constructed an imbalanced dataset of 8035 manually annotated tweets and approximately 516,000 weakly labeled Filipino tweets related to the 2022 Philippine National Election. Leveraging these datasets, we designed a Bidirectional Encoder Representations from Transformers (BERT) and Graph Convolution Network (GCN) model, which uniquely incorporates emojis, hashtags, and user mentions as features to enhance semantic understanding. Differing from the prior literature that focused solely on textual data or discarded non-textual elements, our model integrates these features to achieve a robust performance that outperforms baseline models with a macro-recall score of 64.73% and a macro F 1 -score of 68.72% on the imbalanced dataset. Additionally, we introduce a topic modeling framework that combines BERT embeddings with Latent Dirichlet Allocation (LDA) and Log-Likelihood Ratio (LLR) to yield more distinct topic clusters for deeper sentiment analysis. Our work therefore contributes two novel datasets in Filipino as well as methodologies that bridge sentiment prediction and analysis, and in so doing, provides valuable resources for future research.}, archive = {J_EAAI}, author = {Jean Aristide Aquino and Di Jie Liew and Yung-Chun Chang}, doi = {10.1016/j.engappai.2025.110317}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110317}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph-aware pre-trained language model for political sentiment analysis in filipino social media}, volume = {146}, year = {2025}, }
• (2025). A single-valued neutrosophic affinity propagation approach for engineering geological zoning of open-pit mine slopes. EAAI, 146, 110316. (www)

  @article{ , abstract = {Engineering geological zoning is an important foundation for assessing the stability of open-pit mine (OPM) slopes. Because of the complexity and uncertainty of geological conditions, this work brings great challenges to mining engineers and researchers. This paper presents a single-valued neutrosophic affinity propagation (SVNS-AP) approach for engineering geological zoning of OPM slopes. The approach utilizes the concept of a single-valued neutrosophic set (SVNS) to express the inconsistent and indeterminate information present in the influencing factors of engineering geological conditions through truth, indeterminate and falsity membership functions. Then, the similarity measure of SVNS is integrated into the affinity propagation (AP) algorithm to calculate the degree of similarity between the data points. Finally, the modified silhouette index is used to evaluate the clustering results and decide the optimal number of clusters. The practical application results of the engineering geological zoning of Lanping lead-zinc OPM demonstrate that the SVNS-AP method is an effective way for engineering geological zoning of OPM slopes in the uncertain environment. Clustering results based on datasets in the literature and the UC Irvine Machine Learning Repository show that the proposed method can be used as a general clustering algorithm.}, archive = {J_EAAI}, author = {Jibo Qin and Xiaoming Sun and Shigui Du and Jun Ye}, doi = {10.1016/j.engappai.2025.110316}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110316}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A single-valued neutrosophic affinity propagation approach for engineering geological zoning of open-pit mine slopes}, volume = {146}, year = {2025}, }
• (2025). Short-term offshore wind speed forecasting approach based on multi-stage decomposition and deep residual network with self-attention. EAAI, 146, 110313. (www)

  @article{ , abstract = {Wind energy is one of the widely used renewable energy systems. Wind speed forecasting is used to produce of wind energy and to ensure the sustainability of the power system. However, offshore wind speed forecasting is a challenging task with complex variables and highly nonlinear temporal dynamics of the ocean. This paper proposes a hybrid and robust offshore wind speed forecasting approach based on multi-stage decomposition, deep convolutional neural network (CNN), and extreme learning machine (ELM). Unlike conventional preprocessing for forecasting of renewable energy problems, the proposed approach combines two efficient decomposition methods as complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and ensemble empirical mode decomposition (EEMD). This method can decompose high-frequency and low-frequency components of the wind speed. While high-frequency components are decomposed with the EEMD, low-frequency components are directly sent to the ELM model. The obtained mode functions from the EEMD are then fed to the designed network for forecasting. The CNN model is constructed with the deep residual network and self-attention (SA) mechanism to improve the network performance. In the comparative evaluations, while other approaches give lower forecasting performance between 0.8233 and 2.1885 for the root mean square error (RMSE), the proposed method presents the lowest RMSE value as 0.5400. The experimental results show that the proposed method exhibits more accurate and robust forecasting performance compared with other model combinations and deep learning models.}, archive = {J_EAAI}, author = {Hakan Acikgoz and Deniz Korkmaz}, doi = {10.1016/j.engappai.2025.110313}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110313}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Short-term offshore wind speed forecasting approach based on multi-stage decomposition and deep residual network with self-attention}, volume = {146}, year = {2025}, }
• (2025). Incorporating prior knowledge of collision risk into deep learning networks for ship trajectory prediction in the maritime internet of things industry. EAAI, 146, 110311. (www)

  @article{ , abstract = {Artificial intelligence (AI) has played a key role in advancing autonomous navigation for unmanned ships, where ship trajectory prediction is crucial for ensuring maritime safety. As the shipping industry grows and the number of ships increases, especially with autonomous ships operating in complex environments, collision risks have become a major concern. Accurate trajectory prediction, supported by advanced AI techniques, is crucial for the safe operation of these ships. While current models predict ship trajectories with high precision using Automatic Identification System (AIS) data, they often fail to incorporate prior knowledge of collision risks and struggle to model ship interactions that could lead to collisions. To overcome these limitations, the DGCN-Transformer (Dynamic Graph Convolution Network-Transformer) model is proposed. This model enhances the accuracy and reliability of ship trajectory predictions by incorporating collision risk modeling into the prediction framework. It uses the Quaternion Ship Domain (QSD) to model potential collision scenarios, integrating an advanced understanding of ships' spatial and kinematic properties. The model integrates QSD-based prior knowledge into an advanced Graph Convolutional Network (GCN) for spatial modeling, while the Transformer component captures and analyzes temporal features, overcoming the limitations of traditional Long Short-Term Memory (LSTM) networks. Experiments with AIS data from Tianjin, Caofeidian, and Chengshanjiao ports demonstrate that the DGCN-Transformer model outperforms state-of-the-art models, significantly improving trajectory prediction accuracy. Specifically, at Tianjin Port, the DGCN-Transformer model reduces Final Displacement Error (FDE) by 36.1%, Maximum Displacement Error (MDE) by 15.4%, and Average Displacement Error (ADE) by 50% compared to the best baseline model, highlighting the model's effectiveness in enhancing the safety of autonomous ship navigation.}, archive = {J_EAAI}, author = {Yu Zhang and Ping Tu and Zhiyuan Zhao and Xuan-Yan Chen}, doi = {10.1016/j.engappai.2025.110311}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110311}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Incorporating prior knowledge of collision risk into deep learning networks for ship trajectory prediction in the maritime internet of things industry}, volume = {146}, year = {2025}, }
• (2025). A recurrent neural network for computational physiology and its application to intracoronary image-derived fractional flow reserve. EAAI, 146, 110309. (www)

  @article{ , abstract = {Background Traditional data-driven neural networks struggle to extract physical properties from complex visual data, particularly in medical imaging. An example is calculating fractional flow reserve (FFR) from coronary artery images, which involves the discovery of physical knowledge from visual data. Objective This study proposes a novel deep learning framework to calculates the distribution of FFR along coronary vessels from intracoronary images, integrating both geometric features and blood flow dynamics. Problem Existing FFR estimation methods rely on computationally intensive three-dimensional reconstructions, which may lack accuracy in estimating FFR distributions across vascular segments. Methodology A multi-input, multi-output deep bidirectional recurrent neural network is introduced, directly estimating FFR values from intracoronary images by combining vessel geometry with physical knowledge of blood flow, without the need for three-dimensional reconstructions. Findings The network model demonstrates strong performance on synthetic and clinical datasets, with correlations of 0.91 and 0.85 to computational fluid dynamics simulations for synthetic and clinical data, respectively. It also achieves area under the receiver operating characteristic curve values of 0.97 and 0.94 for ischemic lesion detection. Conclusion This framework enhances FFR estimation and ischemic lesion identification accuracy while reducing computational costs compared to three-dimensional reconstruction methods. Recommendations The model shows promise for clinical applications in coronary artery diseases and could be integrated into clinical decision support systems to improve diagnostic workflows.}, archive = {J_EAAI}, author = {Sun Zheng and Jiao Wenbin and Wang Shuyan}, doi = {10.1016/j.engappai.2025.110309}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110309}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A recurrent neural network for computational physiology and its application to intracoronary image-derived fractional flow reserve}, volume = {146}, year = {2025}, }
• (2025). Understanding the impacts of negative advanced driving assistance system warnings on hazardous materials truck drivers’ responses using interpretable machine learning. EAAI, 146, 110308. (www)

  @article{ , abstract = {In recent years, Artificial Intelligence (AI) has significantly enhanced road safety, with Explainable Artificial Intelligence (XAI) providing essential transparency and trust. Our research utilizes AI to improve Advanced Driving Assistance Systems (ADAS) by investigating the gap in Forward Collision Warning (FCW): the impact of previous negative warnings (false and nuisance warnings) on drivers’ response times to subsequent accurate FCWs. By integrating XAI methods, we offer insights into the factors affecting driver behavior and system trust. Utilizing extensive dataset that encompasses various driving scenarios and driver behaviors, we constructed a gradient-boosting machine model to forecast driver response times. To explain the underlying mechanics of the model, the Shapley Additive Explanations (SHAP) framework was employed, enabling a comprehensive interpretation of feature importance and inter-feature interactions. Key findings reveal that increased speeds heighten driver responsiveness due to amplified alertness, whereas slower speeds lead to delayed reactions. The influence of previous negative warnings, significantly extends response times to accurate warnings. Additionally, older drivers require longer response times. The relationship between the driving period and previous warning judgment profoundly affects subsequent driver responsiveness, indicating trust dynamics with FCW systems. By using interpretable machine learning, we provide insights into ADAS functionality, suggesting pathways for FCW responsiveness and contributing to the field of XAI applications. In the validation experiment, our approach improved driver response times, reducing the average time from 2.1 s to 1.6 s. The proportion of ignored warnings decreased from 12% to 6%, and the driver acceptance rate increased from 59% to 71%.}, archive = {J_EAAI}, author = {Yichang Shao and Yueru Xu and Zhirui Ye and Yuhan Zhang and Weijie Chen and Nirajan Shiwakoti and Xiaomeng Shi}, doi = {10.1016/j.engappai.2025.110308}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110308}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Understanding the impacts of negative advanced driving assistance system warnings on hazardous materials truck drivers’ responses using interpretable machine learning}, volume = {146}, year = {2025}, }
• (2025). SPARDA: Sparsity-constrained dimensional analysis via convex relaxation for parameter reduction in high-dimensional engineering systems. EAAI, 146, 110307. (www)

  @article{ , abstract = {Effective analysis of high-dimensional systems with intricate variable interactions is crucial for accurate modeling and engineering applications. Previous methods using sparsity techniques or dimensional analysis separately often face limitations when handling complex, large-scale systems. This study introduces a sparsity-constrained dimensional analysis framework that integrates the classical Buckingham Pi theorem with sparse optimization techniques, enabling precise nondimensionalization. The framework, formulated as a convex optimization problem, addresses computational challenges associated with sparsity in high-dimensional spaces. Rigorously tested across various datasets, including the Fanning friction factor for rough pipe flow, an international standards-based dataset of physical quantities and units, and experimental data from flow boiling studies, this method successfully identified critical dimensionless groups that encapsulate core system dynamics. This approach not only offers a more compact and interpretable representation than conventional methods but also retains more characteristics of function variability. It proves particularly effective in systems governed by high-dimensional interactions, demonstrating a lower failure rate and mean relative error compared to an algorithm for comparison. The methodology is applicable to the modeling and analysis of complex engineering physical systems such as nuclear power, wind tunnel design, and marine engineering, as well as in designing scaled verification experiments.}, archive = {J_EAAI}, author = {Kuang Yang and Qiang Li and Zhenghui Hou and Haifan Liao and Chaofan Yang and Haijun Wang}, doi = {10.1016/j.engappai.2025.110307}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110307}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SPARDA: Sparsity-constrained dimensional analysis via convex relaxation for parameter reduction in high-dimensional engineering systems}, volume = {146}, year = {2025}, }
• (2025). High-order graph convolutional networks for circular ribonucleic acid and disease association prediction incorporating multiple biological relationships. EAAI, 146, 110303. (www)

  @article{ , abstract = {Background The search for circular Ribonucleic Acid (circRNA) associated with complex diseases holds considerable importance for disease diagnosis, treatment and research, helping to improve the early recognition and therapeutic efficacy of diseases, deepen the understanding of disease mechanisms, and provide guidance for new drug development. Methods This study presents an innovative high-order graph convolutional neural network, which leverages Gaussian kernels to compute the second-order proximity between nodes, thereby capturing long-range dependencies more effectively. Based on the topological structure of nodes in the graph, the model derives high-order embeddings, which not only enhance the preservation of the global network structure but also overcome the limitations of traditional methods that focus solely on local neighborhoods. Furthermore, by integrating this model with heterogeneous networks composed of multiple biological relationships, we successfully implement accurate predictions of circRNA-disease associations. Results This study achieved an area under the curve (AUC) of 0.9491 and an accuracy of 0.9920 on the constructed benchmark dataset, significantly outperforming existing methods in predictive performance, while most of the candidate circRNAs screened in the case studies of breast neoplasms and glioma have been confirmed in the literature. Conclusions This method provides a new perspective for integrating heterogeneous biological data in the study of complex disease-related circRNAs, and will advance further research and practical applications in this field.}, archive = {J_EAAI}, author = {Hao Liu and Chen Chen and Xiaoyi Lv and Jin Gu and Enguang Zuo and Chenjie Chang and Ying Su and Cheng Chen}, doi = {10.1016/j.engappai.2025.110303}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110303}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {High-order graph convolutional networks for circular ribonucleic acid and disease association prediction incorporating multiple biological relationships}, volume = {146}, year = {2025}, }
• (2025). Unsupervised motion-based anomaly detection with graph attention networks for industrial robots labeling. EAAI, 146, 110298. (www)

  @article{ , abstract = {As automated labeling on products in intelligent manufacturing grows in importance, detecting anomalies in the end-effectors used for industrial robots labeling is essential for maintaining production line stability and efficiency. Considering the distinct characteristics of specific movements in the labeling process, different motions, such as moving, labeling and rolling, contribute different effects to end-effector abnormalities. It's challenging to distinguish between normal and anomalies, instead of treating all motions as a homogeneous whole. Also, real-world industrial scenarios often lack the sufficient data on abnormal states, and resource constraints limit computation efficiency. In view of this, this paper aims to develop a task-specific anomaly detection solution tailored to the distinct motions of industrial robots labeling. To achieve this goal, an unsupervised, motion-based anomaly detection framework is proposed. The raw sensor signals from each motion are segmented and a group of encoder networks are employed to extract latent representations for each motion. Then, these motions are modeled as nodes in a graph, where a feature fusion module based on a Graph Attention Network (GAT) captures the interrelationships between them. A memory-augmented reconstruction module with multi-scale skip connections enhances the model's ability to detect anomalies. Finally, an anomaly detection module identifies abnormal states of the end-effector. Experimental validations are conducted on a dataset from real-world steel coil labeling task. The results show that the proposed framework can achieve an average performance of 98.24% with an inference time of 15 ms, also demonstrating the effectiveness of its structural design and key modules.}, archive = {J_EAAI}, author = {Jinrui Han and Zhen Chen and Di Zhou and Bing Hu and Tangbin Xia and Ershun Pan}, doi = {10.1016/j.engappai.2025.110298}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110298}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised motion-based anomaly detection with graph attention networks for industrial robots labeling}, volume = {146}, year = {2025}, }
• (2025). Innovative integration of machine learning and colorimetry for precise potential of hydrogen monitoring in printed hydrogel sensors. EAAI, 146, 110293. (www)

  @article{ , abstract = {Proper potential of hydrogen (pH) monitoring finds wide applications in environmental monitoring, clinical diagnostics, and a variety of industrial processes. However, traditional pH sensors normally present several challenges related to adaptability, portability, and environmental compatibility. In addition, the recently developed hydrogel-based sensors have manifested several advantages due to the flexibility and biocompatibility of the material in a wide variety of applications. While much advancement has been made in integration techniques, further advances need improvement in precision and reliability. The present work describes a novel methodology of pH sensing through integration of hydrogel-based sensors with machine learning algorithms. pH-sensitive dye-impregnated hydrogel sensors have been fabricated using three-Dimensional (3D) printing technology, whereby colorimetric data analysis is combined with five machine learning models, namely Decision Trees, eXtreme Gradient Boosting, K-Nearest Neighbours, Random Forests, and Neural Networks, in the classification of pH based on Red, Green, Blue (RGB) data. The sensor designed can detect pH between 4 and 10 pH with high speed, stability, and reversibility. With precision, recall, and F1-scores all above 99%, this shows how efficient the classification approach is based on RGB and gives weight to the potential of the developed sensors for real-time applications in monitoring and diagnostics, hence making a big contribution to the evolution of pH sensing and paving the way for smarter, more adaptable sensor solutions.}, archive = {J_EAAI}, author = {Abdelrahman Sakr and Ahmed R. El shamy and Haider Butt}, doi = {10.1016/j.engappai.2025.110293}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110293}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Innovative integration of machine learning and colorimetry for precise potential of hydrogen monitoring in printed hydrogel sensors}, volume = {146}, year = {2025}, }
• (2025). Context-aware inverse reinforcement learning for modeling individuals’ daily activity schedules. EAAI, 146, 110279. (www)

  @article{ , abstract = {Understanding individual and crowd dynamics in urban environments is critical for numerous applications, such as urban planning, traffic forecasting, and location-based services. Therefore, accurately modeling individuals' daily activity schedules is essential. Traditional methods, like utility-based and rule-based approaches, rely on expert knowledge and presumed model structures. While machine learning methods offer flexibility, they often ignore explicit behavioral mechanisms, particularly comprehensive discussion and integration of context related to individuals' daily travel. To address these, we propose a framework that integrates travel context with deep Inverse Reinforcement Learning (IRL), learning temporal patterns from sociodemographics, start time and duration of the current activity, travel modes, and land use. Specifically, individuals' activity schedules are initially formulated as a Markov Decision Process to simulate travelers’ sequential decision-making processes, laying the groundwork for the IRL framework; Then, a context-aware IRL method is proposed to model individuals' travel decision-making from observed temporal trajectories. Finally, we validate the proposed model by demonstrating its superior performance over discrete choice model and several well-known imitation learning benchmarks in tasks such as policy performance comparison, reward recovery, model generalizability, and computational efficiency using travel behavior datasets. This approach provides meaningful behavioral insights and paves the way for Artificial Intelligence-driven activity schedulers modeling.}, archive = {J_EAAI}, author = {Dongjie Liu and Dawei Li and Kun Gao and Yuchen Song and Zijie Zhou}, doi = {10.1016/j.engappai.2025.110279}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110279}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Context-aware inverse reinforcement learning for modeling individuals’ daily activity schedules}, volume = {146}, year = {2025}, }
• (2025). Learning multi-color curve for image harmonization. EAAI, 146, 110277. (www)

  @article{ , abstract = {Due to the varying shooting conditions, composite images often lack realism between the foreground and the back ground. As an important and challenging visual task, image harmonization can effectively improve visual effect of composite images. Currently, image harmonization methods have achieved satisfied performance on public dataset. However, in some challenging examples with substantial color disparities between the foreground and the background, existing methods get poor results. To solve this problem, we propose a Multi-color Curve Net that processes images through multiple color spaces to capture richer color information. Our Multi-color Curve Net performs multi-stage curve learning in different color spaces with the encoder composed of modified Transformer blocks. Simultaneously, we introduce a Multi-color Integration Module to effectively fuse the information extracted from different color spaces and further improve the results by a lightweight Fine-grained Optimization Module. The Multi-color Curve Net gains high performance while maintaining a small parameter scale. Experiments on benchmark demonstrate that the Multi-color Curve Net outperforms state-of-the-art methods in terms of peak signal to-noise ratio (PSNR), structural similarity (SSIM) and foreground mean squared error (fMSE) with fewer parameters. The code for our method is available at https://github.com/gmrj2024/MC2Net .}, archive = {J_EAAI}, author = {Jingrong Yuan and Hao Wu and Lidong Xie and Lei Zhang and Jichen Xing}, doi = {10.1016/j.engappai.2025.110277}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110277}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Learning multi-color curve for image harmonization}, volume = {146}, year = {2025}, }
• (2025). Spatial-temporal relation inference transformer combined with dynamic relationship and static causality for batch process modeling and the application of erythromycin fermentation. EAAI, 146, 110271. (www)

  @article{ , abstract = {In practical production, batch processes are typically characterized by high nonlinearity, dynamics, and coupling. Existing data-driven modeling methods often focus solely on the temporal relationships within variables, overlooking the spatial relationships among them. To address these challenges, this paper proposes a grey-box model named dynamic and static graph Transformer (DSGraphformer), which makes three significant contributions: (1) introducing a dynamic relationships extraction algorithm based on self-attention mechanisms, which transforms the relationships between variables into multiple graphs and utilizes graph attention networks to form dynamic graphs. (2) combining mechanistic knowledge and causal discovery algorithms to construct a static knowledge graph, proposing a prior knowledge fusion modeling scheme based on mutual attention mechanisms; and (3) establishing a sequential encoding framework for temporal and spatial relationships, parallelizing dynamic and static graph. Experiments on a real-world erythromycin fermentation process dataset demonstrate that DSGraphformer significantly outperforms existing methods in various prediction tasks.}, archive = {J_EAAI}, author = {Yifei Sun and Xuefeng Yan}, doi = {10.1016/j.engappai.2025.110271}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110271}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatial-temporal relation inference transformer combined with dynamic relationship and static causality for batch process modeling and the application of erythromycin fermentation}, volume = {146}, year = {2025}, }
• (2025). MIRROR: Multi-scale iterative refinement for robust chinese text recognition. EAAI, 146, 110270. (www)

  @article{ , abstract = {Text recognition has become a key area of research due to its wide applications in various fields. As an important branch of computer vision, Chinese text recognition has gained increasing research and practical value. However, the existing Chinese text recognition methods are still limited. This paper proposes an innovative Chinese text recognition method, M ulti-Scale I terative R efinement for Ro bust Chinese Text R ecognition (MIRROR). The model significantly improves the recognition accuracy of Chinese text through advanced algorithms and structural design. The MIRROR model consists of two core components: a feature extractor and a Next-Character Decoder. Specifically, this paper proposes a Spatial Local Self-Attention Module to enhance the model’s ability to model long-distance dependencies in complex character sequences, addressing the problem of complex distributions in medium-to-long distance Chinese character sequences. The Character Refinement Module effectively captures multi-scale information, handles stroke feature differences, and resolves inter-class similarity issues. By combining multi-scale feature extraction with iterative optimization for feature refinement, the model identifies common features across different styles of the same character, solves the intra-class variation problem, and improves model robustness. In addition, this paper introduces a Three-Dimensional Weight Attention Module to refine the granularity of character features. Experiments show that MIRROR significantly outperforms baseline models on Chinese benchmark datasets. On scene datasets, performance improves by 3.08% (from 76.90% to 79.98%), on web datasets by 1.46% (from 70.43% to 71.89%), on document datasets by 0.38% (from 98.72% to 99.10%), and on handwriting datasets by 9.29% (from 50.26% to 59.55%).}, archive = {J_EAAI}, author = {Hengnian Qi and Qiuyi Xin and Jiabin Ye and Hao Yang and Kai Zhang and Chu Zhang and Qing Lang}, doi = {10.1016/j.engappai.2025.110270}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110270}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MIRROR: Multi-scale iterative refinement for robust chinese text recognition}, volume = {146}, year = {2025}, }
• (2025). Adaptive prompt guided unified image restoration with latent diffusion model. EAAI, 146, 110267. (www)

  @article{ , abstract = {Recently, Diffusion Models (DMs) have witnessed the remarkable success in image restoration tasks. However, DMs are not flexible and adaptive in dealing with uncertain multiple forms of image degradation (e.g., noise, blur and so on) due to the lack of degradation prior, resulting in undesirable boundary artifacts. In addition, DMs require a large number of inference iterations to restore clean image, which consumes massive computational resources. To address the forementioned limitations, we propose an adaptive unified two-stage restoration method based on latent diffusion model, termed APDiff that can effectively and adaptively handle real-world images with various degradation types. Specifically, in Stage I, we pre-train a Degradation Adaptive Prompt Learning Network (DAPLNet-S1) to obtain degradation prompt by exploring differences between low quality (LQ) and ground truth (GT) images adaptively. Then, we encode it into the latent space as key discriminant information for different degraded images. In Stage II, we propose a latent diffusion model to directly estimate a degradation prompt similar in pre-train DAPLNet-S1 only using LQ images. Meanwhile, to restore different degradation images effectively, we design a Prompt Guided Fourier Transformer Restorer to integrate the extracted prompt, which enhances characterization ability of model for global frequency feature and local spatial information. Since the generated prompts are low-dimensional latent vector representations, this can significantly reduce computational complexity of diffusion model. Thus, during the inference process, our method takes only 0.09 s to restore an image of SPA+. Extensive experiments demonstrate that APDiff achieves state-of-the-art performance for multi-degradation tasks.}, archive = {J_EAAI}, author = {Xiang Lv and Mingwen Shao and Yecong Wan and Yuanjian Qiao and Changzhong Wang}, doi = {10.1016/j.engappai.2025.110267}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110267}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive prompt guided unified image restoration with latent diffusion model}, volume = {146}, year = {2025}, }
• (2025). Picking point localization method of table grape picking robot based on you only look once version 8 nano. EAAI, 146, 110266. (www)

  @article{ , abstract = {Automatic localization of the picking point for table grape is the key to achieving intelligent harvesting. Aiming at the problem, the You Only Look Once version 8 nano-Deformable Convolutional Networks-Wise Intersection over Union-Fourth Detection Layer (YOLO v8n-DWF) network was designed for the table grape picking robot to realize the detection of table grapes and the localization of picking points. The Deformable Convolutional Networks (DCN) model was used to enhance the robustness of table grape detection and improve the detection precision of grape stems. To improve the detection precision of table grape and stem and reduce the impact of low-quality data on model generalization ability, Wise-Intersection over Union version 3 ( WIou v3 ) was applied. In addition, aiming at the problem of difficultly identifying due to small target and low pixel of table grape stem, a fourth detection layer for small target detection was added to improve the recognition and detection ability of the network model for table grape stem. Further, a more accurate geometric localization method of picking points was proposed to achieve fast picking of table grapes. Finally, the results showed that the detection precision, recall, mean Average Precision 50 ( mAP 50 ) and mean Average Precision 50-95 ( mAP 50-95 ) of the YOLO v8n-DWF model was 97.9%, 95.3%, 97.6% and 85.4%, respectively. In addition, the success rate of the geometric method based on the results of identification for YOLO v8n-DWF was 88.24%, and the average picking success rate of table grapes in field experiments was 87.40%. It can fully meet the requirements of intelligent picking of table grapes.}, archive = {J_EAAI}, author = {Yanjun Zhu and Shunshun Sui and Wensheng Du and Xiang Li and Ping Liu}, doi = {10.1016/j.engappai.2025.110266}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110266}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Picking point localization method of table grape picking robot based on you only look once version 8 nano}, volume = {146}, year = {2025}, }
• (2025). An improved framework for breast ultrasound image segmentation with multiple branches depth perception and layer compression residual module. EAAI, 146, 110265. (www)

  @article{ , abstract = {Breast cancer is becoming a leading cause of death among women worldwide. Early detection is essential for improving survival rates and facilitating targeted medical treatments. Automated segmentation of breast tumors from ultrasound images is vital for this early diagnosis. To tackle challenges such as low contrast, unclear lesion boundaries, and class imbalance in breast ultrasound images, a multiple branches depth perception network was introduced, using a symmetric encoder and decoder architecture. In the initial feature extraction stage, the network’s encoder employs the multiple branches depth residual block to integrate feature information from multiple branches while employing dilated convolution to capture intricate contextual details, enhancing the characterization of complex features. Subsequently, in the feature recovery stage, the network utilizes dual path depth perception block to mitigate information loss in deep networks by leveraging dual path residual connections, extracting rich textural and structural features from breast ultrasound images. Furthermore, the layer compression residual module and attention refinement module were incorporated within the skip connections to strengthen the contextual relationships between the encoder and decoder, leading to improved segmentation of breast lesions. Extensive qualitative and quantitative evaluations on two challenging public datasets were conducted to assess the effectiveness and generalizability of the proposed approach. The experimental results demonstrate the reliability of the proposed method in clinical treatment, achieving segmentation mean intersection over union scores of 91.11% and 92.28% on these respective datasets.}, archive = {J_EAAI}, author = {Ke Cui and Qichuan Tian and Haoji Wang and Chuan Ma}, doi = {10.1016/j.engappai.2025.110265}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110265}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An improved framework for breast ultrasound image segmentation with multiple branches depth perception and layer compression residual module}, volume = {146}, year = {2025}, }
• (2025). Image–text sentiment analysis based on hierarchical interaction fusion and contrast learning enhanced. EAAI, 146, 110262. (www)

  @article{ , abstract = {As social media continues to evolve, an increasing number of individuals prefer to publish a variety of content that combines different forms of expression on platforms to convey their feelings. In recent times, the study of sentiment within these diverse media formats has gained significant traction. However, some studies have neglected the multilayered interactions between modalities and underutilized the relationships between data samples and between classes. To address this problem, this paper proposes an image–text sentiment analysis method based on layer interaction fusion and contrast learning enhanced. First, the method uses a multi-layer cross-modal interaction module to emphasize the complementarity between modalities. The correlation information between modalities is deeply mined. Then, feature integration is performed by the multimodal fusion module. In addition, the model introduces a comparative learning task to exploit the relationship between samples and between classes. The emotional features under different classes and the key emotional features of the samples are mined. Finally, sentiment classification is performed to realize image–text sentiment analysis. In order to assess the efficacy of the approach, this paper performs extensive experimental validation on a multimodal sentiment dataset. Compared with several baseline models, the model in this paper achieves certain improvements. For example, on the multi-view sentiment analysis single dataset(MVSA-Single dataset), the F1 score is improved by 1.90% compared to the baseline model.}, archive = {J_EAAI}, author = {Hongbing Wang and Qifei Du and Yan Xiang}, doi = {10.1016/j.engappai.2025.110262}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110262}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Image–text sentiment analysis based on hierarchical interaction fusion and contrast learning enhanced}, volume = {146}, year = {2025}, }
• (2025). Neighborhood search with heuristic-based feature selection for click-through rate prediction. EAAI, 146, 110261. (www)

  @article{ , abstract = {Click-through-rate (CTR) prediction is crucial in online advertising and recommender systems. Maximizing CTR has been a major focus, leading to the development of numerous models designed to capture implicit and explicit feature interactions. However, extracting both low-order and high-order interactions remains challenging, as irrelevant features can increase computational costs and reduce prediction accuracy. Feature performance may also vary across predictive models and fluctuate due to traffic changes, making efficient feature selection essential in live environments where both training and inference times are critical. Traditional filter-based feature selection methods often fail to dynamically identify the most impactful features. This paper introduces a greedy heuristic, called Neighborhood Search with Heuristic-based Feature Selection (NeSHFS), to enhance CTR prediction by iteratively refining the feature set. NeSHFS employs a grid-search-like strategy to identify and retain the most relevant features, effectively reducing dimensionality and computational costs. Comprehensive experiments on several public datasets validate this approach, demonstrating improved prediction performance and reduced training times.}, archive = {J_EAAI}, author = {Dogukan Aksu and Ismail Hakki Toroslu and Hasan Davulcu}, doi = {10.1016/j.engappai.2025.110261}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110261}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neighborhood search with heuristic-based feature selection for click-through rate prediction}, volume = {146}, year = {2025}, }
• (2025). A rolling bearing fault diagnosis framework under variable working conditions considers dynamic feature extraction. EAAI, 146, 110255. (www)

  @article{ , abstract = {As a key component in industrial machinery, rolling bearings usually operate at variable speeds. The features in the signal are dynamic due to speed changes, with complementarity and correlation embedded in the different features. However, the utilization of this complementarity and correlation in mining dynamic signal features has been neglected, leading to reduced accuracy of fault classification models and less adaptability to variable working conditions. To address this problem, a multi-scale asymmetric feature reproduction plots-shifted window transformer (MAFRP-ST) framework of rolling bearing fault diagnosis is proposed under variable speed conditions in this study. Specifically, the framework includes dynamic feature capture and dynamic feature learning modules. The dynamic feature capture module is designed to convert signals into multi-scale asymmetric feature reproduction plots (MAFRP) containing features in the time–frequency domain, allowing the deeper dynamic features to be better captured by exploiting the complementarities and correlation. In the dynamic feature learning module, a shifted window (Swin) transformer adapted to dynamic features at different scales is developed, calculating local attention according to the window size in each layer and incrementally increasing the receptive field layer by layer. Compared with recently proposed similar methods, the MAFRP-ST framework improves diagnosis accuracy by about 4.1% and 2.2% on average across two datasets, respectively, and better robustness to noise is demonstrated.}, archive = {J_EAAI}, author = {Wang Jia and Hui Shi and Zengshou Dong and Xiaoyi Zhang}, doi = {10.1016/j.engappai.2025.110255}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110255}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A rolling bearing fault diagnosis framework under variable working conditions considers dynamic feature extraction}, volume = {146}, year = {2025}, }
• (2025). A hybrid probabilistic battery health management approach for robust inspection drone operations. EAAI, 146, 110246. (www)

  @article{ , abstract = {Monitoring the health of remote critical infrastructure poses significant challenges due to limited accessibility and harsh operational environments. Inspection drones are ubiquitous assets that enhance the reliability of critical infrastructures through improved accessibility. However, due to the harsh operation environment, it is crucial to monitor their health to ensure successful inspection operations. The battery is a key component that determines the reliability of the inspection drones and, with an appropriate health management approach, contributes to reliable and robust inspections. This paper introduces a novel hybrid probabilistic approach for predicting the end-of-discharge (EOD) voltage of lithium polymer (Li-Po) batteries in inspection drones. The proposed approach integrates Monte Carlo (MC) dropout based Convolutional Neural Networks (CNN) with electrochemistry-based battery discharge model. This integration employs an error-correction configuration that combines electrochemistry-based EOD prediction with probabilistic error correction using CNN with MC dropout. The approach is designed to infer aleatoric and epistemic uncertainty, facilitating robust battery discharge predictions through uncertainty-aware predictions. The proposed approach is empirically evaluated using a dataset comprising EOD voltage measurements under varying load conditions. The dataset, obtained from real inspection drones during offshore wind turbine inspections, underscores the practical applicability of the proposed approach. Comparative analysis with various probabilistic methods, including Quantile Linear Regression, Quantile Regression Forest, and Quantile Gradient Boosting, demonstrates a 14.8% improvement in probabilistic accuracy compared to the best-performing method. Additionally, the estimation of different uncertainties enhances the diagnosis of battery health states, contributing to more reliable inspection operations and highlighting the practical value of the work.}, archive = {J_EAAI}, author = {Jokin Alcibar and Jose I. Aizpurua and Ekhi Zugasti and Oier Peñagarikano}, doi = {10.1016/j.engappai.2025.110246}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110246}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid probabilistic battery health management approach for robust inspection drone operations}, volume = {146}, year = {2025}, }
• (2025). Polarization-based camouflaged object detection with high-resolution adaptive fusion network. EAAI, 146, 110245. (www)

  @article{ , abstract = {In comparison to traditional object detection or segmentation tasks, Camouflaged Object Detection (COD) poses greater challenges, as humans are often perplexed or deceived by the inherent similarities between foreground objects and their background surroundings. Polarization information serves as a valuable asset for discerning the attributes of objects with varied characteristics and surface texture. Taking inspiration from the polarization vision systems observed in animals, this study presents the High-Resolution Intensity & Polarization Fusion (HIPF) Net, a high-efficiency cross-modal fusion network that leverages trichromatic intensity and linear orthogonal polarization cues to produce a scene representation that is rich in texture and edge details. Specifically, the Early Adaptive Stokes Fusion (EASF) module maximizes the utilization of information from linear orthogonal polarization images. Subsequently, the Mix-Attention Feature Interaction Module (MAI) is introduced to facilitate complementary interaction among low-level features. Additionally, the Attentional Receptive Field Block (ARFB) enables the model to uncover concealed cues effectively, capturing objects of various sizes. Finally, the Weighted Cross-Level Decoder(WCFD) is designed to dynamically fuse and assign weights to cross-level contextual information for robust detection. Training and extensive validation of our model are performed on the polarization-based dataset as well as non-polarization-based datasets, with experimental results demonstrating that HIPFNet consistently outperforms state-of-the-art methods. Source codes are available at https://github.com/CVhfut/HIPFNet .}, archive = {J_EAAI}, author = {Xin Wang and Junfeng Xu and Jiajia Ding}, doi = {10.1016/j.engappai.2025.110245}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110245}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Polarization-based camouflaged object detection with high-resolution adaptive fusion network}, volume = {146}, year = {2025}, }
• (2025). Paying more attention to local contrast: Improving infrared small target detection performance via prior knowledge. EAAI, 146, 110244. (www)

  @article{ , abstract = {The data-driven methods for InfraRed Small Target Detection (IRSTD) have achieved promising results. However, these methods typically incorporate modules with high computational complexity, which enhance performance at the expense of computational efficiency. Utilizing human expert knowledge to assist data-driven methods in better learning with less costs is worthy of exploration. To effectively guide the model to focus on targets’ spatial features, this paper proposes the Local Contrast Attention Enhanced infrared small target detection Network (LCAE-Net), combining prior knowledge with data-driven deep learning methods. LCAE-Net is a U-shaped neural network model which consists of two developed modules: a Local Contrast Enhancement (LCE) module and a Channel Attention Enhancement (CAE) module. The LCE module takes advantage of prior knowledge, leveraging handcrafted convolution operators to acquire Local Contrast Attention (LCA), which could realize background suppression while enhancing the potential target region, thus guiding the neural network to pay more attention to potential infrared small targets’ location information. To effectively utilize the response information throughout the downsampling progresses, the CAE module is proposed to achieve the information fusion among feature maps’ different channels. Experimental results indicate that our LCAE-Net outperforms comparison methods on the three public datasets, and its detection speed could reach up to 70 Frames Per Second (FPS). Meanwhile, our model has a parameter count and Floating-Point Operations (FLOPs) of 1.945 Million (M) and 4.862 Giga (G) respectively, which is suitable for deployment on edge devices. Our code will be available at https://github.com/boa2004plaust/LCAENet .}, archive = {J_EAAI}, author = {Peichao Wang and Jiabao Wang and Yao Chen and Rui Zhang and Yang Li and Zhuang Miao}, doi = {10.1016/j.engappai.2025.110244}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110244}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Paying more attention to local contrast: Improving infrared small target detection performance via prior knowledge}, volume = {146}, year = {2025}, }
• (2025). Multi-step difference-driven domain adversarial network for few-sample fault detection in dynamic industrial systems. EAAI, 146, 110242. (www)

  @article{ , abstract = {The escalating production demands for manufacture result in heightened complexity in industrial processes, which leads to frequent changes in operating conditions, thus making few-sample scenarios commonplace. Although many deep learning methods achieve good performance in fault detection tasks, they mostly rely on sufficient data. Therefore, the lack of adequate data presents challenges for accurately representing the process. Moreover, the inherent interplay in processes and among equipment often manifests in data with dynamic characteristic. To address these challenges, we propose a dynamic domain adversarial network (DDAN) for dynamic few-sample fault detection. DDAN based on knowledge transfer, aiming to facilitate modeling data-poor domain with cross-domain information from the data-rich domain. It consists of three main components, a feature extractor, a data reconstructor, and a domain discriminator. To effectively extract features from dynamic samples in few-sample scenarios, a multi-step difference method is introduced. Combined with self-attention, the feature extractor highlights the most significant difference block in the dynamic representations. The output of data reconstructor is utilized for fault detection tasks, while the domain discriminator is applied for domain adaptation with a rebalancing loss. The proposed method is validated on a numerical case and a real-world alumina evaporation process. The experimental results demonstrate an average improvement in the fault detection rate of 2.9%, with an improvement exceeding 6% for latent variable faults.}, archive = {J_EAAI}, author = {Ruiyi Fang and Kai Wang and Xiaofeng Yuan and Zeyu Yang and Yalin Wang and Chunhua Yang}, doi = {10.1016/j.engappai.2025.110242}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110242}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-step difference-driven domain adversarial network for few-sample fault detection in dynamic industrial systems}, volume = {146}, year = {2025}, }
• (2025). Periodic event-triggered adaptive neural output feedback tracking control of unmanned surface vehicles under replay attacks. EAAI, 146, 110237. (www)

  @article{ , abstract = {This paper proposes a periodic event-triggered adaptive neural output feedback tracking control scheme for unmanned surface vehicles under replay attacks, where actuator saturation constraint and internal/external uncertainties are involved. To reduce attack signals entering the control system, an independent adaptive neural state observer is developed to recover the unavailable real velocities and mismatched compound uncertainties. Under the backstepping design framework, the adaptive neural-based single-parameter-learning method is involved to reconstruct the internal/external uncertainties, and an anti-replay-attacks output feedback tracking control law is devised. Furthermore, in the controller-actuator channel, a smooth saturation model is introduced and a periodic event-triggering mechanism is established to relieve the physical constraint of actuators. Theoretical analysis and simulation results verify the effectiveness of the developed scheme.}, archive = {J_EAAI}, author = {Guibing Zhu and Zhengyue Xu and Yun Gao and Yalei Yu and Lei Li}, doi = {10.1016/j.engappai.2025.110237}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110237}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Periodic event-triggered adaptive neural output feedback tracking control of unmanned surface vehicles under replay attacks}, volume = {146}, year = {2025}, }
• (2025). Development of optimized ensemble machine learning-based character segmentation framework for ancient tamil palm leaf manuscripts. EAAI, 146, 110235. (www)

  @article{ , abstract = {Background Character Segmentation plays a crucial role in analyzing the intricate script inscribed on these historical documents. Due to the unique script style and the physical characteristics of the palm leaves, character segmentation is considered to be challenging. Aim : An efficient character segmentation approach from palm-leaf manuscripts is developed in this work. Method ology: Initially, the images of Tamil palm leaf manuscripts are gathered manually and pre-processed with optimal binary thresholding and morphological operation. Then, the pre-processed images are utilized for Line segmentation by the Projection Profile method. The line-segmented images are fed into the feature extraction process. These extracted features are subjected to character segmentation by developing an Ensemble Machine Learning Structure (EMLS). EMLS is motivated by Bayesian Learning (BL), Support Vector Machine (SVM), and Artificial Neural Network (ANN). Moreover, parameter optimization is performed using the Hybridization of Sandpiper with Fireworks Algorithm (HSFA). Therefore, the developed method is suitable for real-time applications like mobile document scanning and educational-based applications. Result The experimental analysis is made to declare the efficiency of the developed approach, and given the accuracy to be 95.53%. Conclusion The developed model provides the character-segmented outcome and offers a promising tool for palm-leaf character recognition.}, archive = {J_EAAI}, author = {Mary Selvan and K. Ramar}, doi = {10.1016/j.engappai.2025.110235}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110235}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Development of optimized ensemble machine learning-based character segmentation framework for ancient tamil palm leaf manuscripts}, volume = {146}, year = {2025}, }
• (2025). Novel machine intelligent expedition with adaptive autoregressive exogenous neural structure for nonlinear multi-delay differential systems in computer virus propagation. EAAI, 146, 110234. (www)

  @article{ , abstract = {Computer viruses are significant from the perspective of reliable computer security for control infrastructure development that connotes the essential requirements for understanding virus spread and growth. This study investigates a delayed epidemiological computer virus model by applying artificial intelligence-inspired computing to analyze the behavior of a multi-delay differential system, incorporating the influence of the latent period on the dynamics of susceptible, infected, quarantined, and recovered computers using an adaptive autoregressive exogenous neural structure with Levenberg-Marquardt backpropagation. The reference data for executing networks is generated using the Adams numerical solver, incorporating parameters such as the computer infection rate, the saturation coefficient for calculating inhibitory effects, the immune loss rate of recovered computers, the number of infected computers in quarantine, the recovery rate, the mortality rate for each differential class, and delays associated with temporary immunity and the incubation period. The designed networks operate on generated synthetic data that are randomly distributed for testing, validation, and training samples to determine the approximate response of the nonlinear delay computer virus systems. The predictive solutions consistently aligned with the reference numerical outcomes indicating an error with negligible magnitude. The accuracy, stability, convergence, and efficiency of the designed intelligent networks are established on a thorough investigation by employing a variety of assessment metrics in terms of mean square error-based convergence curves, time series analysis of predicted outcomes, control parameter adaptation, error frequency distribution analysis, and exhaustive statistics on input-output and cross-correlations.}, archive = {J_EAAI}, author = {Nabeela Anwar and Aqsa Saddiq and Muhammad Asif Zahoor Raja and Iftikhar Ahmad and Muhammad Shoaib and Adiqa Kausar Kiani}, doi = {10.1016/j.engappai.2025.110234}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110234}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Novel machine intelligent expedition with adaptive autoregressive exogenous neural structure for nonlinear multi-delay differential systems in computer virus propagation}, volume = {146}, year = {2025}, }
• (2025). Non-parametric semi-supervised chiller fault diagnosis via variational compressor under severe few labeled samples. EAAI, 146, 110233. (www)

  @article{ , abstract = {Intelligent fault diagnosis for chiller is essential for energy efficiency optimization and health management within Heating, Ventilation, and Air Conditioning systems. Deep learning-based chiller fault diagnosis methods have demonstrated competitive performance. However, certain challenges remain: (1) in large-scale chiller systems, obtaining labeled data is costly, leading to scarce labeled datasets; (2) large amounts of unlabeled data have not been fully explored. This paper proposes a semi-supervised fault diagnosis algorithm based on system complexity quantification. Firstly, a variational autoencoder-based lossless compressor is trained using unlabeled data in an unsupervised manner. Subsequently, Kolmogorov complexity is approximated via the compressor to achieve entropy-based quantification of complexity. Using the normalized information distance algorithm, an information distance matrix is then calculated, which is combined with a K-nearest neighbors classifier for fault diagnosis. The proposed method is validated using the ASHRAE 1043-RP and HY-31C datasets, and the experimental results indicate that the proposed method can use the severe few labeled samples to obtain a better diagnostic accuracy.}, archive = {J_EAAI}, author = {Huazheng Han and Xuejin Gao and Huayun Han and Huihui Gao and Yongsheng Qi and Kexin Jiang}, doi = {10.1016/j.engappai.2025.110233}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110233}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Non-parametric semi-supervised chiller fault diagnosis via variational compressor under severe few labeled samples}, volume = {146}, year = {2025}, }
• (2025). Periodic decomposition and feature enhancement fusion for traffic forecasting. EAAI, 146, 110229. (www)

  @article{ , abstract = {With the rapid acceleration of urbanization, traffic prediction plays a crucial role in smart city development. This paper proposes an architecture called Periodic Decomposition and Feature Enhancement Fusion (PDGM) aimed at addressing the periodicity issue overlooked in existing traffic prediction methods. PDGM utilizes downsampling techniques to decompose the original traffic data into periodic components and enhances missing data through feature enhancement fusion, thereby improving the accuracy of traffic data prediction. Experimental results of this study demonstrate that PDGM outperforms state-of-the-art baseline models on three benchmark datasets, offering new possibilities for traffic data analysis and prediction tasks.}, archive = {J_EAAI}, author = {Xiaofei Kong and Hua Wang and Mingli Zhang and Fan Zhang}, doi = {10.1016/j.engappai.2025.110229}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110229}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Periodic decomposition and feature enhancement fusion for traffic forecasting}, volume = {146}, year = {2025}, }
• (2025). A human-robot interaction system for automated chemical experiments based on vision and natural language processing semantics. EAAI, 146, 110226. (www)

  @article{ , abstract = {Using collaborative robots to replace researchers in performing repetitive and hazardous chemical experiments can effectively enhance experimental efficiency. However, this technology still faces several challenges, including understanding researchers' natural language instructions, autonomously generating action sequences, and more. Therefore, we developed a general control framework for robots in automated chemical experiments based on visual and natural language semantic information. Firstly, starting with the recognition of keywords within Chinese language instructions, we established a domain dictionary for chemical experiment operations and proposed an instruction understanding model based on the bidirectional long-short-term memory and conditional random field(BiLSTM-CRF), enhancing the robot's cognitive ability towards user instructions. Then, a rule matching method for chemical experimental information and a multimodal information feature matching mechanism were established for command content verification and the automatic generation of multiple types of structured language. At the same time, a robot feedback mechanism was added, enabling human-computer interaction and establishing closed-loop control of the system. Finally, propose a robot action sequence generation mechanism based on hierarchical finite state machines(HFSM), transforming structured language into operational strategies for chemical experiments required by the robot. Experimental results show that on the instruction task comprehension dataset created in this paper, the proposed method improves the F1 score by up to 4.44% in the instruction keyword extraction task compared to other models. In addition, compared to traditional manual teaching control, this method significantly reduces time costs. This verifies that the method effectively enhances the robot's ability to comprehend Chinese instructions and generates reliable executable action sequences.}, archive = {J_EAAI}, author = {Zhuang Yang and Yu Du and Dong Liu and Kesong Zhao and Ming Cong}, doi = {10.1016/j.engappai.2025.110226}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110226}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A human-robot interaction system for automated chemical experiments based on vision and natural language processing semantics}, volume = {146}, year = {2025}, }
• (2025). Hybrid pathfinding optimization for the lightning network with reinforcement learning. EAAI, 146, 110225. (www)

  @article{ , abstract = {Payment channel networks, such as Bitcoin’s Lightning Network, have emerged to address blockchain scalability issues, enabling rapid transactions. Despite their potential, these networks often experience payment failures due to delays in pathfinding, unreliable routes and infrastructure issues, resulting in excessive carbon emissions. Current reinforcement learning solutions for payment channel networks mainly address issues like payment channel balance and routing fees but often overlook the infrastructure-related causes of payment failure. This paper introduces a novel reinforcement learning-based architecture that combines reinforcement learning agent with native deterministic pathfinding algorithms. This hybrid approach leverages the fast, complete solutions of deterministic algorithms while adapting to the network’s dynamic and probabilistic nature of payments to significantly enhance payment success rates. Experiments on real network snapshots show that this approach outperforms native pathfinding algorithms and state-of-the-art static optimization methods, providing improved reliability and efficiency in dynamic network conditions. In scenarios with payment failure rates greater than 5%, the proposed approach achieves a 10% higher payment success rate than existing methods, while maintaining balanced performance on economic key metrics such as the payment fee and throughput, and on sustainability key metrics such as payment path length, number of inter-country/continental hops, and average carbon intensity.}, archive = {J_EAAI}, author = {Danila Valko and Daniel Kudenko}, doi = {10.1016/j.engappai.2025.110225}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110225}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid pathfinding optimization for the lightning network with reinforcement learning}, volume = {146}, year = {2025}, }
• (2025). Development and implementation of a robotic moving target system for object recognition testing. EAAI, 146, 110223. (www)

  @article{ , abstract = {While artificial intelligence (AI) has made great progress in recent years, the scarcity of image datasets still limits their practical use in military domain. This paper proposes a robotic moving target system to build the image training dataset and test the object recognition algorithms. The robotic moving target system consists of human–machine interface and autonomous robotic moving platform. Once the working mode is set through the human–machine interface, robotic moving platform will move autonomously to simulate the object images in arbitrary poses. The path tracking controller is proposed to track the desired routes accurately. The effectiveness of the robotic moving target system was verified by following three types of routes in simulation and real experiments. The results show that robotic moving platform can move in arbitrary poses while it follows preset routes autonomously. The linear and angular velocities are adjusted smoothly. The average trajectory tracking error is less than 0.1 m. The function of it fully meets the requirement of building image training datasets and testing object recognition algorithms.}, archive = {J_EAAI}, author = {Haojie Zhang and Rongmin Liang and Bo Zhao and Chuankai Liu}, doi = {10.1016/j.engappai.2025.110223}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110223}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Development and implementation of a robotic moving target system for object recognition testing}, volume = {146}, year = {2025}, }
• (2025). Adaptive kalman-informed transformer. EAAI, 146, 110221. (www)

  @article{ , abstract = {The extended Kalman filter (EKF) is a widely adopted method for sensor fusion in navigation applications. A crucial aspect of the EKF is the online determination of the process noise covariance matrix reflecting the model uncertainty. While common EKF implementation assumes a constant process noise, in real-world scenarios, the process noise varies, leading to inaccuracies in the estimated state and potentially causing the filter to diverge. Model-based adaptive EKF methods were proposed and demonstrated performance improvements to cope with such situations, highlighting the need for a robust adaptive approach. In this paper, we derive an adaptive Kalman-informed transformer (A-KIT) designed to learn the varying process noise covariance online. Built upon the foundations of the EKF, A-KIT utilizes the well-known capabilities of set transformers, including inherent noise reduction and the ability to capture nonlinear behavior in the data. This approach is suitable for any application involving the EKF. In a case study, we demonstrate the effectiveness of A-KIT in nonlinear fusion between a Doppler velocity log and inertial sensors. This is accomplished using real data recorded from sensors mounted on an autonomous underwater vehicle operating in the Mediterranean Sea. We show that A-KIT outperforms the conventional EKF by more than 49.5% and model-based adaptive EKF by an average of 35.4% in terms of position accuracy.}, archive = {J_EAAI}, author = {Nadav Cohen and Itzik Klein}, doi = {10.1016/j.engappai.2025.110221}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110221}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive kalman-informed transformer}, volume = {146}, year = {2025}, }
• (2025). Fault diagnosis of motor bearing in complex scenarios based on mamba and indicative contrastive learning. EAAI, 146, 110216. (www)

  @article{ , abstract = {Motors are critical components of rotating machinery, and bearings are essential elements of motors. Therefore, monitoring the health of motor bearings is crucial. This paper addresses three major challenges in motor bearing fault diagnosis: noise interference, sample imbalance, and cross-domain diagnosis. To overcome these challenges and achieve high-precision fault diagnosis, This paper propose a novel method that integrates a Global-Local Fusion Feature Extractor (GLFFE) and Indicative Contrastive Learning (ICL). First, to mitigate the issues of noise interference and sample imbalance, design a Bidirectional Dual-Head Gated Mamba model (BDMamba) and a Residual Local Feature Extraction Module (RLFEM) for extracting comprehensive global and local fault features. The dual-branch architecture significantly enhances the extraction of local features while preserving the global modeling capabilities of Mamba. Next, proposed a Feature Fusion Mamba Module (FFMM) to effectively combine global and local features, thereby enriching feature diversity and reducing redundancy. To address the challenge of cross-domain fault diagnosis, integrate contrastive learning with working condition label prompts to enhance the model’s adaptability to multiple operating conditions. This approach also addresses the limitations of traditional contrastive learning, such as the requirement for large sample sizes. Finally, extensive experiments are conducted on two datasets to evaluate the robustness of the proposed method in cross-domain fault diagnosis under noise, sample imbalance, and variations in load and speed. The results demonstrate that the proposed method achieves high fault diagnosis accuracy and robustness and exhibits strong cross-domain transfer capabilities.}, archive = {J_EAAI}, author = {Jun Xu and Yunji Zhao and Wenming Bao and Chao Hao}, doi = {10.1016/j.engappai.2025.110216}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110216}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis of motor bearing in complex scenarios based on mamba and indicative contrastive learning}, volume = {146}, year = {2025}, }
• (2025). Automated classification of thyroid disease using deep learning with neuroevolution model training. EAAI, 146, 110209. (www)

  @article{ , abstract = {Background: Thyroid disease is a common endocrine disorder; its timely and accurate diagnosis is important. Using clinical and laboratory data input, we developed an artificial neural network (ANN) for thyroid disease classification, incorporating an evolutionary algorithm for network optimization. Methods: The proposed model combined ANN with a genetic algorithm (GA), which iteratively modified the weights and biases of the ANN architecture. The weights, encoded as genes in a chromosome and represented as one-dimensional vectors, were updated in each iteration. Binary cross-entropy loss was used as the fitness function to calculate the suitability of solutions in the genetic algorithm. The model was trained and tested on an open-access Hypothyroid disease dataset comprising multiparametric variables of 3772 samples (291 thyroid patients and 3481 controls). Results: Our model attends % 99 . 14 accuracy for binary classification (thyroid disease vs. normal), outperforming published models. Conclusion: Incorporating GA optimization into ANN enabled the model to explore diverse solutions and escape local optima more effectively, leading to better performance and generalizability. The excellent results support the feasibility of implementing the proposed model for thyroid disease screening in the clinical setting.}, archive = {J_EAAI}, author = {Mohammad Rashid Dubayan and Sara Ershadi-Nasab and Mariam Zomorodi and Pawel Plawiak and Ryszard Tadeusiewicz and Mohammad Beheshti Roui}, doi = {10.1016/j.engappai.2025.110209}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110209}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated classification of thyroid disease using deep learning with neuroevolution model training}, volume = {146}, year = {2025}, }
• (2025). Artificial intelligence based-improving reservoir management: An attention-guided fusion model for predicting injector–producer connectivity. EAAI, 146, 110205. (www)

  @article{ , abstract = {The existing oil reservoir demonstrated suboptimal inter-well connectivity, leading to irregular depletion and reduced overall production efficiency. This article demonstrates the Attention-Guided Fusion Model for Injector–Producer Connectivity Estimation (AGFM). The model has an attention mechanism in the first path, pulling discernment from the relationships between injectors and producers through the training phase, extracting the attention weight. This attention weight is then devoted to the second path, utilising a Long-Short-Term Memory (LSTM)-based architecture. The first path is only to the training stage. In contrast, the second path is used during training and testing, improving the ability of the model to find a more significant representation of the data. This makes the model robust enough to predict reservoir performance and interconnectivity, giving valuable insights to optimise field operations. The AGFM undergoes an evaluation with two different injection liquids (carbon dioxide ( C O 2 ) and water) in three scenarios: all water and all CO2 alternating between water and CO2 as a flooding liquid. The evaluation emphasises the efficacy of the model in all scenarios, making it a practical tool for estimating reservoir connectivity and enhancing oil recovery strategies. The water alternating gas (WAG) process performed high accuracy rates, with 82.1% for oil production, 86.8% for water production, and 86.9% for gas production. Our proposed method consistently demonstrates superior performance through comprehensive experimentation and rigorous analysis compared to existing approaches. The results reveal spatial interwell connectivity, confirming the efficacy and potential of our method as a more effective solution for reservoir recovery.}, archive = {J_EAAI}, author = {Ahmed Saihood and Tariq Saihood and Sabah Abdulazeez Jebur and Christine Ehlig-Economides and Laith Alzubaidi and Yuantong Gu}, doi = {10.1016/j.engappai.2025.110205}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110205}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence based-improving reservoir management: An attention-guided fusion model for predicting injector–producer connectivity}, volume = {146}, year = {2025}, }
• (2025). A fine-tuning approach based on spatio-temporal features for few-shot video object detection. EAAI, 146, 110198. (www)

  @article{ , abstract = {This paper describes a new Fine-Tuning approach for Few-Shot object detection in Videos that exploits spatio-temporal information to boost detection precision. Despite the progress made in the single image domain in recent years, the few-shot video object detection problem remains almost unexplored. A few-shot detector must quickly adapt to a new domain with a limited number of annotations per category. Therefore, it is not possible to include videos in the training set, hindering the spatio-temporal learning process. We propose augmenting each training image with synthetic frames to train the spatio-temporal module of our method. This module employs attention mechanisms to mine relationships between proposals across frames, effectively leveraging spatio-temporal information. A spatio-temporal double head then localizes objects in the current frame while classifying them using both context from nearby frames and information from the current frame. Finally, the predicted scores are fed into a long-term object-linking method that generates object tubes across the video. By optimizing the classification score based on these tubes, our approach ensures spatio-temporal consistency. Classification is the primary challenge in few-shot object detection. Our results show that spatio-temporal information helps to mitigate this issue, paving the way for future research in this direction. FTFSVid achieves 41.9 AP50 on the Few-Shot Video Object Detection (FSVOD-500) and 42.9 AP50 on the Few-Shot YouTube Video (FSYTV-40) dataset, surpassing our spatial baseline by 4.3 and 2.5 points. Additionally, FTFSVid outperforms previous few-shot video object detectors by 3.2 points on FSVOD-500 and 14.5 points on FSYTV-40, setting a new state-of-the-art.}, archive = {J_EAAI}, author = {Daniel Cores and Lorenzo Seidenari and Alberto Del Bimbo and Víctor M. Brea and Manuel Mucientes}, doi = {10.1016/j.engappai.2025.110198}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110198}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A fine-tuning approach based on spatio-temporal features for few-shot video object detection}, volume = {146}, year = {2025}, }
• (2025). Beyond spatial neighbors: Utilizing multivariate transfer entropy for interpretable graph-based spatio–temporal forecasting. EAAI, 146, 110161. (www)

  @article{ , abstract = {Spatio–temporal forecasting is a challenging task that requires modeling complex interactions between multiple time series. While graph-based models have emerged as compelling tools for this task, their effectiveness heavily depends on the underlying graph structure that captures spatial dependencies but ignores the temporal relationships. To address this challenge, we propose the Multivariate Transfer Entropy-Multivariate Time series forecasting with Graph Neural Networks (MTE-MTGNN), a hybrid approach that combines statistical and deep learning methods. MTE-MTGNN introduces an interpretable graph construction layer founded on Multivariate Transfer Entropy, which effectively captures both spatial and temporal dependencies in the data. Empirical evaluations across five benchmark datasets demonstrate the superiority of our proposed approach in terms of predictive accuracy. The model shows particular strength in few-shot scenarios where traditional forecasting approaches typically struggle, achieving performance improvements of up to 3% on the RRSE metric in the exchange rate dataset and up to 4% on the correlation metric in the Hungarian Chickenpox dataset compared to state-of-the-art baselines. The findings witnessed across different experiments translate into significant practical benefits for real-world engineering applications and different domains.}, archive = {J_EAAI}, author = {Safaa Berkani and Adil Bahaj and Bassma Guermah and Mounir Ghogho}, doi = {10.1016/j.engappai.2025.110161}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110161}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Beyond spatial neighbors: Utilizing multivariate transfer entropy for interpretable graph-based spatio–temporal forecasting}, volume = {146}, year = {2025}, }
• (2025). A survey on learning with noisy labels in natural language processing: How to train models with label noise. EAAI, 146, 110157. (www)

  @article{ , abstract = {When applying deep neural network language models to related systems (e.g., question answering systems, chatbots, and intelligent assistants), many datasets contain different types or degrees of label noise. Label noise can lead to a decline in model performance and an increase in resource consumption. Therefore, learning with noisy labels is becoming an important task in Natural Language Processing (NLP). This paper aims to collect, analyze, and evaluate methods for learning with label noise in NLP. First, we analyze the relationship between data feature extraction, prediction output, and optimization in the context of noise robustness to help researchers understand the mechanisms behind noise generation. Based on this, we classified the noise processing methods into five types according to the training process: feature vector, transition matrix, prediction confidence, loss improvement, and data weighting. We analyze each method and conduct a systematic evaluation across six metrics. In addition, we summarized the commonly used resources such as datasets, open source codes, etc. Finally, we also analyzed the challenges faced in current research and the potential opportunities. As a comprehensive survey, this work will help researchers and industry developers to understand the current state of research and unique challenges facing label-noise learning, which facilitate the selection and combination of different methods in applications to further advancements.}, archive = {J_EAAI}, author = {Han Zhang and Yazhou Zhang and Jiajun Li and Junxiu Liu and Lixia Ji}, doi = {10.1016/j.engappai.2025.110157}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110157}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A survey on learning with noisy labels in natural language processing: How to train models with label noise}, volume = {146}, year = {2025}, }
• (2025). Multi-scale patch transformer with adaptive decomposition for carbon emissions forecasting. EAAI, 146, 110153. (www)

  @article{ , abstract = {Rapid urbanization and industrialization have led to a significant increase in carbon emissions, posing a challenge for sustainable environmental management. However, current research predominantly focuses on traditional forecasting models that often overlook the complexity and dynamic nature of environmental data. To address this, a novel multi-scale patch transformer with adaptive decomposition (MPDformer) has been developed specifically for forecasting carbon emissions. This model introduces an adaptive decomposition method that dynamically assesses the noise level, trend, and stationarity of data to select the most appropriate decomposition technique. In addition, the use of a Transformer with multi-scale patches can optimize the use of information at different granularities in the decomposed sub-series for time series prediction of carbon emissions. Experimental validations have shown that this method possesses an exceptional capability to discern complex temporal dependencies within fluctuating environmental data, consistently outperforming comparative models across a range of carbon emissions datasets and various forecasting horizons. These results indicate the potential for more accurate and reliable carbon emissions forecasts, which can contribute to better-informed decisions in sustainable energy planning and environmental management.}, archive = {J_EAAI}, author = {Xiang Li and Lei Chu and Yujun Li and Fengqian Ding and Zhenzhen Quan and Fangx Qu and Zhanjun Xing}, doi = {10.1016/j.engappai.2025.110153}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110153}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale patch transformer with adaptive decomposition for carbon emissions forecasting}, volume = {146}, year = {2025}, }
• (2025). Explainable reinforcement learning for powertrain control engineering. EAAI, 146, 110135. (www)

  @article{ , abstract = {In this paper we demonstrate a practical post-hoc approach for explainable reinforcement learning (RL) in vehicle powertrain control. The goal is to exploit the advantages of RL yet obtain a solution that is feasible to implement in safety-critical control engineering problems. This means finding a solution that balances optimal product design with the required engineering effort, while maintaining the transparency necessary for safety-critical applications. Our method is based on initially training a neural network based RL policy and converting it into a look-up table, using a decision tree (DT) as an intermediary. The DT is limited to a certain depth, resulting in a look-up table of manageable size that can be directly tested, implemented and evaluated by control engineers. In order to evaluate this approach, a set of RL expert policies were used to train DTs with increasing depth, showing the regions where the DT solution can outperform benchmarks while still remaining small enough to translate to a manageable look-up table. Our approach involves standard Python libraries, lowering the barrier for implementation. This approach is not just relevant to powertrain control, but offers a practical approach for all regulated domains which could benefit from application of RL.}, archive = {J_EAAI}, author = {C. Laflamme and J. Doppler and B. Palvolgyi and S. Dominka and Zs.J. Viharos and S. Haeussler}, doi = {10.1016/j.engappai.2025.110135}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110135}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable reinforcement learning for powertrain control engineering}, volume = {146}, year = {2025}, }
• (2025). Intelligent exogenous networks with bayesian distributed backpropagation for nonlinear single delay brain electrical activity rhythms in parkinson's disease system. EAAI, 145, 110281. (www)

  @article{ , abstract = {In this study, a novel intelligent adaptive exogenous network backpropagated with a Bayesian distributive scheme is introduced for nonlinear Parkinson's disease systems (NPDS) represented with three differential classes governing the brain's electrical activity rhythms (BEAR) at different cerebral cortex positions considering single and multiple delays in one continuing response. The computing structure is formulated with a multi-layer architecture of nonlinear autoregressive exogenous networks (NARX) with backpropagation through a Bayesian distributed algorithm (BDA). The synthetic datasets for the execution of NARX-BDA are acquired through the Adams numerical solver for NPDS involving single and multiple delays in one variable of BEAR for different sensor locations on the cerebral cortex. The designed computing structure of NARX-BDA is operated arbitrarily for acquired datasets and used for training samples for network formulation on mean square error (MSE) sense while testing samples to validate the performance on unbiased inputs. The exhaustive numerical experimentation studies are conducted for NARX-BDA in solving delayed variants of NPDS, and comparative studies are carried out through numerical solutions by means of convergence curves on MSE for training and testing instances, absolute error, error histograms, statistical studies on regression and correlation measurements to certify the performance.}, archive = {J_EAAI}, author = {Roshana Mukhtar and Chuan-Yu Chang and Muhammad Asif Zahoor Raja and Naveed Ishtiaq Chaudhary and Nabeela Anwar and Iftikhar Ahmad and Chi-Min Shu}, doi = {10.1016/j.engappai.2025.110281}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110281}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent exogenous networks with bayesian distributed backpropagation for nonlinear single delay brain electrical activity rhythms in parkinson's disease system}, volume = {145}, year = {2025}, }
• (2025). Self-supervised deep contrastive and auto-regressive domain adaptation for time-series based on channel recalibration. EAAI, 145, 110280. (www)

  @article{ , abstract = {Time-series based unsupervised domain adaptation (UDA) techniques have been widely adopted to the applications of intelligent systems, such as sleep staging, fault diagnosis, and human activity recognition. However, recently methods have overlooked the importance of temporal feature representations and the distribution discrepancies across domains, which deteriorated UDA performance. To address these challenges, we proposed a novel S elf-supervised D eep C ontrastive and A uto-regressive D omain A daptation (SDCADA) model for cross-domain time-series classification. Specifically, the cross-domain mixup preprocessing strategy is applied to reduce sample-level distribution discrepancy, then we proposed to introduce the channel recalibration module for adaptively selecting discriminative representations. Afterwards, the auto-regressive discriminator and teacher model are proposed to reduce the distribution discrepancies of feature representations. Finally, a total of six losses, including contrastive and adversarial learning, are weighted and jointly optimized to train the SDCADA model. The proposed SDCADA model has been systematically experimented on four cross-domain time-series benchmarked datasets, and its classification performance surpasses several recently proposed state-of-the-art models. Moreover, it effectively captures discriminative and comprehensive cross-domain time-series feature representations with parameter insensitivity.}, archive = {J_EAAI}, author = {Guangju Yang and Tian-jian Luo and Xiaochen Zhang}, doi = {10.1016/j.engappai.2025.110280}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110280}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-supervised deep contrastive and auto-regressive domain adaptation for time-series based on channel recalibration}, volume = {145}, year = {2025}, }
• (2025). DeepEvader: An evasion tool for exposing the vulnerability of deepfake detectors using transferable facial distraction blackbox attack. EAAI, 145, 110276. (www)

  @article{ , abstract = {Deepfakes are highly realistic fabricated media created using sophisticated generative techniques that are massively used to spread disinformation in cyberspace. Implementing a deepfake detector is crucial to identify and counter these threats, thereby ensuring the integrity of digital media. However, these detection systems are susceptible to adversarial attacks that exploit vulnerabilities to circumvent identification. The present study investigates the vulnerability of deepfake detectors to adversarial attacks by proposing a new Facial Distraction Black-Box Attack (FDB attack) framework that is visually realistic, resilient, and demonstrates formidable attacking capabilities. The proposed black-box attack is capable of successfully evading deepfake detectors without the need for access to the target detector's parameters or architectural specifications. It exhibits high transferability across a variety of deepfake detectors, including end-to-end deep learning, fused, and unified models, utilizing five standard datasets. The proposed attack beats the performance of the best detector and reduces the accuracy of the detectors significantly from 99.9% to 49.7%. Rigorous experimentation was performed to show the effectiveness of the proposed attack in comparison with state-of-the-art attacks. Additionally, we have developed a specialized penetration testing tool named DeepEvader to uncover and analyze the vulnerabilities of existing deepfake detectors. By exposing weaknesses in current detection methodologies, our work highlights an urgent need for robust detection mechanisms to combat deceptive digital content effectively. Our research reveals flaws in existing detection methods, emphasizing the immediate requirement for more effective and robust detection techniques to combat deepfakes.}, archive = {J_EAAI}, author = {Qurat Ul Ain and Ali Javed and Aun Irtaza}, doi = {10.1016/j.engappai.2025.110276}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110276}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DeepEvader: An evasion tool for exposing the vulnerability of deepfake detectors using transferable facial distraction blackbox attack}, volume = {145}, year = {2025}, }
• (2025). Thermo-mesh transformer network for generalizable three-dimensional temperature prediction with mechanism-based representation. EAAI, 145, 110274. (www)

  @article{ , abstract = {The application of artificial intelligence for predicting physical fields holds significant potential in military and manufacturing domains. However, physical phenomenon prediction networks face challenges in accurately modeling physical laws across diverse scenarios while ensuring computational efficiency and generalization. This paper presents a novel Thermo-Mesh Transformer Network (TMTN) for universal three-dimensional (3D) temperature reconstruction from a single two-dimensional (2D) image. To enhance the network's understanding of heat transfer phenomena, a multi-scale feature extraction module is proposed to effectively capture and integrate both local and global features from the mesh and the image of the complex thermodynamic scenario. A mesh-enhanced transformer block (METB) is designed to implicitly represent the energy conservation equation, combining a Graph Convolutional Network (GCN) for modeling the diffusion term and a Transformer for modeling the source term. This innovative mechanism-based representation architecture directly integrates physical laws, enabling TMTN to predict heat transfer phenomena across various shapes and conditions, including untrained models and out-of-range conditions. Additionally, a parameter-efficient design based on an improved Laplacian matrix is proposed, leveraging the PointNet framework and farthest point sampling to efficiently represent global vertex relationships via Transformer with reduced parameters. Experimental results demonstrate that TMTN achieves a 38% reduction in error compared to state-of-art methods and exhibits robust performance under out-of-range and untrained conditions. TMTN offers valuable engineering applications in military infrared reconnaissance and the thermal management of civilian systems like energy storage and heating devices.}, archive = {J_EAAI}, author = {Jincheng Chen and Feiding Zhu and Yuge Han and Dengfeng Ren}, doi = {10.1016/j.engappai.2025.110274}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110274}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Thermo-mesh transformer network for generalizable three-dimensional temperature prediction with mechanism-based representation}, volume = {145}, year = {2025}, }
• (2025). Deep learning based methodological approach for prediction of dynamic modulus and phase angle of asphalt concrete. EAAI, 145, 110269. (www)

  @article{ , abstract = {The present work proposes a deep learning approach to predict dynamic modulus and phase angle of asphalt mixtures. The dynamic modulus and phase angle data reported in national cooperative highway research program 9–19 were utilized to validate the proposed approach. Within this database 201 distinct asphalt mixtures were selected. Dynamic modulus and phase angle mastercurves were constructed for individual mixtures using combination of two sigmoidal functions and two temperature shift factor determination approaches. The input variables for deep learning model consisted of information regarding reduced frequency, binder properties, aggregate gradation, and mixture volumetrics. When compared to other input variables, reduced frequency and binder properties were highly correlated with dynamic modulus and phase angle. Further, recursive feature elimination was used to rank all input variables. Using these ranked input variables, deep learning-based models for predicting dynamic modulus and phase angle were developed. The deep learning architecture (finalized through exhaustive optimization) dependent on parameter under consideration, and mastercurve construction approach was adopted. Detailed statistical analysis indicated dynamic modulus predictive models performed better when compared to phase angle predictive models. The numerical values of goodness of fit indicators showed that accuracy of deep learning-based model was dependent on mastercurve construction approach adopted. Overall results indicated that the deep learning-based models can predict dynamic modulus and phase angle with good accuracy. The output from the proposed deep learning models can be used as direct input into pavement design framework, which can result in accurate prediction of pavement performance.}, archive = {J_EAAI}, author = {Nishigandha Rajeshwar Jukte and Aravind Krishna Swamy}, doi = {10.1016/j.engappai.2025.110269}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110269}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning based methodological approach for prediction of dynamic modulus and phase angle of asphalt concrete}, volume = {145}, year = {2025}, }
• (2025). A hierarchical deep reinforcement learning method for coupled transportation and power distribution system dispatching. EAAI, 145, 110264. (www)

  @article{ , abstract = {The randomness and dimensionality growth of variables in the Coupled transportation and power distribution systems (CTPS) pose challenges for effectively solving CTPS dispatching tasks. This paper presents a hierarchical deep reinforcement learning (HDRL) method, which disperses the action and state space of CTPS onto decision-making layer and autonomous optimization layer. The Cloud DRL model in the decision-making layer is responsible for the load assignment task of charging stations. The distribution network (DN) and transportation network (TN) DRL models in the autonomous optimization layer are responsible for optimizing the DN and TN respectively. A layer-wise training method is adopted to alleviate the asynchronous convergence problem of HDRL. Firstly, the Gurobi assists in achieving the efficient training of Cloud DRL model by ensuring the reward effectiveness of autonomous optimization layers. Meanwhile, the differential evolution (DE) algorithm assists in optimizing the diversity and focalization of the Transitions by controlling distribution patterns of species initialization, during the pre-sampling and training stage. Then, the trained Cloud DRL model is frozen to train the DN and TN DRL models. This method is tested on two different sizes of CTPS. Simulation analysis shows that this method improves the training performance of the HDRL model.}, archive = {J_EAAI}, author = {Qi Han and Xueping Li and Liangce He}, doi = {10.1016/j.engappai.2025.110264}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110264}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hierarchical deep reinforcement learning method for coupled transportation and power distribution system dispatching}, volume = {145}, year = {2025}, }
• (2025). Next-generation healthcare: Digital twin technology and monkeypox skin lesion detector network enhancing monkeypox detection - Comparison with pre-trained models. EAAI, 145, 110257. (www)

  @article{ , abstract = {The rise of digital healthcare has led to the adoption of various technologies aimed at enhancing health operations, patient well-being, and healthcare costs. Digital Twin (DT) technology is a pivotal innovation in this domain. Monkeypox virus (MPXV), a zoonotic virus, poses a significant public health risk, particularly in remote regions of Central and West Africa. Early diagnosis of monkeypox lesions is crucial but challenging due to similarities with other skin conditions. Many studies have employed deep-learning models to detect the monkeypox virus. However, those models often require substantial storage space. This research introduces the Monkeypox Skin Lesion Detector Network (MxSLDNet), an automated digital twin framework designed to enhance digital healthcare operations by enabling early detection and classification of monkeypox and non-monkeypox lesions. Monkeypox Skin Lesion Detector Network (MxSLDNet) significantly advances monkeypox lesion identification, outperforming conventional models like Visual Geometry Group 19 (VGG-19), Densely Connected Network 121 (DenseNet-121), Efficient Network B4 (EfficientNet-B4) and Residual Network 101 (ResNet-101) regarding precision, recall, F1-score, and accuracy, while requiring less storage. This innovation addresses the critical issue of storage demands, making the Monkeypox Skin Lesion Detector Network (MxSLDNet) a viable solution for early monkeypox lesion detection in resource-limited healthcare settings. Utilizing the “Monkeypox Skin Lesion Dataset” with 1428 monkeypox and 1764 non-monkeypox images, Monkeypox Skin Lesion Detector Network (MxSLDNet) achieves high recall, precision, and F1-scores of 0.96, 0.95, and 0.95, respectively. Integrating digital twins into healthcare promises to create a scalable, intelligent, and comprehensive health ecosystem, enhancing treatments by connecting patients and healthcare providers.}, archive = {J_EAAI}, author = {Vikas Sharma and Akshi Kumar and Kapil Sharma}, doi = {10.1016/j.engappai.2025.110257}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110257}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Next-generation healthcare: Digital twin technology and monkeypox skin lesion detector network enhancing monkeypox detection - Comparison with pre-trained models}, volume = {145}, year = {2025}, }
• (2025). Robust nonlinear control of permanent magnet synchronous motor drives: An evolutionary algorithm optimized passivity-based control approach with a high-order sliding mode observer. EAAI, 145, 110256. (www)

  @article{ , abstract = {Permanent Magnet Synchronous Machines (PMSMs) have revolutionized motor design by replacing traditional components like rotor windings, brushes, and sliding contacts with permanent magnets. This innovation has significantly improved operational efficiency and reduced maintenance needs. However, controlling PMSMs remains challenging due to the changing dynamics of the machine over time and its sensitivity to different environmental conditions. To tackle these challenges, this study presents a novel nonlinear control approach called passivity-based control (PBC). Unlike conventional methods, PBC manages both the electrical and mechanical dynamics of the system, focusing on energy flow and dissipation to maintain stability. To make the control more robust, the approach combines a nonlinear observer and a high-order sliding mode controller (HSMC), which enhance the system's ability to handle disturbances and parameter changes. Additionally, the study uses Genetic Algorithm (GA) optimization to fine-tune the parameters of the PBC, observer, and HSMC. This optimization improves the motor's tracking accuracy and robustness against external disruptions. The result is a control framework that preserves the natural dynamics of PMSMs while improving their stability and performance. Experimental validation using the platform for real-time simulation (OPAL-RT) and real world on a PMSM using dSPACE DS1202 board demonstrates that this method outperforms existing techniques under a variety of operating conditions, highlighting its effectiveness and reliability.}, archive = {J_EAAI}, author = {Youcef Belkhier and Siham Fredj and Haroon Rashid and Mohamed Benbouzid}, doi = {10.1016/j.engappai.2025.110256}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110256}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust nonlinear control of permanent magnet synchronous motor drives: An evolutionary algorithm optimized passivity-based control approach with a high-order sliding mode observer}, volume = {145}, year = {2025}, }
• (2025). Energy performance prediction of centrifugal pumps based on adaptive support vector regression. EAAI, 145, 110247. (www)

  @article{ , abstract = {It is of great significance to speed up the development and optimization of pumps with energy performance prediction methods. Machine learning is widely used for performance prediction of centrifugal pumps due to its fast and accurate predictions. However, the prediction model performance distinctly for the different geometry and performance parameters. This paper proposes an adaptive support vector regression (SVR) model for predicting centrifugal pump energy performance, which incorporates input-output correlation analysis and differential evolution to automatically adjust the input parameter weights. The model's performance was validated against experimental data, yielding mean absolute residuals (MAR) of 0.174 for head, 0.113 for power, and 1.658 for efficiency. Additionally, the model achieved an R 2 of 0.995 and a mean square error (MSE) of 2.99. In multi-operation conditions, by adjusting the parameter vector, the adaptive SVR reduced the mean absolute relative error (MARE) of head, power, and efficiency to 0.443%, 1.07%, and 6.63%, respectively, representing improvements of 79.6%, 86.2%, and 31.6% compared to the original SVR model. The proposed model also outperformed the adaptive least squares support vector regression (LSSVR).}, archive = {J_EAAI}, author = {Huican Luo and Peijian Zhou and Jiayi Cui and Yang Wang and Haisheng Zheng and Yantian Wang}, doi = {10.1016/j.engappai.2025.110247}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110247}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Energy performance prediction of centrifugal pumps based on adaptive support vector regression}, volume = {145}, year = {2025}, }
• (2025). Driver's trust assessment based on situational awareness under human-machine collaboration driving. EAAI, 145, 110243. (www)

  @article{ , abstract = {The human-machine co-driving can effectively take into account the superior characteristics of human driver and agent, and generate the ‘1 + 1>2’ collaboration mode. However, drivers with autonomous consciousness and different behaviors actively intervene in the control of vehicles, which will trigger the conflict between human and intelligent driving system. The human-machine mutual trust has gradually developed one of the key technologies to mitigate human-machine conflicts. In this study, from the perspective of machine trust towards human, a comprehensive trust evaluation model (CTEM) for intelligent systems towards driver intervention behavior was established. Based on the driver's situational awareness (SA) recovery process, the model focused on the hierarchical decision-making model from perception to cognition. A residual convolutional neural network based on attention mechanism was proposed to identify active intervention and accidental touches. Then the perceived trust assessment (PTA) model was established on the basis of visual geometry group16 (VGG16) network. For the cognitive trust assessment (CTA) model, the long short-term memory (LSTM) codec structure was employed to predict the vehicle trajectory, and the risk field was structured to quantify the risk value of the future trajectory. The CTEM model was constructed by integrating the PTA and CTA models, which serve as its foundational components. Finally, the driving simulator experiments was implement to verify the proposed model, and the results demonstrated that the CTEM could availably distinguish and assess the driver's SA ability. The construction of trust evaluation model will provide effective support for the improvement of human-machine co-driving safety.}, archive = {J_EAAI}, author = {Qinyu Sun and Hang Zhou and Rui Fu and Yaning Xu and Chang Wang and Yingshi Guo}, doi = {10.1016/j.engappai.2025.110243}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110243}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Driver's trust assessment based on situational awareness under human-machine collaboration driving}, volume = {145}, year = {2025}, }
• (2025). Heterogeneous graph contrastive learning-based transductive health condition assessment of francis turbine unit. EAAI, 145, 110240. (www)

  @article{ , abstract = {To assess the Francis turbine unit's (FTU) degradation status from the onsite data without state labeling, a series of health benchmark model (HBM) driven health condition assessment (HCA) methods have been proposed. However, some limitation still exists, including: 1) Multiple similar HBMs are repeatedly constructed and trained to process multiple heterogeneous monitoring signals, however, there is a lack of a unified model that can handle the full task flow directly. 2) Simple linear methods are used to fuse multiple heterogeneous signals to construct performance degradation indexes (PDIs), ignoring temporal changes in the relationships within the signals and resulting in inadequate unit state representation. In this paper, a heterogeneous graph contrastive learning-based transductive health condition assessment of FTU considering multi-source monitoring signals is proposed. First, multi-source heterogeneous signals are converted into a series of heterogeneous signal graphs via the designed similarity-based edge-connection function considering the temporal representation differences. Then, an innovative graph-level heterogenous signal fusion method is proposed to represent unit conditions completely. By concatenating adjacency matrices of the heterogeneous signal graphs, multiple edge connections within the assessment period data are fused, obtaining a unique interactive graph with balanced and extended edge connections. Further, GCL-driven graph representator model is used to maximize the differences between the healthy and degraded interactive graph (IGs), and then the distances between the model readout vectors of the above two graphs is calculated as IPDIs, considering the signal correlations influenced by unit state changes. Verification experiments show that the proposed transductive HCA method effectively evaluates FTU's degradation without state labeling.}, archive = {J_EAAI}, author = {Fengyuan Zhang and Jie Liu and Yujie Liu and Yuxin Li and Ran Duan and Zhidi Chen and Xingxing Jiang}, doi = {10.1016/j.engappai.2025.110240}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110240}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Heterogeneous graph contrastive learning-based transductive health condition assessment of francis turbine unit}, volume = {145}, year = {2025}, }
• (2025). Omni-scale spatio-temporal attention network for impact localization of sandwich composite panels. EAAI, 145, 110238. (www)

  @article{ , abstract = {Sandwich composite panels (SCPs) have been widely used in aerospace, shipbuilding, and other fields due to their excellent flexural rigidity, designability, and other advantages. However, SCPs are susceptible to damage from external impacts during service, resulting in internal damage that is difficult to detect, and seriously affecting the equipment's structural performance and service life. Accurate impact localization is a prerequisite for condition monitoring and damage detection of SCPs. However, existing methods cannot achieve accurate impact localization of SCPs since the core material of SCPs absorbs impact energy and suppresses signal transmission. Hence, this paper investigates the utilization of deep learning technology to analyze impact signals, aiming to achieve accurate localization. Firstly, an impact signal acquisition scheme is designed. Then, this paper designs an omni-scale spatio-temporal attention network (OSTNet). It has the capability to extract all temporal scale features and spatio-temporal correlations of impact signals, thereby effectively handling SCPs impact signals with concentrated energy and low discriminability. Moreover, OSTNet also processes the impact signals in frequency domain to further improve accuracy. Finally, we set up the experimental platform and established an impact localization dataset of SCPs to test the localization effect of OSTNet. Experimental results show that the maximum, minimum, and average location errors of OSTNet are 7.410 mm, 1.145 mm, and 5.157 mm, respectively, which are 11.93 mm, 1.776 mm, and 4.372 mm lower than the optimal values of classical comparison models SFNet, PZTNet, LSTM, and RNN, respectively. And, OSTNet outperforms the recently published MNQGN and DAGNN methods, with 2.378 mm, 5.070 mm, and 1.070 mm decreasing, respectively. Moreover, the statistical tests also strongly confirm that OSTNet exhibits significant differences compared to the 6 comparative networks. Furthermore, the effectiveness of OSTNet is validated by employing impact signal data with varying energy levels, resulting in satisfactory localization performances being achieved.}, archive = {J_EAAI}, author = {Yang Zhang and Bo Yang and Shilong Wang and Fan Mo and Fengyang Bi and Yan He}, doi = {10.1016/j.engappai.2025.110238}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110238}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Omni-scale spatio-temporal attention network for impact localization of sandwich composite panels}, volume = {145}, year = {2025}, }
• (2025). A learning-based algorithm for turn-based orbital pursuit-evasion problem with reaction-time delay. EAAI, 145, 110231. (www)

  @article{ , abstract = {In this paper, we propose an artificial intelligence-based methodology to investigate the impulsive orbital pursuit-evasion problem, taking into account the often-neglected factor of reaction time delay. This study defines this scenario as a Delayed Turn-based Orbital Pursuit-Evasion Game (DT-OPEG) and establishes relevant concepts and definitions based on orbital dynamics and game theory. Subsequently, considering several constraints inherent in real-world missions, including nonlinear orbital dynamics, maneuvering capabilities, fuel reserves, and mission duration, we formulate the problem modeling for DT-OPEG. To address the complexity of this problem, especially the challenge of incorporating action delays, we propose a Delayed Turn-based Multi-Agent Deep Deterministic Policy Gradient (DT-MADDPG) algorithm. The establishment process of this algorithm includes establishing a Turn-based Markov Decision Process (T-MDP) model with reaction-time delay, constructing a turn-based training framework, developing a network architecture based on MADDPG, and designing reward functions. Finally, simulation analyses are conducted for both two-dimensional and three-dimensional DT-OPEG scenarios, confirming the effectiveness of the proposed algorithm and demonstrating the winning mechanism in this type of game.}, archive = {J_EAAI}, author = {Liran Zhao and Qinbo Sun and Sihan Xu and Zhaohui Dang}, doi = {10.1016/j.engappai.2025.110231}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110231}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A learning-based algorithm for turn-based orbital pursuit-evasion problem with reaction-time delay}, volume = {145}, year = {2025}, }
• (2025). A multi-scale deep feature memory and recovery network for multi-sensor fault diagnosis in the channel missing scenario. EAAI, 145, 110228. (www)

  @article{ , abstract = {In the field of data-driven intelligent fault diagnosis, the monitoring data provided by single sensor are typically inadequate to reveal the complex states of large-scale equipment comprehensively. Intelligent fault diagnosis techniques based on the fusion of multi-sensor signals have achieved considerable success. Nevertheless, the primary multi-channel fault diagnosis methods based on multi-sensor signals have not yet effectively resolved the problem of sudden sensor failures. Given the challenges mentioned above, we introduce the channel missing scenario to emulate the situation where partial channels are suddenly missing during online inference. To alleviate the impact of missing channels, we propose a multi-scale deep feature memory and recovery network (MDFMR). The feature memory and channel recovery mechanisms of MDFMR can improve the robustness of the model under channel missing scenario. The experiments were conducted on two rotating machinery datasets. Experimental results demonstrated that in the most severe channel missing scenario, conventional multi-channel diagnosis methods become unreliable, while MDFMR maintains a diagnosis accuracy of over 95%.}, archive = {J_EAAI}, author = {Tianao Zhang and Li Jiang and Jie Liu and Xin Zhang and Qing Zhang}, doi = {10.1016/j.engappai.2025.110228}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110228}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-scale deep feature memory and recovery network for multi-sensor fault diagnosis in the channel missing scenario}, volume = {145}, year = {2025}, }
• (2025). Hybrid graph convolutional and deep convolutional networks for enhanced pavement crack detection. EAAI, 145, 110227. (www)

  @article{ , abstract = {Pavement crack detection plays a crucial role in ensuring traffic safety and prolonging service life. Cracks often exhibit irregular curved contours, with their local details submerged within complex background textures. Accurate segmentation of crack contours necessitates a model capable of capturing both long-range correlations and local details. While various methods have been proposed for complex crack detection challenges, most methods in the literature often encounter difficulties when dealing with the dual demands of global correlations and local details. This paper proposes a hybrid model that integrates graph convolutional networks (GCNs) for modeling long-range correlations and deep convolutional networks (DCNs) for capturing local textural details. The proposed model first transforms crack images into graphs using superpixel segmentation, where graph nodes and edges are initialized based on superpixels and their spatial adjacency. A hybrid deep learning architecture, comprising a GCN and an encoder-decoder DCN, is then designed. The GCN learns hidden representations of graph nodes through topology-adaptive convolution and edge message passing, capturing global topological features and long-range spatial dependencies. These features are complementarily enhanced with local spatial features extracted by the encoder-decoder DCN. Experimental results on benchmark pavement crack detection datasets demonstrate that the proposed method achieves competitive detection performance with fewer learnable parameters, accurately capturing both the slender, irregular crack contours and their local textural details. Thus, the method offers a reliable solution for pavement crack detection with improved generalization and interpretability.}, archive = {J_EAAI}, author = {Qingsong Song and Jiashu Tian}, doi = {10.1016/j.engappai.2025.110227}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110227}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid graph convolutional and deep convolutional networks for enhanced pavement crack detection}, volume = {145}, year = {2025}, }
• (2025). Pedestrian re-identification algorithm based on global attention and region ranking. EAAI, 145, 110224. (www)

  @article{ , abstract = {—In practical applications, significant variations in camera parameters and the complexity of shooting environments present substantial challenges for pedestrian re-identification. These challenges include inconsistencies in pedestrian image features across diverse scenes and a limited availability of difficult instances, such as occluded or low-resolution images. To address these issues, we propose a global attention mechanism that fully considers the temporal and spatial consistency of pedestrian information. This mechanism effectively captures regional features that best characterize pedestrian attributes while ensuring that these regions remain temporally and spatially continuous. As a result, robust feature representation is maintained across different scenes, even under occlusion. Furthermore, to overcome difficulties in sample annotation and the ambiguity of boundaries within deep learning-based pedestrian re-identification algorithms, we propose a pedestrian recognition algorithm based on salient region ranking. This algorithm reformulates the identification of salient regions within pedestrian images as a multi-label ranking task, employing ordinal regression to uncover correlations among different category labels, thereby enhancing the robustness of the region saliency classifier. Ablation and comparative experiments conducted on the Market1501, DukeMTMC-reID, and Occluded-Duke datasets demonstrate that our method not only achieves superior recognition performance but also validates its high generalization capability and robustness. This research offers significant theoretical and practical contributions to the further development of deep learning techniques in pedestrian re-identification.}, archive = {J_EAAI}, author = {Feng-Ping An and Jianrong Wang and RuiJun Liu}, doi = {10.1016/j.engappai.2025.110224}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110224}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pedestrian re-identification algorithm based on global attention and region ranking}, volume = {145}, year = {2025}, }
• (2025). Semi-supervised contrastive learning for flotation process monitoring with uncertainty-aware prototype optimization. EAAI, 145, 110222. (www)

  @article{ , abstract = {Froth flotation is a widely employed mineral beneficiation technique, and effective process monitoring is critical for optimizing mineral separation. However, in the industrial process, labeling froth images to create large labeled datasets is both expensive and time-consuming. Semi-supervised deep learning offers a promising solution, but leveraging unlabeled data remains a significant challenge. To this aim, we propose an uncertainty-aware semi-supervised contrastive learning method. Our approach employs a pseudo-labeling module with dropout to generate pseudo labels and estimate uncertainty. Based on the uncertainty scores, the pseudo-labeled data are split into reliable and unreliable sets. A semi-supervised contrastive learning module is developed to exploit pseudo-label information and learn class-aware representations from the reliable set. Additionally, a dynamically weighted consistency learning module is introduced to explore potential classification information in the unreliable data while preventing the model from being misled by low-confidence predictions. Comparison experiments on industrial zinc flotation data show our method achieves 88.12% classification accuracy, surpassing the best alternative by a margin of 5.11%. These results demonstrate that our method generalizes well to unseen test data and outperforms state-of-the-art methods.}, archive = {J_EAAI}, author = {Mingxi Ai and Jin Zhang and Peng Li and Jiande Wu and Zhaohui Tang and Yongfang Xie}, doi = {10.1016/j.engappai.2025.110222}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110222}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised contrastive learning for flotation process monitoring with uncertainty-aware prototype optimization}, volume = {145}, year = {2025}, }
• (2025). Exergy focused optimum solar panel tilt angle determination with improved hybrid model: The case of turkey. EAAI, 145, 110220. (www)

  @article{ , abstract = {The aim of this research is to improve electricity generation efficiency and contribute to reducing energy costs by positioning solar panels at the optimal tilt angle in Turkey, a country with high solar energy potential. To achieve this goal, more advanced methods such as machine learning techniques are preferred instead of traditional methods calculated by mathematical formulae. In a typical Particle Swarm Optimisation (PSO) application, mathematical equations are used as the fitness function, but such an approach can be time, labour, and performance consuming. In this study, a new method called ExtraPSO (Extra Tree with Particle Swarm Optimisation) is developed by using K-Nearest Neighbors, AdaBoost (Adaptive Boosting), Gradient Boosting, Random Forest, ExtraTrees algorithms instead of a mathematical model for the fitness function. As a result of the testing of ExtraPSO, it was observed that a powerful predictive hybrid model was obtained in addition to saving labour and time. The prediction success of the developed ExtraPSO model is determined as 99.0549% with the Accuracy metric. In addition, the optimum tilt angle and maximum exergy amount were determined for Turkey according to the seasons. Considering all seasons and annual average for Turkey, the maximum exergy amount that can be obtained for a solar panel placed at the optimum tilt angle is determined as 3.871 W/m 2 , 5.713 W/m 2 , 7.048 W/m 2 , 5.272 W/m 2 and 5.285 W/m 2 for winter, spring, summer, fall and annual average, respectively.}, archive = {J_EAAI}, author = {Remzi Gürfidan and Fatih Yiğit and Ahmet Kabul}, doi = {10.1016/j.engappai.2025.110220}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110220}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exergy focused optimum solar panel tilt angle determination with improved hybrid model: The case of turkey}, volume = {145}, year = {2025}, }
• (2025). Hybrid traffic flow prediction model for emergency scenarios with scarce historical data. EAAI, 145, 110219. (www)

  @article{ , abstract = {Accurate traffic volume prediction under emergency conditions is critical for enhancing traffic control, disaster response, and route optimization for both government agencies and individual travelers. However, current prediction models often assume homogeneity between historical and future data, rendering them inadequate for anticipating unprecedented traffic disruptions during emergencies. To address this limitation, this study introduces a hybrid predictive model that integrates Prophet and DeepAR (Probabilistic Forecasting with Autoregressive Recurrent Networks) to improve traffic volume forecasting in such scenarios. Prophet captures long-term trends, seasonal variations, and holiday effects, while DeepAR is adept at modeling abrupt nonlinear changes induced by emergencies. This hybrid framework synergizes the strengths of both models, facilitating adaptive learning of anomalous patterns and striking a robust balance between regular traffic dynamics and emergency-induced disturbances. Through modular decomposition, integration of exogenous factors, and dynamic weight adjustments, the hybrid model offers enhanced flexibility and predictive accuracy compared to existing models, particularly under unprecedented conditions. The hybrid model was validated on highways across Luxembourg, Heilongjiang Province, and Minnesota. Results showed the hybrid model outperformed its components and other competing models under different emergency scenarios. In the Luxembourg pandemic scenario, the model reduced mean absolute error (MAE) by 73.60%, root mean square error (RMSE) by 74.22%, and coefficient of variation (CV) by 31.52%, with a 98.55% improvement in R 2 . Comparable enhancements were achieved in Heilongjiang's snowstorms and Minnesota's adverse weather. These results underscore the robustness of the hybrid model and its potential for real-time traffic management in emergency situations.}, archive = {J_EAAI}, author = {Xueyi Gao and Yusheng Ci and Kum Fai Yuen and Lina Wu and Ruimin Li}, doi = {10.1016/j.engappai.2025.110219}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110219}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid traffic flow prediction model for emergency scenarios with scarce historical data}, volume = {145}, year = {2025}, }
• (2025). A hybrid prediction model for marine wind speed considering internal temporal features recombination and external variables association. EAAI, 145, 110218. (www)

  @article{ , abstract = {Marine wind exists widely as a major environmental disturbance in the process of marine resource exploration. Accurate short-term prediction of marine wind speed can ensure the safety of offshore exploration operations. The current wind speed prediction model has prediction lag, and lacks consideration of chaotic characteristics and the influence of external variables. Analyzing the chaotic time series characteristics of marine wind speed and their causative correlations, a hybrid prediction model for marine wind speed is presented in this paper, which incorporates the recombination of internal temporal features and external variables association. Firstly, the marine wind speed is decomposed using the improved complete ensemble empirical mode decomposition with adaptive noise, with features recombined based on a comprehensive evaluation of multiple entropies and hierarchical clustering. Subsequently, for the chaotic components characterized by strong randomness and nonlinearity, successive variational mode decomposition is used for a secondary decomposition, with chaotic reconstruction based on their recurrence plot characteristics. Then, a hybrid prediction model for marine wind speed considering internal temporal features recombination and external variables association is proposed. Finally, experiments are performed by using actual data from a marine buoy. The result shows that the hybrid prediction model can accurately predict the marine wind speed for the next time step, providing advanced environmental perception for marine resource exploration.}, archive = {J_EAAI}, author = {Haoxian Wen and Sheng Du and Chengda Lu and Yawu Wang and Min Wu and Weihua Cao}, doi = {10.1016/j.engappai.2025.110218}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110218}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid prediction model for marine wind speed considering internal temporal features recombination and external variables association}, volume = {145}, year = {2025}, }
• (2025). Deep learning-based modelling of polyvinyl chloride tube-confined concrete columns under different load eccentricities. EAAI, 145, 110217. (www)

  @article{ , abstract = {This study presents a deep learning-based framework for predicting the load-carrying capacity of polyvinyl chloride (PVC) tube-confined concrete columns under various loading conditions. A comprehensive dataset of 200 samples was generated using finite element modeling, incorporating key parameters such as PVC tube thickness, concrete strength, and load eccentricity. Several machine learning algorithms, including Linear Regression (LR), Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), and a novel hybrid Transformer-Convolutional Neural Network (Transformer-CNN) model, were employed for the prediction task. The results demonstrate that the proposed Transformer-CNN model outperforms traditional methods, achieving the lowest root mean squared error of 27.15 kN and the highest coefficient of determination value of 0.9875. The model's robustness was further validated using cross-validation techniques, ensuring its reliability for practical applications. To facilitate usability, a Python-based graphical user interface (GUI) was developed, enabling engineers to apply the model efficiently in real-world scenarios. This study highlights the potential of deep learning in advancing the design and analysis of PVC-confined concrete columns, offering a more accurate and efficient alternative to conventional methods.}, archive = {J_EAAI}, author = {Li Shang and Haytham F. Isleem and Mostafa M. Alsaadawi}, doi = {10.1016/j.engappai.2025.110217}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110217}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning-based modelling of polyvinyl chloride tube-confined concrete columns under different load eccentricities}, volume = {145}, year = {2025}, }
• (2025). Retentive network-based time series anomaly detection in cyber-physical systems. EAAI, 145, 110215. (www)

  @article{ , abstract = {Time series data are ubiquitous in the operation of cyber-physical systems (CPS), encompassing network traffic data, sensor measurements, and other relevant data streams. Intelligent anomaly detection methods are crucial for identifying deviations in these time series, which can enhance system maintainability and reliability. In this study, we propose an unsupervised reconstruction-based anomaly detection method utilizing retentive network (RetNet), a novel variant of Transformer. A one-dimensional convolutional neural network is employed to map the raw time series into the RetNet model space, analogous to word embedding techniques in natural language processing. The rotary position embedding introduced in RetNet can simultaneously incorporate absolute and relative positional information, which is beneficial for modeling temporal dependency in time series. Meanwhile, the multi-scale retention mechanism of RetNet facilitates the learning of informative representations of the dominant normal patterns from training data. The observed anomalies result in larger reconstruction errors, which are subsequently detected by the peaks-over-threshold (POT) method using a dynamic threshold. We evaluate the proposed method on four benchmark datasets. Experimental results demonstrate that the proposed method outperforms the best baseline by 4.07% in terms of the F1 score.}, archive = {J_EAAI}, author = {Zhaoyi Min and Qianqian Xiao and Muhammad Abbas and Duanjin Zhang}, doi = {10.1016/j.engappai.2025.110215}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110215}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Retentive network-based time series anomaly detection in cyber-physical systems}, volume = {145}, year = {2025}, }
• (2025). An improved you only look once model for the multi-scale steel surface defect detection with multi-level alignment and cross-layer redistribution features. EAAI, 145, 110214. (www)

  @article{ , abstract = {Steel surface defects involve a wide variety of sizes and irregular shapes. The performance of the detection models depends greatly on the effective extraction of the cross-scale features. The sequential feature fusion in the traditional Feature Pyramid Network (FPN) is subject to an information loss during the aggregation and transition across shallow and deep features. The classification and localization of steel surface defects using common You Only Look Once (YOLO) detection model may lead to sub-optimal performance when they are involved with the multi-scale objects. To realize a complete shallow-deep feature representation for different scale steel surface defects, an aggregation-redistribution network is introduced into the YOLO detection model to aggregate and refine features across different levels. In the aggregation sub-network, a Multi-level Alignment Module (MAM) is adopted to address the feature misalignment in FPN by aligning the feature maps from level-wise extractors. Therein, the scale deviation at pixels is compensated by the multiple parallel dilated convolutions. Meanwhile, in the redistribution sub-network, a Fusion-Redistribution Module (FRM) is constructed to impose the global information on the fused multi-level features to steer cross-layerly the generation of prediction feature maps for YOLO. The global feature provides a semantic adaptive weight and serves as a complement on the multi-level features through an attention mechanism. Finally, by incorporating the aggregation-redistribution network into YOLOv5, an improved YOLO detection model, i.e., Aggregation-Redistribution YOLO (ARYOLO), is derived for the steel surface defects. The validation results indicate that ARYOLO has achieved a rather satisfactory detection performance for the defects with large variations in scale and shape. It provides a mean average precision of 80.7% on the North Eastern University Surface Defect Dataset (NEU-DET) and 71.2% on the Global 10 Class Metallic Surface Defect Dataset (GC10-DET), giving a great potential in detection and localization tasks.}, archive = {J_EAAI}, author = {Jianhang Huang and Xinliang Zhang and Lijie Jia and Yitian Zhou}, doi = {10.1016/j.engappai.2025.110214}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110214}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An improved you only look once model for the multi-scale steel surface defect detection with multi-level alignment and cross-layer redistribution features}, volume = {145}, year = {2025}, }
• (2025). Human gait recognition based on frontal-view sequence using discriminative optical flow feature representations and learning. EAAI, 145, 110213. (www)

  @article{ , abstract = {Existing gait recognition methods usually employ lateral-view human silhouettes and require a large data collection field, which is not feasible in real-world surveillance applications. To facilitate the practical application of gait recognition, a new frontal-view gait recognition method based on discriminative optical flow feature representations and learning is proposed in this paper. Lower limb region is extracted as the region of interest, and optical flows of different local subregions are calculated for each frame based on original video data. We propose to extract a new frontal-view gait representation, called local fuzzy histogram energy image (LFHEI), which reflects the underlying spatial–temporal dynamics characteristics of corresponding body parts to build a robust gait recognition system. Finally, a new feature learning scheme is proposed, in which LFHEI descriptors are fed into the deep metric learning algorithms to learn underlying nonlinear mapping and project LFHEI descriptors into a discriminative embedding space. Gait recognition can be achieved by using the nearest-neighbor search between a query embedding and candidate embeddings. Experiments on CASIA-A and CASIA-B gait databases show that promising results can be achieved for frontal-view gait recognition. Our proposed method yields average accuracies of 95.0% and 89.9% on CASIA-A and CASIA-B gait databases, with improvements of 4.0% and 1.9% compared to state-of-the-art frontal-view gait recognition techniques. Our proposed method is feasible for frontal-view gait recognition, and has strong robustness and generalization ability to different types of walking conditions and environments.}, archive = {J_EAAI}, author = {Muqing Deng and Yi Zou and Zhi Zeng and Yanjiao Wang and Xiaoreng Feng and Yuan Liu}, doi = {10.1016/j.engappai.2025.110213}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110213}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Human gait recognition based on frontal-view sequence using discriminative optical flow feature representations and learning}, volume = {145}, year = {2025}, }
• (2025). Extended combined method based on cumulative prospect theory with type-2 neutrosophic fuzzy numbers and their application to the prioritization of inefficient land redevelopment. EAAI, 145, 110212. (www)

  @article{ , abstract = {As urbanization progresses rapidly, the expansion of cities has led to increased fragmentation of land use and inefficiencies, exacerbating the imbalance between the supply of and demand for land resources. In this context, the challenge of inefficient land redevelopment has emerged as a pressing issue that requires urgent resolution. Established Multiple-attribute group decision-making (MAGDM) methods have significant limitations in addressing such problems, especially failing to adequately account for the ambiguities and uncertainties inherent in real-world scenarios. Moreover, many of these methods ignore the psychological behaviors of decision makers (DMs) in the face of risks and benefits, or rely on a single method or directly aggregate evaluation information to determine weights. To address the above limitations, this article introduces Cumulative Prospect Theory (CPT)-based Multi-attributive boundary approximation region comparison (MABAC) method with the Type-2 Neutrosophic number (T2NN) for the prioritization of alternatives. The method effectively captures the bounded rationality of DMs through CPT while accurately representing the ambiguity and uncertainty of evaluation information using T2NN. Additionally, we integrate the Game Theory (GT)-based Analytic Hierarchy Process (AHP)-Criteria Importance Though Intercrieria Correlation (CRITIC) combination weighting method to determine attribute weights more scientifically and comprehensively. Finally, we validate the proposed method through a numerical example, conduct sensitivity analysis by changing weights, and perform comparative analysis with three traditional decision-making methods. The results demonstrate the method has significant reliability and effectiveness in addressing MAGDM problems characterized by ambiguity and uncertainty, offering new insights and methods for complex decision-making problems.}, archive = {J_EAAI}, author = {Yangjingyu Zhang and Yuming Zhu}, doi = {10.1016/j.engappai.2025.110212}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110212}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Extended combined method based on cumulative prospect theory with type-2 neutrosophic fuzzy numbers and their application to the prioritization of inefficient land redevelopment}, volume = {145}, year = {2025}, }
• (2025). U-shaped disassembly line balancing problem under interval type-2 trapezoidal fuzzy set: Modeling and solution method. EAAI, 145, 110211. (www)

  @article{ , abstract = {With the accelerated upgrading of mechanical products, the impact of end-of-life (EOL) products on the environment is gradually increasing. Therefore, disassembly and recycling have gradually become hot spots for environmental protection and sustainable manufacturing. The U-shaped disassembly line has better advantages over the linear disassembly line in terms of disassembly task allocation, workstation balancing and disassembly line layout, which is challenging research. when the EOL products seriously deform and fail, the disassembly time is difficult to predict, which will cause disassembly failure and the disassembly line shutdown will affect the production. Moreover, If the uncertainty in the U-shaped disassembly line balancing problem (U-DLBP) are not considered, which ultimately leads to the result deviation from the actual situation. Therefore, this paper proposes an U-DLBP framework to deal with the uncertain U-DLBP and quantify the disassembly risk. an interval type-2 trapezoidal fuzzy set (IT2TFS) was introduced as a modelling tool for uncertain analysis, risk quantification, and improving the accuracy in U-DLBP. An IT2TFS-based U-DLBP mathematical model combining average smoothness, disassembly cost and disassembly energy consumption was constructed. In response to the complexity of the problem, an improved fuzzy multi-objective particle swarm optimization algorithm (FMOPSO) inspired by genetic operators was developed to avoid premature convergence and improving global optimization ability. According to the case study of EOL printer, the performance of the improved FMOPSO algorithm is significantly better than other three algorithms in stability and convergence when solving a practical case. The results confirm that this research provides an effective tool to solve the problem of the IT2TFS-based U-DLBP and some reference gist for designers.}, archive = {J_EAAI}, author = {Honghao Zhang and Zhongwei Huang and Danqi Wang and Guangdong Tian and Wenjie Wang}, doi = {10.1016/j.engappai.2025.110211}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110211}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {U-shaped disassembly line balancing problem under interval type-2 trapezoidal fuzzy set: Modeling and solution method}, volume = {145}, year = {2025}, }
• (2025). Exploratory literature review and scientometric analysis of artificial intelligence applied to geopolymeric materials. EAAI, 145, 110210. (www)

  @article{ , abstract = {The goal of this work is to present and analyze studies that link artificial intelligence (AI) and geopolymer composites through an exploratory literature review. The systematic search comprised articles in the Scopus database, from 2000 to 2023. The results from 48 articles show that AI has been applied mainly to predict compressive strength and slump. Comparing the AI techniques to their R 2 , no clear trend was observed, i.e., there was no overall best algorithm; and their performance was not directly related to database size. Still, Random Forest obtained superior results in articles that compared multiple techniques. Some of the main gaps identified are the lack of studies with geopolymers synthesized in acidic environments or with waste-based or nanomaterials; and that AI has not been largely applied to properties besides compressive strength (e.g., fire resistance, durability, porosity, and thermal properties). Regarding transparency, authors have seldom publicized their codes, impairing proper review of the models. In conclusion, if properly and ethically adopted, AI is a promising tool for the development of geopolymer composites, which can optimize time and employee resources; and many applications remain unexplored.}, archive = {J_EAAI}, author = {Aldo Ribeiro de Carvalho and Romário Parreira Pita and Thaís Mayra de Oliveira and Guilherme Jorge Brigolini Silva and Julia Castro Mendes}, doi = {10.1016/j.engappai.2025.110210}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110210}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploratory literature review and scientometric analysis of artificial intelligence applied to geopolymeric materials}, volume = {145}, year = {2025}, }
• (2025). Research on mathematical model of green behavior propagation based on big data statistical analysis and artificial intelligence algorithm. EAAI, 145, 110208. (www)

  @article{ , abstract = {The mechanism of green behavior propagation is examined by creating a reaction–diffusion model to support the development of the green economy. We analyze the complex dynamics of green behavior propagation on continuous and complex networks based on Turing pattern theory. Further, we construct the evaluation indicator Normalized Positive Prevalent Area ( N P P A ) and set the largest value as the ideal target. To achieve the ideal propagation state of green behavior as quickly as possible, the parameter identification technique based on artificial intelligence algorithms is employed. Specifically, the projected gradient descent (PGD) algorithm and the Barzilai–Borwein (BB) algorithm are applied to design optimal paths with minimal loss for different network structures, respectively. It is found that the results of the BA network and lattice network with 8 neighboring nodes under the BB algorithm are better. In terms of model application, we give the data released by the Ministry of Ecology and Environment of the People’s Republic of China to provide an optimization path for the actual dissemination of green behavior. Next, this work projects and optimizes the carbon emission intensity of 30 provinces in China in 2025 and 2030 in response to the Chinese dual carbon policy. In addition to this, the parameter identification technique based on the artificial intelligence algorithm can also be applied to image recognition with better results.}, archive = {J_EAAI}, author = {Yuxuan Pan and Shuling Shen and Linhe Zhu}, doi = {10.1016/j.engappai.2025.110208}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110208}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on mathematical model of green behavior propagation based on big data statistical analysis and artificial intelligence algorithm}, volume = {145}, year = {2025}, }
• (2025). Event-triggered neuro-optimal fault tolerant control for uncertain macro–micro composite stage system with actuator faults. EAAI, 145, 110207. (www)

  @article{ , abstract = {This paper develops an event-triggered neuro-optimal fault tolerant control (ETNOFTC) method for an uncertain macro–micro composite stage (MMCS) system with actuator faults. By regarding the MMCS system as a two-player system, its control problem is formulated as a nonzero-sum game which aims to cooperatively stabilize the system with the individual cost functions of all players. To stabilize the faulty MMCS system, we develop an ETNOFTC scheme which consists of two parts, i.e., integral sliding mode control and adaptive dynamic programming-based event-triggered optimal control. An integral sliding mode control method is developed to eliminate the effects of actuator faults and uncertainties, and thus a system equivalent to the nominal system is obtained. Then, by employing the critic-only architecture, the event-triggered Hamilton–Jacobi equation is solved by approximating the coupled cost function to obtain the approximate event-triggered optimal control policy of each player. Furthermore, we prove that the closed-loop MMCS system is asymptotically stable by using the Lyapunov’s direct method. Finally, the effectiveness of the developed ETNOFTC scheme is verified.}, archive = {J_EAAI}, author = {Shunchao Zhang and Bo Zhao and Yongwei Zhang}, doi = {10.1016/j.engappai.2025.110207}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110207}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Event-triggered neuro-optimal fault tolerant control for uncertain macro–micro composite stage system with actuator faults}, volume = {145}, year = {2025}, }
• (2025). Knowledge-based real-time scheduling for gas supply network using cooperative multi-agent reinforcement learning and predictive functional range control. EAAI, 145, 110206. (www)

  @article{ , abstract = {In steel enterprises, the real-time scheduling optimization of the gas supply network can provide strong support for stabilizing production and enhancing economic benefits. Due to the coupling of multiple gas/liquid products and numerous units, centralized scheduling methods require large training and coordination overhead. Consequently, the real-time scheduling problem of a multi-product gas supply network is modeled under the cooperative multi-agent reinforcement learning (MARL) architecture, which makes the scheduling strategy of each unit keep the same optimization goal. Decentralized execution mode reduces the computing cost and information exchange compared with centralized execution mode. Different from adding constraint penalty terms as soft constraints, a constraint monitor module is designed by utilizing the process knowledge, ensuring the various production constraints are satisfied. This strategy can reduce trail-and-error costs, making it more conducive to industrial safety. To deal with the unknown disturbance in production (typically gas leakage), a predictive functional range control (PFRC) algorithm is then developed to modify the future gas demand. Finally, case studies are carried out on a real-world gas supply network to verify the performance of the proposed method.}, archive = {J_EAAI}, author = {Pengwei Zhou and Zuhua Xu and Jiakun Fang and Jun Zhao and Chunyue Song and Zhijiang Shao}, doi = {10.1016/j.engappai.2025.110206}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110206}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Knowledge-based real-time scheduling for gas supply network using cooperative multi-agent reinforcement learning and predictive functional range control}, volume = {145}, year = {2025}, }
• (2025). Enhancing healthcare supply chains: A comprehensive evaluation of lean, agile, resilient and green paradigms. EAAI, 145, 110204. (www)

  @article{ , abstract = {The COVID-19 outbreak highlighted the need for efficient and resilient healthcare supply chains (HSCs) that can handle surges in demand and infectious waste. Many studies have evaluated individual lean, agile, resilient, or green (LARG) paradigms in HSCs, but a comprehensive evaluation of all four together is lacking. We present the first attempt to evaluate LARG parameters in HSC performance, using integrated methods including Pythagorean Fuzzy Decision-making Trial and Evaluation Laboratory (PF-DEMATEL), interpretive structural modeling (ISM), and Bayesian network (BN). We identified the most important contributing factors of the LARG HSC through literature review analysis, which was modified and validated through expert interviews. Cause-and-effect diagrams and hierarchical network models of LARG variables were developed using PF-DEMATEL and ISM methods to determine the interdependent relationships between factors. Mapping the outcomes to the BN method provided a quantification of the intensity of coupling relationships between variables. The key factors identified include Consistent Medical Service (CMS), Flexibility (F), Emergency Medical Services in Distressing Situations (EMSDS), and Medical Equipment Maintenance Program (MEMP). The study’s results, based on data from five hospitals, indicate a 62% probability of achieving high-level LARG performance. The proposed model provides managers with the most critical parameters of the LARG HSC and enhances their understanding of the relationships between these parameters. The proposed model serves as a practical guide for improving healthcare service delivery and meeting patient needs.}, archive = {J_EAAI}, author = {Salman Nazari-Shirkouhi and Samirasadat Samadi}, doi = {10.1016/j.engappai.2025.110204}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110204}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing healthcare supply chains: A comprehensive evaluation of lean, agile, resilient and green paradigms}, volume = {145}, year = {2025}, }
• (2025). Simultaneously local and global contrastive learning of graph representations. EAAI, 145, 110203. (www)

  @article{ , abstract = {Graph Representation Learning (GRL) is crucial for understanding complex graph-structured data. This is a hot topic in Artificial Intelligence research, particularly due to its wide applicability in various engineering contexts, such as social network analysis, bioinformatics, and recommender systems. However, existing methods often overlook the multi-scale properties of graphs and the integration of semantic information, which limits the ability to capture the deeper information of the data. To address this issue, we propose a novel approach, Simultaneously Local and Global Contrastive Learning of Graph Representations (SLGGCL), for the graph-node classification problem. Specifically, SLGGCL generates subgraphs for local views and diffusion graphs for global views by leveraging a combination of strategies, including edge perturbation, attribute masking, and graph diffusion. Then, we apply a contrastive learning model to capture the information among multi-scale graphs to improve the classification accuracy. In addition, experimental results on multiple real-world datasets show that SLGGCL outperforms ten state-of-the-art methods, achieving a 1.01% improvement in accuracy on Cora compared to the best-performing method.}, archive = {J_EAAI}, author = {Shuai An and Binsheng Hong and Zhaori Guo and Shunzhi Zhu and Kaibiao Lin and Fan Yang}, doi = {10.1016/j.engappai.2025.110203}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110203}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Simultaneously local and global contrastive learning of graph representations}, volume = {145}, year = {2025}, }
• (2025). Circular spherical fuzzy aggregation operators: A case study of risk assessments on industry expansion. EAAI, 145, 110202. (www)

  @article{ , abstract = {The circular spherical fuzzy set (C-SFS), an advanced concept that integrates the features of picture fuzzy sets (PFS) and spherical fuzzy sets, is presented in this work. Previous models often lack the flexibility and capacity to effectively handle intricate uncertainties and ambiguous data due to limited parameters and constrained frameworks. To address these limitations, the C-SFS model enhances the representation and analysis of imprecise data by introducing additional parameters and a more robust structure. The study aims to provide a comprehensive framework for managing ambiguity and uncertainty in decision-making. Specifically, we explore various aggregation operators within the C-SFS framework and thoroughly evaluate their mathematical properties, such as norms and co-norms. The practical applicability of these operators is illustrated through a case study on risk assessment in manufacturing, demonstrating their ability to integrate multiple factors and resolve complex uncertainties. Comparative analysis further reveals that the C-SFS framework outperforms traditional approaches in ambiguity handling and decision-making efficacy. These findings contribute significantly to fuzzy set theory by offering an innovative method for multi-criteria decision-making with broad potential applications. The study establishes a strong foundation for future research and practical implementations of C-SFS in addressing real-world challenges involving uncertainty and complexity.}, archive = {J_EAAI}, author = {Shahzaib Ashraf and Muhammad Shakir Chohan}, doi = {10.1016/j.engappai.2025.110202}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110202}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Circular spherical fuzzy aggregation operators: A case study of risk assessments on industry expansion}, volume = {145}, year = {2025}, }
• (2025). Entropy-based fuzzy 1norm twin random vector functional link networks for binary class imbalance learning. EAAI, 145, 110201. (www)

  @article{ , abstract = {Real-world datasets may contain dissimilar numbers of positive and negative class samples. Class imbalance is a common issue in classification tasks, where one class (the majority) significantly outnumbers another (the minority). This can lead to conventional models failing to predict the minority class, which is more important accurately. Hence, in this work we have proposed a model with high generalization capability on imbalanced datasets to address this problem. The proposed Entropy-based Fuzzy 1-norm Twin Random Vector Functional Link Network (EFTRVFL 1norm ) solves the binary class imbalance learning problem. In the training phase, this model evaluates entropy values based on the concept of k nearest neighbour of two samples that belong to two distinct classes. To improve model's performance, samples with higher class certainty or lower entropy values are associated with larger fuzzy memberships. Entropy-driven fuzzy membership pays more consideration to the samples with higher class probability by providing relatively large fuzzy members to majority samples compared to minority class samples. Moreover, the model becomes robust and generalized in the presence of a 1-norm which generates sparse output vectors. The performance analysis is undertaken based on two evaluators: Area under the curve and Geometric mean by comparing the results with some related models such as twin support vector machine, kernel ridge regression (KRR), affinity-based fuzzy KRR (AFKRR), random vector functional link layer network (RVFL), intuitionistic fuzzy RVFL classifier (IFRVFLC) and 1-norm twin RVFL (TRVFL 1norm ). The comparing results demonstrate the model's classification efficiency for computing binary imbalanced classification problems.}, archive = {J_EAAI}, author = {Chittabarni Sarkar and Deepak Gupta and Barenya Bikash Hazarika and Rajat Subhra Goswami}, doi = {10.1016/j.engappai.2025.110201}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110201}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Entropy-based fuzzy 1norm twin random vector functional link networks for binary class imbalance learning}, volume = {145}, year = {2025}, }
• (2025). Thermal behavior prediction of the spindle-bearing system based on the adaptive thermal network modeling method. EAAI, 145, 110199. (www)

  @article{ , abstract = {Differences in comprehension of the heat transfer mechanism in spindle-bearing systems result in diverse thermal network modeling approaches. This leads to variations in simulation results of thermal prediction. To eliminate this epistemic uncertainty and determine the optimal thermal network model for the spindle-bearing system under various operating conditions, this study proposes an adaptive modeling method. Firstly, common thermal network models based on different mechanistic understandings are briefly introduced. Next, a convolutional neural network (CNN) surrogate model based on incremental learning is proposed to accurately approximate numerous thermal networks with limited simulations. Furthermore, considering that the spindle-bearing system comprises a multi-support structure, the grey relational analysis is utilized to mitigate the interaction effects of different support bearings on thermal characteristic analysis. Moreover, based on the CNN surrogate model, a two-step optimization method is proposed to obtain a thermal network model which balances accuracy and structural simplicity. Finally, the effectiveness and accuracy of the optimal thermal network model are validated through the thermal characteristic experiments of the spindle system under various operating conditions.}, archive = {J_EAAI}, author = {Ziquan Zhan and Shaoke Wan and Xiaohu Li and Bin Fang}, doi = {10.1016/j.engappai.2025.110199}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110199}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Thermal behavior prediction of the spindle-bearing system based on the adaptive thermal network modeling method}, volume = {145}, year = {2025}, }
• (2025). Essential anchor graph learning for incomplete multi-view clustering. EAAI, 145, 110194. (www)

  @article{ , abstract = {Incomplete multi-view clustering focuses on exploring complementary information from diverse views when some views are missing. Recently, many anchor-based incomplete multi-view clustering methods have been presented. However, most existing approaches neglect the recovery of missing view information and the exploration of complementary information in anchor graph learning. In this work, we present a novel incomplete multi-view clustering method based on the anchor graph framework to resolve these problems. Concretely, the missing instances in the incomplete data are recovered by the anchor graph learning, while the recovered complete data matrix performs the learning of anchors and the construction of anchor graphs in a joint framework, which makes the learned anchors more discriminative. Moreover, we apply the orthogonal transformation on the anchor graph, which can mine the complementary and consensus information from different views. The clustering results on several benchmark datasets substantiate the effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Peng Song and Jinshuai Mu and Yuanbo Cheng and Zhaohu Liu and Wenming Zheng}, doi = {10.1016/j.engappai.2025.110194}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110194}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Essential anchor graph learning for incomplete multi-view clustering}, volume = {145}, year = {2025}, }
• (2025). Adaptive control method for ship curved hybrid laser-arc welding through lightweight encoder-decoder architecture. EAAI, 145, 110193. (www)

  @article{ , abstract = {The proliferation of three-dimensional welding seams spans automobile, aerospace, pipeline, and shipbuilding. An advancement in curved seam welding is the evolution towards hybrid laser-arc welding with decreased heat input and defects, particularly in ship-curved blocks characterized by susceptibility to deformation and processing errors. These blocks are crucial for ship quality but require precise weld position and orientation control. While existing seam tracking methods recall the region of interest or segment clean profile regions under weld interference and reprocess features based on the predicted region. In this paper, we introduce a deep-learning architecture that can regress features with an accuracy of 3.41 pixels for feature points and 1.47 pixels for center lines. The adaptive system performs weld position and parameter adjustment in which a hybrid laser-arc welding system integrated with a single structured-light camera acquires images during welding, and a computer integrated with the proposed model serves to regress feature points for seam tracking, calculate seam coordinate systems for pose adaption, and identify gaps for wire feed speed adjustment. Two experiments were conducted to demonstrate the adaptive system: the first involved butt and corner welding of a 1-m-long, 8-mm-thick curved plate with a center angle of 12°, and the second focused on welding a curved block with a length of 3m. The proposed method sets the stage for an integrated implementation of seam tracking, pose adaption, and parameter adjustment.}, archive = {J_EAAI}, author = {Sen Zhang and Yansong Zhang and Haoyuan Bai}, doi = {10.1016/j.engappai.2025.110193}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110193}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive control method for ship curved hybrid laser-arc welding through lightweight encoder-decoder architecture}, volume = {145}, year = {2025}, }
• (2025). A framework for compression index prediction considering geographical information and feature missing. EAAI, 145, 110192. (www)

  @article{ , abstract = {Using machine learning methods to predict the soil compression index ( C c ) has become a research hotspot due to its efficiency in computation and automatic data pattern mining. However, existing methods often overlook missing input features and fail to incorporate geographical information of the soil location, which limits data utilization efficiency and accuracy enhancement. To address these shortcomings, we propose a framework for C c prediction that considers geographical information and complements missing features. Firstly, geographical information, namely latitude ( Lat ), longitude ( Lon ), annual average maximum temperature ( T max ), and annual average minimum temperature ( T min ), is included according to the soil sample's collection site, as supplementary features in addition to liquid limit ( LL ), plasticity index ( PI ), natural water content ( w ), and initial void ratio ( e 0 ) as adopted in the past studies. Subsequently, a missing feature completion method called “Mutual Prediction” is introduced to reduce the number of scenarios needing consideration for missing features. Finally, we construct a machine learning-based model named “Trunk-Branch” (TB model) to achieve C c prediction. The results show: (1) Incorporating geographical information increases the coefficient of determination ( R 2 ) of C c from 0.751 to 0.901, with Lon contributing the most. (2) “Mutual Prediction” reduces the scenarios needing consideration for missing features from 15 to 3 and enhances the R 2 from 0.746 to 0.811. (3) Corresponding to these three missing scenarios, the TB model achieves R 2 values of 0.612, 0.720, and 0.901, respectively.}, archive = {J_EAAI}, author = {Yuan-en Pang and Xu Li and Jin-peng Xin and Jia-ting Wang and Hao Cai}, doi = {10.1016/j.engappai.2025.110192}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110192}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A framework for compression index prediction considering geographical information and feature missing}, volume = {145}, year = {2025}, }
• (2025). Granular ball-based fuzzy multineighborhood rough set for feature selection via label enhancement. EAAI, 145, 110191. (www)

  @article{ , abstract = {Recently, in artificial intelligence, existing feature selection schemes have usually neglected the different distributions of features and description degrees of labels with respect to samples without considering label enhancement. In addition, some granular ball models do not compute the similarity of labels between samples, which leads to division errors in granular balls. To overcome these drawbacks, this paper studies granular ball-based adaptive fuzzy multineighborhood rough set and develops multilabel feature selection via label enhancement. First, the overall label set is developed on the basis of the description degree of label to granular ball. To develop the similarity of samples under the label space, the Jaccard similarity of labels is designed for the purity of granular balls, and then the multilabel samples with similar labels are clustered into the same granular ball. Second, to adapt the distribution of different features, the maximum and minimum feature standard deviations are applied to construct an adaptive fuzzy multineighborhood radius set. Adaptive fuzzy multineighborhood granules of samples can be extracted from adaptive fuzzy neighborhood classes to construct upper and lower approximations, and then granular ball-based adaptive fuzzy multineighborhood rough set can be established. Third, the fuzzy similarity of samples relative to the label set is considered to provide a label enhancement strategy to compute the label proportion, the fuzzy decision similarity under the label space is defined, and several label enhancement-based uncertainty measures are provided. By combining the dependence degree and knowledge granularity, a mixed measure is proposed to compute feature scores. Finally, the feature score vector of each granular ball is synthesized to select the optimal feature subset; then, a feature selection methodology via label enhancement is proposed. Experiments show that this constructed algorithm is effective on 14 multilabel datasets.}, archive = {J_EAAI}, author = {Lin Sun and Wenjuan Du and Weiping Ding and Qian Long and Jiucheng Xu}, doi = {10.1016/j.engappai.2025.110191}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110191}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Granular ball-based fuzzy multineighborhood rough set for feature selection via label enhancement}, volume = {145}, year = {2025}, }
• (2025). Bipolar augmentation: Lightweight artificial intelligence model training for anomaly detection under color cast scenarios. EAAI, 145, 110184. (www)

  @article{ , abstract = {Anomaly detection is an important research field for Industrial Internet of Things (IIoT). The factory has to detect the abnormal events during the manufacturing process. Deep learning is becoming a core technology for anomaly detection. The real world applications have many unwanted and irreversible data corruptions. Building a robust AI (Artificial Intelligence) model to against these data corruptions is crucial for smart manufacturing. In order to build an anomaly detection model, we have to collect the images for model training. However, the real world factories would have various illumination conditions and the white balance function of camera may not be well configured. Therefore, the anomaly detection model should be robust to the color cast images. Although we can expand the training data by collecting all the possible color cast images, the increased efforts might not be affordable. Hence, we proposed a lightweight data augmentation approach to develop a robust AI model. In order to investigate the research issues from color cast images, we build a color cast dataset for anomaly detection. Meanwhile, our proposed bipolar augmentation is able to provide similar performance but requires much less training efforts. The performance gap between our proposed method and the ideal but costly training process is around 0 . 75 % ∼ 1 . 8 % , while the training time acceleration of our method is up to 15 . 15 × or even 30 . 63 × . We expected this work can inspire more research efforts to address the color cast issues in anomaly detection.}, archive = {J_EAAI}, author = {Yu-Hsun Lin and Yu-Mo Lin}, doi = {10.1016/j.engappai.2025.110184}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110184}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bipolar augmentation: Lightweight artificial intelligence model training for anomaly detection under color cast scenarios}, volume = {145}, year = {2025}, }
• (2025). Estimating room acoustic descriptors from bag-of-vectors representation with transformers. EAAI, 145, 110183. (www)

  @article{ , abstract = {In this paper, we propose a novel deep learning method for room acoustic descriptor estimation. Certain descriptors are highly important in assessing acoustic quality, therefore estimating them during the planning phase is a crucial part of designing indoor spaces. Traditional approaches rely on either computationally expensive numerical methods, or statistical formulae with insufficient accuracy. Our solution is FRAPPE (fast room acoustic prediction and parameter estimation), which applies lightweight transformer-based neural networks to estimate acoustic descriptors in rectangular rooms, utilizing a “bag-of-vectors” representation that is capable of capturing diverse interior designs. We employ transformers without positional encoding, highlighting the broad applicability of the architecture outside of traditional domains. FRAPPE achieves high accuracy and operates at near-instant speed, providing a better cost–accuracy balance than either ray tracing methods or empirical formulae. It is the first transformer-based approach that is applicable in the design phase, and it offers a more general deep learning solution for acoustic descriptor estimation than any prior methods. The accuracy, inference speed and versatility of FRAPPE makes it a valuable innovation for architectural design, supporting better decisions during the early stages of room planning.}, archive = {J_EAAI}, author = {Bence Bakos and Gábor Hidy and Bálint Csanády and Csaba Huszty and András Lukács}, doi = {10.1016/j.engappai.2025.110183}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110183}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Estimating room acoustic descriptors from bag-of-vectors representation with transformers}, volume = {145}, year = {2025}, }
• (2025). Transfer learning enhanced deep neural network application in gauss quadrature for isogeometric analysis. EAAI, 145, 110182. (www)

  @article{ , abstract = {Artificial neural networks (ANNs) have gained significant popularity in computational mechanics due to their capacity to learn intricate correlations between inputs and outputs, making them suitable for a wide range of applications in the field. The number of Gauss quadrature points (nGQPs) required for the numerical integration of elemental quantities influences the accuracy and computational cost of non-uniform rational B-spline (NURBS) based isogeometric analysis (IGA). In this work, we propose four deep neural network (DNN) models to determine the optimum nGQPs for evaluating the element stiffness matrix in IGA. Out of the four DNN models, the hybrid DNN model, which combines data-balancing through the synthetic minority oversampling technique (SMOTE) and cost-sensitive learning, performs best in predicting nGQPs from an imbalanced data set. The robustness of this approach is further increased by the use of transfer learning to predict nGQPs for elements with different order NURBS. Moreover, the use of transfer learning bypasses the exhaustive search for the optimum values of hyperparameters of a DNN model. We solve three linear elastic problems where the stiffness matrices of the elements are computed with nGQPs predicted by the hybrid DNN model. Instead of assigning fixed nGQPs for the complete analysis, it is shown that the proposed approach significantly reduces the computational time while achieving the desired accuracy.}, archive = {J_EAAI}, author = {Dipjyoti Nath and Debanga Raj Neog and Sachin Singh Gautam}, doi = {10.1016/j.engappai.2025.110182}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110182}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transfer learning enhanced deep neural network application in gauss quadrature for isogeometric analysis}, volume = {145}, year = {2025}, }
• (2025). Multi-step look ahead deep reinforcement learning approach for automatic train regulation of urban rail transit lines with energy-saving. EAAI, 145, 110181. (www)

  @article{ , abstract = {Urban rail transit train operations are frequently disturbed by external factors, making schedule adherence challenging. These disturbances can cause cascading train delays, leaving many passengers stranded and compromising system safety, stability, and efficiency. Furthermore, increasing energy consumption concerns in train operations necessitates the incorporation of energy-saving considerations in formulating train regulation strategies. To address the practical concern, this paper introduces an innovative automatic train regulation learning environment, which models the train traffic dynamic, passenger flow, and energy consumption within a Markov decision process. In automatic train regulation learning environment, the automatic train regulation system functions as the learning agent interacting with the environment. A novel multi-step look ahead deep deterministic policy gradient algorithm is designed for the agent to decide real-time regulation strategies. This algorithm, enabling the anticipation of future states, ensures robust and reliable decision-making for train operations under disturbed conditions. Real-world experiments including comparisons with other commonly-used automatic train regulation strategies demonstrates the effectiveness of the proposed method. The generated regulation strategy shows significant improvements in delay reduction, passenger service quality, and energy savings. Actually, the proposed methodology not only enhances urban rail transit system performance but also extends deep reinforcement learning applications in intelligent transportation, offering an innovative and promising solution to transportation operational challenges.}, archive = {J_EAAI}, author = {Yunfeng Zhang and Shukai Li and Yin Yuan and Lixing Yang}, doi = {10.1016/j.engappai.2025.110181}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110181}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-step look ahead deep reinforcement learning approach for automatic train regulation of urban rail transit lines with energy-saving}, volume = {145}, year = {2025}, }
• (2025). Process planning of parameter intelligent adjustment for batch machining based on historical data segmented modeling. EAAI, 145, 110180. (www)

  @article{ , abstract = {Improving machining benefit while maintaining stable quality and performance is a prerequisite for improving intelligent manufacturing and enterprise competitiveness. This paper attempts to explore the method of segmented optimization of process parameters to improve benefit (such as machining time and tool life) while maintaining quality and performance within a given range during the tool life cycle in batch machining. An intelligent parameter adjustment framework is proposed, which utilizes the historical machining data to construct the process parameter adjustment time series (i.e., process planning) in the tool life cycle, and by online adding data from new tool machining processes, the time series can be modified. The key technology of the framework is to determine the adjustment point and the corresponding process parameters. A segmented prediction model based on historical data is presented to predict the adjustment point during the new tool machining process of, and a double depth Q-network (DDQN) is utilized to obtain the optimized machining parameters at the adjustment point. Based on a milling experiment of TC18, the effectiveness of the proposed method is validated by the prediction and optimization of the first adjustment point. The adjustment point is predicted to be the 58th machining of the new tool, after adjusting the process parameters obtained through DDQN at the adjustment point, the quality and cutting force remain within the set threshold, cutting time decreases by 0.2% and tool life increases by 14.42%, which shows that the proposed methods achieve quality assurance and benefit enhancement during the machining process.}, archive = {J_EAAI}, author = {Juan Lu and Shiying Tu and Ying Li and Liang Zhang and Xiaoping Liao}, doi = {10.1016/j.engappai.2025.110180}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110180}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Process planning of parameter intelligent adjustment for batch machining based on historical data segmented modeling}, volume = {145}, year = {2025}, }
• (2025). Lightweight defect detection network based on steel strip raw images. EAAI, 145, 110179. (www)

  @article{ , abstract = {Efficient defect detection on hot rolled steel strips is important for industrial production. However, existing defect detection methods are not lightweight enough and lack detection ability for actual steel production environments. A lightweight detection framework based on raw steel strip defect images is proposed to address this gap. A new strip surface defect detection dataset comprising 2650 raw images with five defect types is established. Considering the unique features of strip images, a new strip image augmentation strategy is employed to enhance training sample diversity. Then, a novel lightweight model is introduced. The model consists of the Location Enhanced Ghost Network (LEG-Net) and the Refine Grouped Spatial Network (RGS-Net). The LEG-Net incorporates Ghost modules and a new Location Enhanced Attention. The lightweight backbone effectively reduces the number of parameters. The RGS-Net neck part consists of a slim neck and Efficient Channel Attention. The RGS-Net increases the extraction of channel information and then realizes the defect recognition between different defect types and backgrounds by adjusting the spatial receptive field mechanism. The mean Average Precision (mAP) accuracy of the proposed lightweight model is 76.9% and the speed is 33 Frames Per Second (FPS). Compared to existing models on three datasets, the proposed network delivers superior detection accuracy while incurring lower computational costs. It effectively identifies challenging defects, such as small targets and large fuzzy samples. Furthermore, the proposed framework closely resembles the actual steel production environment, thereby advancing the industrial application of intelligent defect detection.}, archive = {J_EAAI}, author = {Yue Huang and Zhen Chen and Zhaoxiang Chen and Di Zhou and Ershun Pan}, doi = {10.1016/j.engappai.2025.110179}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110179}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight defect detection network based on steel strip raw images}, volume = {145}, year = {2025}, }
• (2025). A novel semi-local centrality to identify influential nodes in complex networks by integrating multidimensional factors. EAAI, 145, 110177. (www)

  @article{ , abstract = {This study addresses the critical problem of identifying influential nodes in complex networks, a task that plays a pivotal role in understanding network dynamics, optimizing information spread, and controlling epidemic outbreaks. Although semi-local centrality metrics are a valid approach for identifying influential nodes, they face challenges such as inefficiency when dealing with large-scale networks and neglecting semantic relationships, often relying on unidimensional criteria that limit their effectiveness. To tackle this challenge, this study presents a novel Semi-Local Centrality metric designed to identify influential nodes in complex networks by incorporating Multidimensional Factors (SLCMF). SLCMF combines structural, social, and semantic factors to find seed nodes in complex networks. To improve scalability, SLCMF utilizes distributed local subgraphs and redefines semi-local centrality by employing the average shortest path theory. Additionally, SLCMF incorporates a semantic graph embedding model by an augmented graph to capture distant and latent relationships among nodes. Extensive experiments on real-world networks demonstrate the effectiveness and efficiency of the proposed centrality metric, showcasing its superior performance in ranking influential nodes. Specifically, SLCMF outperforms the best traditional and advanced centrality metrics, improving Kendall's correlation coefficient by 8.94% and 1.61%, respectively. Additionally, the proposed metric demonstrates enhanced efficiency, reducing runtime by 4.7% and 0.21% compared to the top-performing traditional and advanced metrics, respectively.}, archive = {J_EAAI}, author = {Kun Zhang and Zaiyi Pu and Chuan Jin and Yu Zhou and Zhenyu Wang}, doi = {10.1016/j.engappai.2025.110177}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110177}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel semi-local centrality to identify influential nodes in complex networks by integrating multidimensional factors}, volume = {145}, year = {2025}, }
• (2025). Enhanced feedback analysis of vertical load reliability parameters for airplane landing gear using an improved generative adversarial network and explainable artificial intelligence techniques. EAAI, 145, 110175. (www)

  @article{ , abstract = {Effective feedback analysis of critical equipment data is essential for improving performance and optimizing design parameters in aviation systems. This study presents a novel framework that integrates an improved generative adversarial network (GAN) with explainable artificial intelligence techniques (XAI) to evaluate the reliability of the vertical load for airplane landing gear. By utilizing limited data from the Quick Access Recorder (QAR), the improved GAN generates extensive synthetic data to expand the dataset and strengthen the analysis. Each parameter's importance and influence on vertical load reliability are then evaluated through the Shapley Additive Explanations (SHAP) method, a key approach in XAI. Validation using landing gear data from a typical civil airplane demonstrates the effectiveness of this method and confirms the viability of explainable artificial intelligence for parametric feedback analysis. The results highlight the impact of each parameter on vertical load reliability, providing valuable insights to support enhanced design and operational efficiency of landing gear.}, archive = {J_EAAI}, author = {Weihuang Pan and Yunwen Feng and Cheng Lu and Jiaqi Liu and Jingcui Liang}, doi = {10.1016/j.engappai.2025.110175}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110175}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced feedback analysis of vertical load reliability parameters for airplane landing gear using an improved generative adversarial network and explainable artificial intelligence techniques}, volume = {145}, year = {2025}, }
• (2025). Ensemble modeling for predicting head vibration based on driving seating conditions: Towards adaptive seating systems. EAAI, 145, 110174. (www)

  @article{ , abstract = {Personalized adaptive seating systems represent a frontier in automotive engineering, promising to enhance driver comfort and safety by adapting to individual characteristics and driving seating conditions. This study investigated the application of ensemble modeling techniques to predict tri-axial head vibration in drivers along the fore-and-aft (X), lateral (Y), and vertical (Z) axes. We addressed the complex interactions between individual characteristics, seating condition, and whole-body vibration exposure. We analyzed data from 15 male subjects (mean age: 21.93 ± 1.03 years; body mass index: 24.13 ± 2.62 kg/m 2 ) across various seating conditions. Statistical analysis showed significant effects of seating conditions and participant characteristics on acceleration responses across the X, Y, and Z axes (p < 0.05). We employed Random Forest, Bagging, Gradient Boosting, and Adaptive Boosting algorithms to predict tri-axial acceleration responses. The Random Forest model outperformed others, achieving R-Squared values of 0.879, 0.922, and 0.871 for X, Y, and Z axes, respectively. Feature importance analysis demonstrated that combinations of anthropometric and seating variables were most influential, with importance scores up to 0.99 for key feature combinations. The analysis emphasized varying impacts of features across axes, with body mass index and backrest angle showing consistently strong influences. The findings demonstrate the superiority of ensemble methods over traditional approaches in capturing nonlinear relationships affecting driver comfort and safety, offering potential for real-time monitoring and personalized ergonomic interventions in automotive engineering.}, archive = {J_EAAI}, author = {Amandeep Singh and Naser Nawayseh and Subhash Rakheja}, doi = {10.1016/j.engappai.2025.110174}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110174}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ensemble modeling for predicting head vibration based on driving seating conditions: Towards adaptive seating systems}, volume = {145}, year = {2025}, }
• (2025). An efficient m-step lookahead rollout algorithm for profit-oriented selective disassembly sequence planning with operation stochastic failure. EAAI, 145, 110173. (www)

  @article{ , abstract = {To fully reclaim large amounts of used electro-mechanic products, an effective and high-quality disassembly sequence is crucial that separates a product into parts/components one-by-one for recovery. However, it becomes rather challenging to efficiently determine the optimal/near-optimal disassembly sequence for used products since the quality conditions of used products are uncertain and disassembly operations that constitute the disassembly sequence might fail. In this paper, the stochastic failure characteristics of disassembly operations is taken into account and a selective disassembly sequence planning is studied for maximizing the recovering profit of used products. First, we formally model a profit-oriented selective disassembly sequence planning with operation stochastic failure (PSDSP-OSF), in which the expected recovering profit of a disassembly sequence is formulated based on the stochastic failure characteristics of disassembly operations. Then, we simplify and reformulate PSDSP-OSF according to the disassembly rule. A m -step lookahead rollout algorithm is proposed to efficiently solve PSDSP-OSF, in which every disassembly operation is selected by a lookahead decision rule that looking ahead from the current decision stage to future m decision stages. Finally, three different scales of products are selected as case studies and two well-known existing methods are employed to compare with the proposed approach. The computational results demonstrate that our proposed approach is significantly superior to the existing methods in solution quality, computational efficiency, and algorithm robustness.}, archive = {J_EAAI}, author = {Yaping Ren and Leilei Meng and Guangdong Tian and Zhiwu Li and Yun Li}, doi = {10.1016/j.engappai.2025.110173}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110173}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient m-step lookahead rollout algorithm for profit-oriented selective disassembly sequence planning with operation stochastic failure}, volume = {145}, year = {2025}, }
• (2025). Domain-adaptive person re-identification without cross-camera paired samples. EAAI, 145, 110171. (www)

  @article{ , abstract = {Existing person re-identification (re-ID) mainly focuses on pedestrian identity matching across cameras in adjacent areas. However, in reality, it is inevitable to face the problem of pedestrian identity matching across long-distance scene. The cross-camera pedestrian samples collected from this scene often have no positive samples. It is extremely challenging to use cross-camera negative samples to train a re-ID model to achieve cross-region pedestrian identity matching. To solve this problem, a novel domain-adaptive person re-ID method that focuses on cross-camera consistent discriminative feature learning under the supervision of unpaired samples is proposed. This method mainly includes category synergy co-promotion module (CSCM) and cross-camera consistent feature learning module (CCFLM). In CSCM, a task-specific feature recombination mechanism is proposed. This mechanism first groups features according to their contributions to specific classification tasks. Then an interactive promotion learning scheme between feature groups is developed and embedded to enhance feature discriminability. Since the control parameters of the specific task model are reduced after division by task, the generalization ability of the model is improved. In CCFLM, instance-level feature distribution alignment and cross-camera identity consistent learning are constructed. Therefore, the supervised model training is achieved under the style supervision of the target domain by exchanging styles between source-domain samples and target-domain samples, and the challenges caused by the lack of cross-camera paired samples are solved by utilizing cross-camera similar samples. In experiments, the effectiveness of the proposed method is demonstrated through four experimental settings. The source code of the proposed method can be available at https://github.com/lhf12278/CSC-FLM .}, archive = {J_EAAI}, author = {Huafeng Li and Yanmei Mao and Yafei Zhang and Guanqiu Qi and Zhengtao Yu}, doi = {10.1016/j.engappai.2025.110171}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110171}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Domain-adaptive person re-identification without cross-camera paired samples}, volume = {145}, year = {2025}, }
• (2025). Corrigendum to “Online model-based anomaly detection in multivariate time series: Taxonomy, survey, research challenges and future directions” [Eng. appl. artif. intell. 138 (2024) 109323]. EAAI, 145, 110170. (www)

  @article{ , archive = {J_EAAI}, author = {Lucas Correia and Jan-Christoph Goos and Philipp Klein and Thomas Bäck and Anna V. Kononova}, doi = {10.1016/j.engappai.2025.110170}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110170}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Corrigendum to “Online model-based anomaly detection in multivariate time series: Taxonomy, survey, research challenges and future directions” [Eng. appl. artif. intell. 138 (2024) 109323]}, volume = {145}, year = {2025}, }
• (2025). A deep dive into automated sexism detection using fine-tuned deep learning and large language models. EAAI, 145, 110167. (www)

  @article{ , abstract = {The issue of sexism in online content has recently been a significant concern. With the increasing number of online interactions and the rise of social media platforms, the need for automated techniques to identify and classify sexism has become more critical than ever. This paper addresses this problem by fine-tuning deep-learning models for sexism classification using “ MultiHate ”. It is a comprehensive dataset created by curating ten different datasets on sexism. The dataset consists of 1.76 M English texts labelled as sexist and not sexist, then fine-tuned two deep learning models, Convolutional Neural Networks-Bidirectional Long Short-Term Memory and Generative Pre-trained Transformer 2, which accurately detect and classify sexism. A comparative analysis has been conducted on several machine learning and deep learning models using the MultiHate dataset. Investigation reveals that the Generative Pre-trained Transformer 2 model outperforms other models with an accuracy of 92%, while the Convolutional Neural Networks-Bidirectional Long Short-Term Memory model achieved an accuracy of 90% using precision, recall, and F1 scores as performance metrics. The models’ performances are promising, indicating that automated techniques can be employed to classify sexist content effectively. A comprehensive error analysis of the models’ performance has been presented, highlighting their limitations and challenges. The computational time required for training and testing the models is a significant challenge, especially for larger datasets.}, archive = {J_EAAI}, author = {Advaitha Vetagiri and Partha Pakray and Amitava Das}, doi = {10.1016/j.engappai.2025.110167}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110167}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep dive into automated sexism detection using fine-tuned deep learning and large language models}, volume = {145}, year = {2025}, }
• (2025). Integrated energy management for hybrid electric vehicles: A bellman neural network approach. EAAI, 145, 110166. (www)

  @article{ , abstract = {In light of growing concerns surrounding environmental pollution, resource depletion, and global warming, hybrid electric vehicles (HEVs) have gained significant popularity. An effective energy management strategy (EMS) is pivotal in enhancing fuel efficiency and overall performance in HEVs. While dynamic programming (DP) and Pontryagin’s minimum principle (PMP) are classical methods for achieving optimal solutions, drawbacks such as ‘curse of dimensionality’ and initial state guess usually make their applications challenging. On the other hand, the rapid development of neural network based learning techniques brings new aspects. Physics-informed neural networks (PINNs) have emerged as a promising technology. Considering the key role of Bellman optimality in describing the optimal EMS, this paper proposes the Bellman optimal energy management (BOEM) algorithm by leveraging the power of PINNs to approximate the solution of the underlying Hamilton–Jacobi–Bellman (HJB) partial differential equation (PDE) of the EMS. BOEM not only yields highly accurate optimal solutions but also avoids the challenges associated with co-state initial guess dependency. Both theoretical analysis and comprehensive simulation results are presented to show the effectiveness and distinctive qualities of the proposed BOEM method. Additionally, the simulations reveal the distinctive characteristics including accuracy, data efficiency, and extra parameter for adapting different driving preferences, underlining its potential to revolutionize energy management in HEVs.}, archive = {J_EAAI}, author = {Fangyuan Li and Lefei Gao and Yubo Zhang and Yanhong Liu}, doi = {10.1016/j.engappai.2025.110166}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110166}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrated energy management for hybrid electric vehicles: A bellman neural network approach}, volume = {145}, year = {2025}, }
• (2025). A two-stage classification-regression method for prediction of flexural strength of fiber reinforced polymer strengthened reinforced concrete beams. EAAI, 145, 110164. (www)

  @article{ , abstract = {Accurate prediction of the flexural strength of externally bonding fiber reinforced polymer strengthened reinforced concrete beams (FSRCBs) is challenging for both traditional semi-empirical equations and machine learning models because it involves classification of failure modes in advance. Machine learning models that can predict the flexural strength without knowing the failure modes are scare to date. This study proposed a novel two-stage machine learning, classification-regression method, to first predict the failure mode of FSRCBs and then use the predicted failure mode as an additional input to predict the flexural strength. Using a database of 702 FSRCBs collected from the literature, the accuracy of the failure mode predicted by the proposed model reached 96.5%. With the predicted failure mode, the XGBoost (extreme gradient boosting) and AdaBoost (adaptive boosting) regression model were used to predict the flexural strength with an R 2 of 0.966 and 0.973, respectively. Shapley additive explanation (SHAP) analysis was employed to ensure interpretability of the predictions generated by the proposed method. An explicit and practical equation was derived to predict the flexural strength based on results of machine learning, which showed higher prediction accuracy compared with traditional equations. A graphic users' interface with the well-trained model is also provided.}, archive = {J_EAAI}, author = {Hao Xu and Xingxing Zou and Lesley H. Sneed}, doi = {10.1016/j.engappai.2025.110164}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110164}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A two-stage classification-regression method for prediction of flexural strength of fiber reinforced polymer strengthened reinforced concrete beams}, volume = {145}, year = {2025}, }
• (2025). Comparing passive and active fault-tolerant control of swashplate actuator faults in a medium-scale rotorcraft unmanned aerial vehicle with experimental validation. EAAI, 145, 110163. (www)

  @article{ , abstract = {This paper compares two fault-tolerant control strategies for swashplate actuator faults in a medium-scale rotorcraft: super-twisting sliding mode control (STSMC) and dynamic neural network (DNN)-based feedback linearisation control combined with fault detection and isolation module. Rotor speed is used as a redundant control option. Computational intelligence techniques are used to optimise the controllers parameters, with STSMC tuned using single-objective optimisation (SOO) and the DNN using multi-objective optimisation (MOO). Validation is done using a hardware-in-the-loop experimental rig. Results show acceptable fault recovery up to 40% loss-of-effectiveness (LOE) for the STSMC and beyond 50% LOE for the dynamic neural network-based feedback linearisation (DNNFBL) controller.}, archive = {J_EAAI}, author = {Lindokuhle J. Mpanza and Jimoh O. Pedro}, doi = {10.1016/j.engappai.2025.110163}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110163}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comparing passive and active fault-tolerant control of swashplate actuator faults in a medium-scale rotorcraft unmanned aerial vehicle with experimental validation}, volume = {145}, year = {2025}, }
• (2025). CarNet: A generative convolutional neural network-based line-of-sight/non-line-of-sight classifier for global navigation satellite systems by transforming multivariate time-series data into images. EAAI, 145, 110160. (www)

  @article{ , abstract = {Urban environments present significant challenges to commercial Global Navigation Satellite Systems (GNSS) receivers due to degraded satellite visibility and Non-line-of-sight (NLOS) receptions. Mitigating NLOS receptions for GNSS is essential, especially for safety-critical and reliability-critical location-based applications. Traditional physical error channel propagation modeling encountered bottlenecks since the NLOS and multipath errors cannot be modeled accurately in complex urban environments. Data-driven methods show significant potential for effectively classifying GNSS Line-of-sight (LOS) and NLOS. This paper proposes the CarNet - a generative Convolutional Neural Network (CNN)-based GNSS LOS/NLOS classifier by transforming multivariate time-series data into images. CarNet comprises two modules: an image generator and an image classifier. The image generator enriches and augments the original 1-dimension feature vector into 2-dimension feature maps and the image classifier uses an inception-based CNN to realize multi-scale feature extraction and classification. The proposed architecture is trained and tested on more than 6 h of real vehicle data collected in different challenging environments (about 1.6 million samples). A thorough benchmark is conducted, comparing CarNet against the existing mainstream Artificial Intelligence (AI) methods. The results with cross-validation on unseen data indicate that CarNet achieves the highest accuracy, i.e., 81.47% while maintaining the optimal balance between precision for both classes: 83.3% for LOS and 70.99% for NLOS. Finally, positioning accuracy is assessed using a reweighting strategy based on the LOS/NLOS information predicted by CarNet. The assessment of total datasets shows that CarNet weighting can achieve the best accuracy compared to the traditional weighting schemes based on signal-to-noise ratio or satellite elevation. CarNet shows strong potential for embedding into GNSS receivers to enhance positioning accuracy in complex urban environments, benefiting a wide range of location-based applications such as autonomous driving, emergency response, and urban logistics.}, archive = {J_EAAI}, author = {Ni Zhu and Ruiwen He and Zhiqiang Wang}, doi = {10.1016/j.engappai.2025.110160}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110160}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {CarNet: A generative convolutional neural network-based line-of-sight/non-line-of-sight classifier for global navigation satellite systems by transforming multivariate time-series data into images}, volume = {145}, year = {2025}, }
• (2025). A hardware accelerator to support deep learning processor units in real-time image processing. EAAI, 145, 110159. (www)

  @article{ , abstract = {Deep neural networks are becoming crucial in many cyber–physical systems involving complex perceptual tasks. For those embedded systems requiring real-time interactions with dynamic environments, as autonomous robots and drones, it is of paramount importance that such algorithms are efficiently executed onboard on properly designed hardware accelerators to meet the required performance specifications. In particular, some neural network architectures for object detection and tracking, as You Only Look Once (YOLO), include heavy computational stages that need to be executed before and after the model inference. Such stages are typically not incorporated in traditional accelerators and are executed on general-purpose processors, thus introducing a bottleneck in the overall processing pipeline. To overcome such a problem, this paper presents a general-purpose accelerator on a field-programmable gate array (FPGA) able to run pre-processing and post-processing operations typically required by vision tasks. The proposed solution has been tested in combination with a YOLO object detector accelerated on an Advanced Micro Devices (AMD) Xilinx Kria KR260 board mounting an UltraScale+ multiprocessor system-on-chip, achieving a significant improvement in terms of both timing performance and power consumption, and enabling onboard visual processing into drones. The proposed solution is able to boost the traditional object detection process by a factor of 4.4, allowing the execution of the full processing pipeline at 60 frames per second (fps), versus 13.6 fps reachable without the proposed accelerator. As a result, this work enables the use of high-speed cameras for developing more reactive systems that can respond to incoming events with lower latency.}, archive = {J_EAAI}, author = {Edoardo Cittadini and Mauro Marinoni and Giorgio Buttazzo}, doi = {10.1016/j.engappai.2025.110159}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110159}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hardware accelerator to support deep learning processor units in real-time image processing}, volume = {145}, year = {2025}, }
• (2025). A hybrid deep learning method for the real-time prediction of collision damage consequences in operational conditions. EAAI, 145, 110158. (www)

  @article{ , abstract = {Ship collisions can result in catastrophic outcomes, necessitating effective real-time collision risk assessment methods for proactive risk management. These methods need to rapidly evaluate both the probability of collision and the potential damage dimensions (length, height, and penetration) in real conditions. Existing frameworks often underestimate collision damage consequences during operational risk assessments. This paper presents a hybrid deep learning approach for the real-time prediction of collision damage dimensions under real ship operation conditions. Collision scenarios are identified using Automatic Identification System (AIS) data, with damage extents simulated through the Super Element (SE) method. A comprehensive database of collision scenarios and corresponding damage assessments is developed, sourced from realistic operational data of Ro-Pax ship in the Gulf of Finland. The deep learning model is trained and validated using this dataset, ensuring the model's relevance and practical applicability. Extensive comparative analyses and generalization tests demonstrate the high accuracy of the model in predicting ship collision damages in diverse ship operational conditions. In addition, traditional simulation methods for evaluating damage dimensions require approximately 10 min, whereas the trained deep learning model reduces the time to less than 0.1 s, enabling real-time potential collision consequence assessment in real operational conditions. The proposed model may provide significant insights for ship operators, enhancing ship safety and supporting intelligent decision-making in ship operations.}, archive = {J_EAAI}, author = {Mingyang Zhang and Hongdong Wang and Fabien Conti and Teemu Manderbacka and Heikki Remes and Spyros Hirdaris}, doi = {10.1016/j.engappai.2025.110158}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110158}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid deep learning method for the real-time prediction of collision damage consequences in operational conditions}, volume = {145}, year = {2025}, }
• (2025). Adversarial and self-adaptive domain decomposition physics-informed neural networks for high-order differential equations with discontinuities. EAAI, 145, 110156. (www)

  @article{ , abstract = {The technique of solving differential equations using physics-informed neural networks (PINNs) has received extensive attention and application. Analogous to the concept of adaptive mesh refinement in finite element methods, the PINNs framework should be tailored to the specific characteristics of the problem to further improve performance. However, the adaptive techniques in PINNs focus primarily on sampling and weighting, lacking the capability for adaptive domain decomposition. To address the limitations of PINNs in solving high-order problems with discontinuities, this paper proposes a novel type of adversarial and self-adaptive domain decomposition physics-informed neural networks (AS-PINNs). AS-PINNs leverage residual differences between subdomains to implement adversarial training among subnetworks, enabling the automatic and universal adjustment of subdomain interface positions and the capture of discontinuities with varying characteristics, representing a novel neural network approach. Compared to traditional domain decomposition methods, AS-PINNs eliminate the need for imposing tedious boundary and interface conditions through the loss function, particularly in high-order differential equations, thereby significantly reducing the complexity of the loss function and intrinsically improving accuracy. The results show that the self-adaptive adjustment of subdomain and the network structure makes AS-PINNs to be tailored to specific engineering problems, such as multi-material issues, high-order problems. As the order of the differential equations increases, the accuracy and speed advantages of AS-PINNs become more pronounced. For sixth-order differential equations, the solution speed of AS-PINNs is nine times faster compared to traditional domain decomposition PINNs. Code available at: https://github.com/Ning343/AS-PINNs.git .}, archive = {J_EAAI}, author = {Mingsheng Peng and Hesheng Tang and Yingwei Kou}, doi = {10.1016/j.engappai.2025.110156}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110156}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adversarial and self-adaptive domain decomposition physics-informed neural networks for high-order differential equations with discontinuities}, volume = {145}, year = {2025}, }
• (2025). A dual-discriminator network based on sobel gradient operator for digital twin-assisted fault diagnosis. EAAI, 145, 110155. (www)

  @article{ , abstract = {Intelligent fault diagnosis technology holds great potential for industrial applications, but it heavily depends on high-quality datasets to improve model robustness and diagnostic performance. The scarcity of adequate fault data in industrial settings severely limits its adoption. As an emerging technology, Digital Twin (DT) provides a novel solution by enhancing model performance by integrating virtual and physical data. Despite its significant potential, the notable distributional differences between virtual data and physical data can lead to severe negative knowledge transfer during the data fusion process, ultimately compromising diagnostic accuracy. Hence, a dual-discriminator generative adversarial network based on the Sobel gradient operator (DDGAN-Sobel) is proposed to tackle this problem. Firstly, a DT model of the shaft system is created to generate virtual operation data. Next, an innovative dual discriminator network structure is designed, with two discriminators respectively focusing on the global and local features of the synthetic data. Meanwhile, the Sobel gradient operator is introduced to capture mutation frequency features, achieving the directional transmission of specified fault information. Finally, a transfer learning framework for domain-adaptive fusion is proposed, which enhances data diversity by utilizing synthetic data to address the overfitting problem. The validation test is conducted on a custom shaft system testing platform. The results reveal that DDGAN-Sobel attains average diagnostic accuracies of 99.46%, 100%, and 98.47% under three different conditions, which is superior to the current leading algorithms. These satisfactory outcomes underscore the potential of the proposed method in industrial applications.}, archive = {J_EAAI}, author = {Shuyang Luo and Jiachang Qian and Xufeng Huang and Qi Zhou and Jiexiang Hu}, doi = {10.1016/j.engappai.2025.110155}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110155}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual-discriminator network based on sobel gradient operator for digital twin-assisted fault diagnosis}, volume = {145}, year = {2025}, }
• (2025). Combining informed data-driven anomaly detection with knowledge graphs for root cause analysis in predictive maintenance. EAAI, 145, 110152. (www)

  @article{ , abstract = {Industry 4.0 has facilitated the access to sensor and actuator data from manufacturing systems, leading to studies on data-driven anomaly detection, but limited attention has been paid to finding root causes and automating this process using formalized expert knowledge. This is crucial due to the scarcity of qualified engineers and the time-consuming nature of diagnosing issues in large production systems. To address this gap, we present a framework that combines data-driven anomaly detection with a knowledge graph that provides domain knowledge by leveraging typical explanations of such models (i.e.,data streams potentially caused the detection) for further diagnosis. The framework’s usefulness to infer affected components or data set labels has been evaluated using two deep anomaly detection approaches. For knowledge-based diagnosis, three query strategies that utilize various knowledge graph relationships are implemented through three Artificial Intelligence (AI) techniques. The proposed anomaly detection approach, informed by integrating expert knowledge via the graph structure of the knowledge graph and node embeddings for encoding time series, outperforms baselines and a deep autoencoder in detecting anomalies and in identifying anomalous data streams. In subsequent diagnosis, it achieves the best performance on a complete knowledge graph in combination with a graph pattern matching query by identifying the label or affected component in 60% of detected anomalies by providing 4.1 labels or 2.3 components until the correct one is identified. In case of a corrupted one, Symbolic-Driven Neural Reasoning (SDNR) and Case-Based Reasoning (CBR) with knowledge graph embeddings demonstrate advantages by halving the number of incorrect labels and unaffected components.}, archive = {J_EAAI}, author = {Patrick Klein and Lukas Malburg and Ralph Bergmann}, doi = {10.1016/j.engappai.2025.110152}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110152}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Combining informed data-driven anomaly detection with knowledge graphs for root cause analysis in predictive maintenance}, volume = {145}, year = {2025}, }
• (2025). Predictor-based state-constrained bipartite formation control for nonlinear multi-agent systems with disturbances. EAAI, 145, 110150. (www)

  @article{ , abstract = {Constrained states and transient performance are critical in bipartite formation control of multi-agent systems. In this paper, a predictor-based state-constrained bipartite formation control strategy is proposed for a class of nonlinear multi-agent systems with unknown external disturbances. A dual barrier Lyapunov function is proposed to constrain both prediction errors and predictor-based surface errors. With the help of the dual barrier Lyapunov function, a barrier Lyapunov function-based predictor is then constructed. It alleviates oscillations generated by overlarge adaptive gains in identification of state-constrained followers. With a barrier Lyapunov function-based predictor, barrier Lyapunov function neural networks are developed to approximate unknown dynamics of a multi-agent system with state constraints. A barrier Lyapunov function neural network nonlinear disturbance observer is designed for compensating for generalized disturbances including disturbances as well as barrier Lyapunov function neural networks’ identification errors. From analysis, it is proven that the multi-agent system achieves bipartite formation and states of followers satisfy the constrained condition. The effectiveness of the strategy with state constraints is verified via two simulation examples. Compared with the traditional approaches, the proposed strategy reduces the integrated absolute formation error by 31% and the integrated absolute error of identification by 63% during the simulation process.}, archive = {J_EAAI}, author = {Yang Yang and Xiao Wu and Hongyan Yu and Chen Wang}, doi = {10.1016/j.engappai.2025.110150}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110150}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predictor-based state-constrained bipartite formation control for nonlinear multi-agent systems with disturbances}, volume = {145}, year = {2025}, }
• (2025). Explainable optimal random forest model with conversational interface. EAAI, 145, 110134. (www)

  @article{ , abstract = {Machine Learning (ML) models are now a part of our everyday lives and significantly impact each individual. When users are given a clear and relevant explanation for a decision made by the model, they can understand the reasons and logic behind the model’s behavior, which helps to build trust in the model. This article proposes a zeroth level of explainability for an Optimal Random Forest model, which incorporates an Optimal Sparse Decision tree that is self-explainable. The methodology also involves vectorizing subtrees relevant to an instance, followed by dimensionality reduction and clustering. This method enhances explainability, ensuring high fidelity in the explanations provided. It also proposes a conversational interface for explaining the model’s decisions. Experimental analysis was done on four benchmark datasets, namely SMS Spam collection dataset (classification problem of ham or spam messages), Iris dataset (classification problem of Setosa or Non-Setosa based on length and width information of both sepal and petal data), Tic-tac-toe endgame dataset (based on board position content and their corresponding class labels), Heart Disease dataset (classification problem of prone to heart disease or not). Experimental analysis has demonstrated the explainability of the Optimal Random Forest model without any compromise on generalization and performance. Comparisons were also made with Shapley Additive Explanation (SHAP) derived from both a generalized random forest model and a Support Vector Machine model, and also on Rule learning + Voting model. The results have demonstrated the merits of the algorithm.}, archive = {J_EAAI}, author = {Caroline Mary M. and Jennath H.S. and Asharaf S.}, doi = {10.1016/j.engappai.2025.110134}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110134}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable optimal random forest model with conversational interface}, volume = {145}, year = {2025}, }
• (2025). Explainable evaluation of generative adversarial networks for wearables data augmentation. EAAI, 145, 110133. (www)

  @article{ , abstract = {Data augmentation represents an opportunity for Artificial Intelligence (AI) applications, as it aims at creating new synthetic data based on an existing baseline. In this paper, we present a new evaluation framework for Generative Adversarial Networks (GANs), a data augmentation technique, in multivariate data classification contexts. The goal is not limited to assessing the performance variations obtained through GANs, but also to inspect results with explainable AI (XAI) tools, understanding how GANs work and, finally, exploiting them to discover new knowledge. To this aim, we adopt the Logic Learning Machine (LLM) for performance assessment and rule extraction, and introduce a new measure of rule similarity to compare different artificial datasets. We apply the methodology on two case studies, activity recognition and physical fatigue detection, confirming that GANs can help in overcoming limitations of original datasets and lead to new discoveries.}, archive = {J_EAAI}, author = {Sara Narteni and Vanessa Orani and Enrico Ferrari and Damiano Verda and Enrico Cambiaso and Maurizio Mongelli}, doi = {10.1016/j.engappai.2025.110133}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110133}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable evaluation of generative adversarial networks for wearables data augmentation}, volume = {145}, year = {2025}, }
• (2025). Spatial–temporal intention representation with multi-agent reinforcement learning for unmanned surface vehicles strategies learning in asset guarding task. EAAI, 145, 110120. (www)

  @article{ , abstract = {As a typical application of artificial intelligence, autonomous and intelligent Unmanned Surface Vehicles (USVs) hold significant practical value in asset guarding tasks, as they can ensure target security while significantly reducing costs. Each USV is tasked with inferring adversary intentions based on locally observable information and proactively intercepting intruding boats to maximize asset survival time. Most current methods primarily identify and represent intentions through either prior rule matching or low-dimensional behavioral features. However, rule-based methods struggle to handle dynamically changing intents, and behavioral features can easily introduce uncertainty in intention recognition. In this paper, we propose a Spatial–Temporal Intention Representation (STIR) model that effectively conveys the dynamic intentions of various boats in asset guarding tasks, enhancing the learning efficiency of USVs’ strategies. First, utilizing local observation information from USVs, we construct an intention recognition tree that correlates intentions with a priori task background knowledge, thereby reducing uncertainty in intention recognition. Second, we develop a spatial–temporal attention network to dynamically represent intentions in both spatial and temporal dimensions, improving the USVs’ understanding of local scene dynamics. Third, we combine STIR with multi-agent reinforcement learning to train the interception strategies of USVs. In the experiments, we discuss the positive impact of STIR on adversarial strategy learning in the asset guarding task. Simulation results illustrate the advantages of our approach in terms of learning speed and strategy effectiveness.}, archive = {J_EAAI}, author = {Yang Li and Shaorong Xie and Hang Yu and Han Zhang and Zhenyu Zhang and Xiangfeng Luo}, doi = {10.1016/j.engappai.2025.110120}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110120}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatial–temporal intention representation with multi-agent reinforcement learning for unmanned surface vehicles strategies learning in asset guarding task}, volume = {145}, year = {2025}, }
• (2025). Hypergraph temporal multi-behavior recommendation. EAAI, 145, 110112. (www)

  @article{ , abstract = {As the scale of e-commerce and the number of item categories increase, user behaviors become increasingly diverse, and the real relationships between users and items in recommendation systems become considerably more complex. One of the emerging areas of research in this context is a multi-behavior recommendation, which aims to consider various types of user behavior to better predict user preferences by reflecting multiple behavior patterns. A primary challenge in current multi-behavior recommendation tasks is extracting user behavior temporality and behavior discrimination. Most existing studies cannot extract users’ temporal behavioral patterns and analyze the influence and relevance of various types of behaviors. To address this challenge, we propose a hypergraph temporal multi-behavior recommendation framework consisting of a temporal graph convolution network and a behavior-independent hypergraph. Temporal graph convolution network integrates a graph convolution network with a gated recurrent unit to extract the temporality and relationship of user–item interactions, and behavior independent hypergraph groups users and items with similar behavior patterns and analyzes high-order group relationships for user–item interactions. Our proposed framework can capture users’ temporal behavior dynamics and behavior discrimination by reflecting increasingly complex high-order relationships. We performed comparative experiments based on the hit ratio and normalized discounted cumulative gain metrics using three real-world e-commerce datasets and recorded superiority over the baseline model. This proves that the proposed model, hypergraph temporal multi-behavior recommendation, improves the ability to capture the temporality of user behaviors and effectively enhances the differentiation of each behavior.}, archive = {J_EAAI}, author = {Jooweon Choi and JuneHyoung Kwon and Yeonghwa Kim and YoungBin Kim}, doi = {10.1016/j.engappai.2025.110112}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110112}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hypergraph temporal multi-behavior recommendation}, volume = {145}, year = {2025}, }
• (2025). Extended multi-kernel relevance vector machine optimized kriging interpolation for coal seam thickness prediction in coal-bearing strata. EAAI, 145, 110093. (www)

  @article{ , abstract = {During the drilling process of a coal mine roadway drilling rig, coal seam thickness variation affects the efficiency of coal seam mining. However, the cost of geological drilling required for coal seam exploration in practical engineering is high, the sample size of the data is small and the distribution is discrete, and spatial interpolation is required for coal seam thickness prediction in unexplored coal seams. Therefore, this paper proposes an improved method of kriging spatial interpolation for small sample, acquired from geological drilling. Firstly, for the small sample problem, we use a Relevance Vector Machine (RVM) to reconstruct the variogram in kriging interpolation. Secondly, multi-kernel RVM (MKRVM) is used to improve the fitting effect in global and local, respectively. Finally, Particle Swarm Optimization (PSO) is used as an extension of MKRVM to optimize the hyperparameters in the multiple kernel functions and the weights among different kernel functions to improve the fitting effect of the overall model. Through a series of comparative experiments, the superiority of the extended multi-kernel RVM (EMKRVM) method proposed in this paper is verified. At the same time, the method is applied to a practical project, and the results illustrate that our method has lower error in the prediction of coal seam thickness variation in coal-bearing strata, which can provide a better reference basis for the subsequent adjustment of drilling speed, rotation speed, and drilling pressure.}, archive = {J_EAAI}, author = {Luefeng Chen and Mingdi Ma and Min Wu and Witold Pedrycz and Kaoru Hirota}, doi = {10.1016/j.engappai.2025.110093}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110093}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Extended multi-kernel relevance vector machine optimized kriging interpolation for coal seam thickness prediction in coal-bearing strata}, volume = {145}, year = {2025}, }
• (2025). SSmokeDet: A novel network dedicated to small-scale smoke detection. EAAI, 145, 110092. (www)

  @article{ , abstract = {Small smoke detection is essential for the warning of early and distant smoke. However, small-scale smoke occupies few pixels and only provides limited semantic information, causing a considerable challenge for its detection. To this end, we propose a novel network dedicated to small-scale smoke detection (SSmokeDet). Firstly, we put forward a small-net (SNet) backbone to control the receptive field of the model, which facilitates a better observation of the small smoke. Secondly, combined with a residual connection, a multiple spatial pyramid pooling (MultiSPP) is designed to compensate for the lack of small smoke information on the high level by contextual information reinforcement. Lastly, a self-cooperation head (SCHead) is devised for cross-layer communication after refining branching features at different scales. Moreover, an anchor-free mechanism is employed to break the size limitation of predefined anchor boxes and decode the smoke location information directly for the small-scale smoke detection task. Extensive experiments are conducted on both self-made and synthetic databases with various scenes, and the results demonstrate that our SSmokeDet is superior to the state-of-the-art methods. Compared with the baseline, the accuracy of small-scale smoke is effectively improved by 10.2%, and the average precision is increased by 4.9%.}, archive = {J_EAAI}, author = {Jingjing Wang and Li Wang and Runze Zhang and Xiaochuan Li and Baoyu Fan}, doi = {10.1016/j.engappai.2025.110092}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110092}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SSmokeDet: A novel network dedicated to small-scale smoke detection}, volume = {145}, year = {2025}, }
• (2025). Prototypical networks with unlabeled data for few-shot node classification. EAAI, 145, 110088. (www)

  @article{ , abstract = {Due to the high costs of label acquisition in graph, few-shot node classification receives tremendous research interest, which aims to classify unseen labels with limited available nodes given. Although numerous works on few-shot learning have emerged, most of them are established for data following independent identical distribution assumption, which is invalid on graphs where nodes are mutually dependent. Moreover, a majority of existing methods for few-shot node classification are trained with only supervised signals provided by the few labeled samples, neglecting the information implied by a substantial amount of unlabeled data. In this paper, we address the problem of few-shot node classification on the graph, presenting that the distribution of unlabeled nodes can also act as a kind of complementary supervised information to guide the model training. Specifically, we propose a prototype based method that exploits the unlabeled data to provide additional supervised signals for both the encoder and the prototype. We construct a hybrid contrastive loss to leverage self-supervised information from graph topology for encoder enhancement. Meanwhile, the prototype is refined with unlabeled nodes for better reflecting the actual properties of the class. Extensive experiments demonstrate that our method significantly outperforms existing approaches, achieving up to 6.33% higher accuracy and showing superior robustness, with an average performance variation of only 1.58%.}, archive = {J_EAAI}, author = {Ningrui Wang and Yujing Lai and Chuan Chen and Zibin Zheng}, doi = {10.1016/j.engappai.2025.110088}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110088}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prototypical networks with unlabeled data for few-shot node classification}, volume = {145}, year = {2025}, }
• (2025). Research on echo chamber effect and multiple quantum aggregation method in large scale group decision making. EAAI, 145, 110044. (www)

  @article{ , abstract = {Consensus reaching is crucial for group decision making but often neglects the impact of confirmation bias, interference effects, and iterative interactions on the consensus result. To address these gaps, this paper proposes a group consensus decision method incorporating the echo chamber effect and multiple quantum interference. Firstly, to capture the influence of confirmation bias on experts' willingness to adjust, we introduce an echo chamber definition and develop a model to determine adjustment willingness. Secondly, considering iterative interactions between the moderator and decision experts, as well as the interference effects of multi-source information, a bi-level optimization model is proposed to generate adjustment opinions. Then, to facilitate the moderator in making decisions, a maximum deviation evaluation model is constructed, considering both experts' regret or elation over different alternatives and the interference in alternative selection between experts. Finally, the feasibility of the proposed method is demonstrated through case analysis. The results show that the model can reduce the psychological loss of experts by increasing the total cost to a certain extent.}, archive = {J_EAAI}, author = {Yingjie Hu and Jianjun Zhu and Chenyue Tang}, doi = {10.1016/j.engappai.2025.110044}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110044}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on echo chamber effect and multiple quantum aggregation method in large scale group decision making}, volume = {145}, year = {2025}, }
• (2025). A deep learning-based algorithm for fast identification of multiple defects in tunnels. EAAI, 145, 110035. (www)

  @article{ , abstract = {The application of deep learning in tunnel defect detection is common, but it is difficult to take into account both detection accuracy and detection speed. Fast and accurate identification of tunnel damage is the basis for effectively sensing the service state of tunnels as well as subsequent reinforcement, but the commonly used convolutional neural networks and self-attention mechanism networks have their own advantages and disadvantages in terms of speed and accuracy. This study proposes a high-accuracy detection algorithm, Mobile-YOLO (Mobile Friendly-You Look Only Once), based on the YOLOX (You Look Only Once Version X) network architecture to overcome these limitations. Mobile-YOLO integrates an improved MobileViT (Mobile Friendly Vision Transformer) network to extract both local and global features from images, enhancing accuracy and processing speed. A novel GMC (Ghost Multi Concat) block enhances interlayer feature communication and reduces the number of parameters, improving feature extraction. Furthermore, a custom CSA (Channel and Spatial Attention) attention module is included to amplify the significance of the weights in the feature layers. The performance of Mobile-YOLO was evaluated against six other models, on the dataset constructed in this paper, which contains disease images of more than thirty tunnels, Mobile-YOLO achieved a mAP (Mean Average Precision) of 84.69%, an mF1 score of 81.71%, and an FPS (Frames Per Second) of 97.99 f/s, which are different degrees of enhancement compared with the six comparative algorithms, and show better detection accuracy and detection speed, which are more suitable for the disease detection tasks in actual tunnel projects.}, archive = {J_EAAI}, author = {Zhong Zhou and Hongchang Li and Shirong Zhou and Longbin Yan and Hao Yang}, doi = {10.1016/j.engappai.2025.110035}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110035}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep learning-based algorithm for fast identification of multiple defects in tunnels}, volume = {145}, year = {2025}, }
• (2025). Interpretable multi-source data fusion through latent variable gaussian process. EAAI, 145, 110033. (www)

  @article{ , abstract = {With the advent of artificial intelligence and machine learning, various domains of science and engineering communities have leveraged data-driven surrogates to model complex systems through fusing numerous sources of information (data) from published papers, patents, open repositories, or other available resources. However, very little attention has been paid to the differences in quality and comprehensiveness of the known and unknown underlying physical parameters of the information sources, which could have downstream implications during system optimization. Towards resolving this issue, an interpretable multi-source data fusion framework based on Latent Variable Gaussian Process (LVGP) model is proposed. The individual data sources are first labeled as categorical variables and then mapped into a physically meaningful latent space, enabling the development of a source-aware data fusion model. Additionally, a dissimilarity metric based on the learned latent variables of the LVGP is introduced to study and understand the differences between the data sources. The proposed approach is demonstrated on and analyzed through two mathematical and two materials engineering case studies. From the case studies, it is observed that the proposed multi-source data fusion framework provides more accurate predictions for sparse data scenarios compared to single-source or source-unaware data fusion models.}, archive = {J_EAAI}, author = {Sandipp Krishnan Ravi and Yigitcan Comlek and Arjun Pathak and Vipul Gupta and Rajnikant Umretiya and Andrew Hoffman and Ghanshyam Pilania and Piyush Pandita and Sayan Ghosh and Nathaniel Mckeever and Wei Chen and Liping Wang}, doi = {10.1016/j.engappai.2025.110033}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {110033}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpretable multi-source data fusion through latent variable gaussian process}, volume = {145}, year = {2025}, }
• (2025). Fast mining local high-utility itemsets. EAAI, 145, 109960. (www)

  @article{ , abstract = {High-utility itemset mining (HUIM) serves as a refined model addressing the inherent limitation of frequent itemset mining (FIM) which presumes equal importance for all patterns. At its core, HUIM focuses on evaluating the utility of patterns, considering both their intrinsic quantitative values and external factors, thereby revealing their distinct utility in the context of the database. While HUIM has found extensive applications in real-world scenarios, it does not account for the temporal variations in pattern utilities. This led to the introduction of local high-utility itemset mining (LHUIM). LHUIM is designed to identify patterns that, although they might have limited utility across the entire database, exhibit significant utility within certain temporal bounds. More specifically, LHUIM enhances HUIM by gauging a pattern’s utility over specified time spans. In this light, we introduce Fast-LHUIM, an efficient algorithm. Fast-LHUIM leverages two unique data structures: pseudo-projection and vertical lists. It employs two iterative upper bounds strategically to prune unpromising itemsets, optimizing both memory usage and computational overhead. Rigorous evaluations on both real and synthetic datasets highlight Fast-LHUIM’s capability in pinpointing local high utility itemsets (LHUIs) and their pertinent time frames. Significantly, when compared to LHUI-Miner, Fast-LHUIM excels in computational efficiency, with performance improvements ranging from 1.5-fold to several hundred-fold across diverse datasets. All source code and datasets are available at GitHub: https://github.com/DSI-Lab1/Fast-LHUIM .}, archive = {J_EAAI}, author = {Wei Song and Guibin Ren and Wensheng Gan}, doi = {10.1016/j.engappai.2024.109960}, journal = {Engineering Applications of Artificial Intelligence}, month = {4}, pages = {109960}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast mining local high-utility itemsets}, volume = {145}, year = {2025}, }
• (2025). Task-adaptive unbiased regularization meta-learning for few-shot cross-domain fault diagnosis. EAAI, 144, 110200. (www)

  @article{ , abstract = {Model-agnostic meta-learning (MAML) has been successfully integrated with deep learning models and applied to few-shot cross-domain fault diagnosis. However, existing meta-learning strategies often lead to issues such as task conflict and task bias in deep learning models when dealing with new tasks. Therefore, a task-adaptive unbiased regularized meta-learning (TUBR-ML) method is proposed. The proposed method includes the following three innovations: First, we constructed a structured discard feature extractor (SDFE) based on convolutional neural networks (CNN), incorporating a shared feature extraction module to extract correlated features across different tasks, thereby reducing the impact of task differences on training stability and robustness. Second, a task-adaptive attenuation module is introduced to adjust the model's initialization parameters, optimizing the meta-inner loop learning strategy and resolving conflicts between new and old tasks by dynamically adjusting for task-specific requirements. Finally, to address the task bias effect, we propose a task-unbiased regularized loss (TURL), which utilizes a pre-extracted task-unbiased prior to constrain parameter updates, preventing the model from overfitting to existing tasks and enhancing its generalization ability to new tasks. TUBR-ML demonstrates diagnostic accuracy of 92.52%, 96.44%, and 99.26% for the Case Western Reserve University (CWRU) datasets, wind turbine datasets, and rail transit datasets, respectively, under few-shot conditions. Compared to other methods, TUBR-ML shows significant improvement.}, archive = {J_EAAI}, author = {Huaqing Wang and Dongrui Lv and Tianjiao Lin and Changkun Han and Liuyang Song}, doi = {10.1016/j.engappai.2025.110200}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110200}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Task-adaptive unbiased regularization meta-learning for few-shot cross-domain fault diagnosis}, volume = {144}, year = {2025}, }
• (2025). Super-resolution reconstruction of sequential images based on an active shift via a hybrid attention calibration mechanism. EAAI, 144, 110178. (www)

  @article{ , abstract = {Image super-resolution reconstruction converts low-resolution images into high-resolution images, demonstrating extensive potential in processing sequential images. However, most Multi-Image Super-Resolution methods currently face two significant challenges: first, the lack of precision in the shift information between images, as these methods typically rely on algorithms to estimate relative motion. Second, the limited ability to effectively extract subpixel features from low-resolution images directly impacts the richness of details in reconstructed images. This paper proposes a novel active shift-based sequential image super-resolution reconstruction technique to address these issues. This technique integrates hardware control with deep learning algorithms, utilizing a Piezoelectric platform to control camera movement precisely, capturing sequential images with predetermined subpixel shifts, and accurately recording the relative shifts between images. At the algorithmic level, we have designed a hybrid network model that combines a convolutional neural network with a Transformer architecture and integrates channel attention and self-attention mechanisms. This model fully leverages the precise shift information provided by the hardware and significantly enhances the ability to extract image details and overall image quality. Experimental results demonstrate that our method outperforms single-image super-resolution techniques regarding Peak-Signal-to-Noise-Ratio and Structural Similarity Index Measure. To further validate the applicability and effectiveness of this technology, we conducted tests using a resolution test chart, which showed that our technique can increase the resolution of the original imaging system by 25.6%. Therefore, the strategy combining hardware and software proposed in this paper effectively solves critical issues in Multi-Image Super-Resolution tasks and provides new pathways for image processing technologies.}, archive = {J_EAAI}, author = {Qiang Wu and Ziyi Yang and Hongfei Zeng and Jin Zhang and Haojie Xia}, doi = {10.1016/j.engappai.2025.110178}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110178}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Super-resolution reconstruction of sequential images based on an active shift via a hybrid attention calibration mechanism}, volume = {144}, year = {2025}, }
• (2025). Revisiting sensor-based intelligent fall risk assessment for older people: A systematic review. EAAI, 144, 110176. (www)

  @article{ , abstract = {Falls are a major public health concern among older people due to their high prevalence and severe consequences. Identifying individuals at high risk of falling through fall risk assessment is a fundamental step to implement effective fall prevention strategies. Recent advancements in off-the-shelf human sensing technologies have spurred a surge in sensor-based intelligent fall risk assessment. Existing reviews often overlook non-wearable technologies, free-living environments, and geriatric populations. To address these limitations and capture emerging research trends, this systematic review provides a comprehensive analysis of current literature and outlines future research prospects. Thirty-two relevant papers retrieved from major databases were critically reviewed and analyzed, presenting a variety of faller identification criteria, experimental cohorts, sensing devices, test protocols, artificial intelligence (AI) modeling techniques. Even though accumulated evidence from this review demonstrated that sensor technologies (e.g., inertial sensor, depth camera, radar, pressure sensor) combined with AI algorithms hold significant promise for objective, accurate and convenient fall risk assessment, inconsistencies in study methodologies hinder definitive conclusions about their ability to predict future falls. Moreover, the overall performance on large-scale cohorts remains relatively poor, with a mean accuracy of 63.9%. Additionally, practical standalone applications integrating motion sensing, model inference, and diagnostic reports are underdeveloped, hindering widespread deployment of fall risk assessment among older population. Future research should focus on high-risk geriatric populations, contactless and low-cost motion sensing, prospective multi-center protocol design, multifactorial test protocols, advanced AI models with explainable mechanisms, and user-centric applications to enhance fall risk assessment for older people.}, archive = {J_EAAI}, author = {Xiaoqun Yu and Yuqing Cai and Rong Yang and Fengling Ma and Woojoo Kim}, doi = {10.1016/j.engappai.2025.110176}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110176}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Revisiting sensor-based intelligent fall risk assessment for older people: A systematic review}, volume = {144}, year = {2025}, }
• (2025). A cascaded deep neural network for design and verification of surface lattice resonance metasurfaces biosensors. EAAI, 144, 110172. (www)

  @article{ , abstract = {Artificial neural networks (ANNs) have gradually replaced labor-intensive tasks with their revolutionary advantages by extracting features from datasets. It has been shown that researchers can perform only one traditional finite-difference time-domain (FDTD) simulation and then utilize the neural network for data analysis. This study presents the spectral parameter multi-layer perceptron neural network (SPMLP), designed for the rapid prediction of optical responses in biosensing chips characterized by surface lattice resonance (SLR), achieving speeds thousands of times faster than conventional FDTD simulations. To obtain structures with a high-quality factor (Q) and low full width at half maximum (FWHM), multi-objective particle swarm optimization (MOPSO) techniques are employed to develop a high Q and low FWHM (HQLW) algorithm. This algorithm constructs a Transformer-based inverse network that translates structures into spectra and utilizes the spectral and sensing parameter extraction (SASPE) algorithm to derive essential parameters such as resonance wavelength (λres), FWHM, and Q values. After structures are generated through the HQLW algorithm, spectra are obtained via the forward network, with continuous iterations refining structures until the desired parameters are achieved. Unlike previous efforts, the SLR-based neural network allows for user-defined sensing metrics and structural parameters, achieving a prediction accuracy of 97% and effective chips after 400 iterations.}, archive = {J_EAAI}, author = {Yian Liu and Qingfubo Geng and Weihe Zhan and Zhaoxin Geng}, doi = {10.1016/j.engappai.2025.110172}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110172}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cascaded deep neural network for design and verification of surface lattice resonance metasurfaces biosensors}, volume = {144}, year = {2025}, }
• (2025). Optimizing the selection of the secret parameters for public key cryptosystems by using interval linear programming and fully fuzzy linear programming. EAAI, 144, 110168. (www)

  @article{ , abstract = {One of the basic tasks in public key cryptosystems is to determine the secret parameters. This paper proposes multi-objective interval linear programming and multi-objective fully fuzzy linear programming (FFLP) models to determine the optimum interval numbers and triangular fuzzy numbers (TFNs) of the secret parameters in the public key cryptosystems. The public and private (secret) objective function(s) and constraints can be considered such that the optimum interval numbers and optimum TFNs for the secret parameters can be obtained from which the random selection of the secret parameters can be made for each algorithm's execution. Adapting the general model to the Rivest, Shamir, Adleman (RSA) public key cryptosystem is also illustrated. In the interval programming model for the RSA, the linear combination method combines multiple objective functions into a single objective function, and nonlinear constraints are linearized using the Taylor series expansion method. No multiple objective functions are obtained in the FFLP model of the RSA, but the nonlinear constraints are again linearized using the Taylor series expansion method. The numerical application for the RSA public key cryptosystem is made for both models by considering five cases and twenty-four subcases, and the results are analyzed. To the best of our knowledge, there is no study in the literature that considers interval linear programming and/or FFLP to obtain the optimum interval numbers and TFNs for the secret parameters of the public key cryptosystems. The proposed methodology is systematic and generic and can also be adapted for other public key cryptosystems easily.}, archive = {J_EAAI}, author = {Esra Bas}, doi = {10.1016/j.engappai.2025.110168}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110168}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing the selection of the secret parameters for public key cryptosystems by using interval linear programming and fully fuzzy linear programming}, volume = {144}, year = {2025}, }
• (2025). A parameter-free self-training algorithm based on the three successive confirmation rule. EAAI, 144, 110165. (www)

  @article{ , abstract = {Semi-supervised learning is a popular research topic today, and self-training is a classical semi-supervised learning framework. How to select high-confidence samples in self-training is a critical step. However, the existing algorithms do not consider both global and local information of the data. In the paper, we propose a parameter-free self-training algorithm based on the three successive confirmation rule, which integrates global and local information to identify high-confidence samples. Concretely, the local information is obtained by using k nearest neighbors and global information is derived from the three successive confirmation rule. This dual selection strategy helps to improve the quality of high-confidence samples and further improve the performance of classification. We conduct experiments on 14 benchmark datasets, comparing our method with other self-training algorithms. We use accuracy and F-score as performance metrics. The experimental results demonstrate that our algorithm significantly improves classification performance, proving its effectiveness and superiority in semi-supervised learning.}, archive = {J_EAAI}, author = {Jikui Wang and Wei Zhao and Qingsheng Shang and Feiping Nie}, doi = {10.1016/j.engappai.2025.110165}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110165}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A parameter-free self-training algorithm based on the three successive confirmation rule}, volume = {144}, year = {2025}, }
• (2025). An efficient solution to the simple assembly line balancing problem type 1 using iterated local search. EAAI, 144, 110162. (www)

  @article{ , abstract = {The Assembly Line Balancing (ALB) problem involves dividing total assembly operations into tasks and assigning them to workstations so that each workstation has approximately equal times and precedence constraints are met. The most fundamental and well-studied model in ALB is called Simple ALB Problem Type 1 (SALBP-1), which aims to minimize the number of loaded stations given a fixed cycle time. SALBP-1 is proven to be NP-hard and thus its efficient solution is challenging. In this paper, we developed an Iterated Local Search (ILS) algorithm customized for obtaining high-quality and feasible solutions to SALBP-1 by minimizing the number of assembly stations. We tested our method on 270 benchmark problems from Talbot's, Scholl's, and Otto's datasets, as well as on a real antenna assembly case, and compared the results with nine algorithms from the literature. Analyses showed that the ILS outperformed other algorithms by reaching or improving the best-known upper-bound solutions in 269 of the 270 instances within lower computational runtimes. Key components that contribute to the success of the ILS include the way the initial solution is selected, innovative shallow and deep task assignment implementation, and the diversity of neighborhood generating operators (including the novel Float Shift operator), which help the algorithm to converge toward feasible solutions and achieve minimal values for the objective function. The proposed method enables designers and engineers to design more efficient assembly lines and reduce production lead times and overall costs.}, archive = {J_EAAI}, author = {Somayé Ghandi and Ellips Masehian}, doi = {10.1016/j.engappai.2025.110162}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110162}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient solution to the simple assembly line balancing problem type 1 using iterated local search}, volume = {144}, year = {2025}, }
• (2025). Two-stage type-2 fuzzy parabolic double frontier data envelopment analysis. EAAI, 144, 110154. (www)

  @article{ , abstract = {Data Envelopment Analysis (DEA) has traditionally assessed decision-making units (DMUs) by treating them as “black boxes,” ignoring their internal structures and thereby limiting the depth of analysis. To address this limitation, network DEA models have been developed to account for the internal complexities of DMUs. However, conventional DEA models classify DMUs as efficient or inefficient without distinguishing among efficient units. This study introduces an innovative DEA framework that integrates network two-stage DEA and double-frontier DEA with a Pythagorean approach. This model evaluates optimistic and pessimistic efficiencies by optimizing inputs and outputs through perpendicular projections onto the efficient frontier, enabling a more nuanced performance assessment and classification of DMUs. Additionally, the study incorporates a type-2 fuzzy parabolic variable to address real-world uncertainty in DEA applications, further enhancing the network DEA model. A two-stage defuzzification process is employed: the convex reduction method converts the type-2 variable into a type-1 fuzzy number, followed by the center-of-gravity method to obtain a crisp value. The effectiveness of these proposed models is demonstrated through three numerical examples, showcasing their analytical and practical contributions across crisp, type-2 fuzzy, and fuzzified datasets in the insurance, petrochemical, and banking sectors.}, archive = {J_EAAI}, author = {Mohammad Aqil Sahil and Ashutosh Tiwari and Q.M. Danish Lohani}, doi = {10.1016/j.engappai.2025.110154}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110154}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Two-stage type-2 fuzzy parabolic double frontier data envelopment analysis}, volume = {144}, year = {2025}, }
• (2025). Synthesizing spreading-out features for generative zero-shot image classification. EAAI, 144, 110151. (www)

  @article{ , abstract = {Suffering from the singularity of generated samples, the most of the existing generative models cannot achieve satisfactory performance on Generalized Zero-shot Learning tasks. Generative methods applied to generalized zero-shot learning mainly focus on the data generation of the entire class domain, and do not pay enough attention to the intra-instance relationships, which severely limits the robustness of generators. Therefore, we propose a model that mainly utilizes Cramer distance and Wasserstein Generative Adversarial Network -divergence to generate various diversified visual features, which effectively alleviate the domain shift problem and facilitates better classification than traditional Wasserstein distance and its variants. In addition, we also use perturbation-based attack strategy to standardize the range of generated features, so as not to generate too grotesque features and lead to incorrect classification results. Finally, supported by the above, we utilize two different classifiers to obtain better Generative Zero-shot Learning performance, i.e., the traditional softmax classifier and the normalized prototypes classifier. Extensive experimental results show that our proposed method can outperform the most of state-of-the-art methods on five benchmark datasets in generalized zero-shot learning setting and various measurement criteria.}, archive = {J_EAAI}, author = {Jingren Liu and Ke Sun and Zheng Zhang and Yang Long and Wankou Yang and Yunyang Yan and Haofeng Zhang}, doi = {10.1016/j.engappai.2025.110151}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110151}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Synthesizing spreading-out features for generative zero-shot image classification}, volume = {144}, year = {2025}, }
• (2025). Integrating prompt techniques and multi-similarity matching for named entity recognition in low-resource settings. EAAI, 144, 110149. (www)

  @article{ , abstract = {Few-shot Named Entity Recognition (few-shot NER) is a technique that effectively trains models with limited annotated data, aiming to address the issue of low accuracy in traditional named entity recognition tasks due to sparse data. Existing methods often pay little attention to specific entity information when constructing prompts, which can challenge the model's understanding of semantic relationships between labels and entities. Metric learning methods rely on a single similarity metric, potentially limiting their ability to capture complex relationships between support and query sets. To address these issues, this paper proposes a novel method named Integrating Pro mpt Techniques and M ulti- S imilarity M atching for Named Entity Recognition in Low-Resource Settings (ProMSM). This method combines entity-guided masked prompt techniques with a multi-similarity fusion mechanism. The entity-guided masked prompt aims to fully utilize entity information from the support set or source domain training set. It integrates labels and their associated entity instances into input sequences, providing rich contextual information to enhance the model's accurate understanding of semantic relationships between labels and entities. The multi-similarity fusion technique integrates traditional dot product similarity with context-based custom attention scores, enhancing the precision of relationship measurement between tokens in the support and query sets. Finally, ProMSM was evaluated on five widely used public few-shot datasets under cross-label space and domain transfer settings. Extensive experimental results demonstrate that ProMSM outperforms previous few-shot NER methods.}, archive = {J_EAAI}, author = {Jun Yang and Liguo Yao and Taihua Zhang and Chieh-Yuan Tsai and Yao Lu and Mingming Shen}, doi = {10.1016/j.engappai.2025.110149}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110149}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating prompt techniques and multi-similarity matching for named entity recognition in low-resource settings}, volume = {144}, year = {2025}, }
• (2025). Integrating explainable artificial intelligence in machine learning models to enhance the interpretation of elastic behaviors in three-dimensional-printed triangular lattice plates. EAAI, 144, 110148. (www)

  @article{ , abstract = {Three-dimensional printing provides significant manufacturing flexibility, facilitating the creation of intricate geometries. This process commonly involves fabricating porous products constructed from lattice structures comprising numerous unit cells. These lattice structures have gained popularity for their capacity to reduce weight and material consumption while preserving specific mechanical strength. However, optimizing their elastic properties requires a thorough understanding of how unit cell design influences the overall performance of these structures. This study introduces a novel approach combining finite element simulations with machine learning (ML) to predict and analyze the elastic moduli of two-dimensional triangular lattice structures. Using the Extreme Gradient Boosting algorithm, an ML model was developed to predict moduli based on simulation data, aiming to establish a precise relationship between relative density and elastic moduli. A key objective is to provide an efficient estimation tool for researchers and engineers, verified through comparisons with existing closed-form solutions and experimental data. To enhance trust in predictions, Explainable Artificial Intelligence techniques, including SHapley Additive exPlanations and Partial Dependence Plots, were employed to interpret the model's outputs and identify influential design variables. Findings revealed a non-linear relationship between relative density and elastic moduli, with ML models achieving high predictive accuracy. This approach offers valuable insights for optimizing lattice structures across diverse applications, emphasizing the critical design features influencing their elastic behavior.}, archive = {J_EAAI}, author = {Itthidet Thawon and Pana Suttakul and Ramnarong Wanison and Yuttana Mona and Korrakot Yaibuathet Tippayawong and Nakorn Tippayawong}, doi = {10.1016/j.engappai.2025.110148}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110148}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating explainable artificial intelligence in machine learning models to enhance the interpretation of elastic behaviors in three-dimensional-printed triangular lattice plates}, volume = {144}, year = {2025}, }
• (2025). Unveiling learning trends in convolutional neural networks with dynamic mode decomposition. EAAI, 144, 110146. (www)

  @article{ , abstract = {The widespread adoption of pre-trained models in artificial intelligence poses several evaluation challenges, primarily due to limited access to comprehensive information including training data, test data, loss function, and hyperparameter values. Traditional evaluation metrics, like accuracy, precision, and recall, demand the availability of train and test data, making it difficult to determine the expected performance or quality of the deep learning models. This paper addresses this challenge by introducing a comprehensive empirical analysis of Convolutional Neural Networks (CNNs) using Dynamic Mode Decomposition (DMD) theory. Feature maps from the convolutional layers are used to model the trained CNN as a linear dynamical system. By constructing snapshots for DMD analysis, we decompose the system’s dynamics into individual modes and eigenvalues, providing a linear approximation of the intricate CNN behavior. We also introduce DMD-based metrics, including entropy of DMD-eigenvalues (EDMD), DMD eigen gap/spectral gap, and zero amplitude DMD modes, to quantify the effectiveness and quality of the learned CNN (well-trained vs poorly-trained) without the need to access the test data. Results indicate that the entropy of DMD-eigenvalues (EDMD) correlates well with the classical evaluation metrics and also distinguishes between well-trained and poorly-trained models. Additionally, we explored the potential of interpreting CNNs through stable DMD modes and their corresponding eigenvalues. Our approach was thoroughly evaluated on both shallow and deep models across various datasets for classification and segmentation tasks, demonstrating its effectiveness in quantifying CNN learning evolution and task performance. We also conducted a comprehensive analysis to examine the correlation between the proposed metrics and classical evaluation metrics on pre-trained models, specifically using ImageNet for classification (20 models) and Pascal VOC for segmentation (3 models) tasks. The code is available at: https://github.com/sikha2552/DMD_XAI .}, archive = {J_EAAI}, author = {O.K. Sikha and Miguel A. González Ballester and Raul Benitez}, doi = {10.1016/j.engappai.2025.110146}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110146}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unveiling learning trends in convolutional neural networks with dynamic mode decomposition}, volume = {144}, year = {2025}, }
• (2025). Explainable artificial intelligence models in intrusion detection systems. EAAI, 144, 110145. (www)

  @article{ , abstract = {This study introduces the growing necessity of Explainable Artificial Intelligence (XAI) in the context of cybersecurity, particularly in Intrusion Detection Systems (IDSs). As Artificial Intelligence (AI) technologies, especially complex neural networks, become more prevalent and difficult to understand, the opacity of these models poses increasing risks in critical firald like cybersecurity. XAI aims to address these concerns by making AI models and their outputs more interpretable and transparent, thus fostering trust and secure use. Despite the rising interest and application of XAI in various domains, its integration into IDSs remains insufficient. This research provides a comprehensive overview of XAI, examining its definitions, terminologies, and the evolution of the field. It delves into the opportunities, challenges, and research needs within XAI, as well as the latest developments, tools, and technologies for implementing XAI in AI-based IDSs, alongside associated risks. The study also summarizes and demonstrates a gap in the literature regarding comprehensive reviews of IDS solutions that incorporate XAI, aiming to fill this gap by detailing the application of XAI in IDS research. The findings are expected to guide researchers, experts, users, and policymakers in the domains of AI, XAI, and IDS.}, archive = {J_EAAI}, author = {Samed AL and Seref Sagiroglu}, doi = {10.1016/j.engappai.2025.110145}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110145}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable artificial intelligence models in intrusion detection systems}, volume = {144}, year = {2025}, }
• (2025). Improving predictive maintenance: Evaluating the impact of preprocessing and model complexity on the effectiveness of eXplainable artificial intelligence methods. EAAI, 144, 110144. (www)

  @article{ , abstract = {Due to their performance in this field, Long-Short-Term Memory Neural Network (LSTM) approaches are often used to predict the remaining useful life (RUL). However, their complexity limits the interpretability of their results. So, eXplainable Artificial Intelligence (XAI) methods are used to understand the relationship between the input data and the predicted RUL. Modeling involves making choices, such as preprocessing strategies or model complexity. Understanding how these modeling choices affect the effectiveness of XAI methods is crucial. This paper investigates the impact of two modeling aspects: preprocessing multivariate time series and model complexity, precisely the number of hidden layers, on the quality of the explanations provided by three XAI post-hoc local agnostic methods (Local Interpretable Model-Agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP), and Learning to eXplain (L2X) in the context of the RUL prediction. The quality of the XAI methods is evaluated using eleven metrics, categorized under five properties based on the definitions of interpretability and explainability. Experiments on the C-MAPSS dataset for aero-engine prognostics demonstrate that SHAP often provides better explanations when optimized preprocessing parameters are used. However, variations in these preprocessing parameters affect the quality of the explanation. Additionally, the results suggest no significant correlation between the complexity of the LSTM model and explanation quality, although changes in the number of layers notably influence the precision of SHAP’s explanations.}, archive = {J_EAAI}, author = {Mouhamadou Lamine Ndao and Genane Youness and Ndeye Niang and Gilbert Saporta}, doi = {10.1016/j.engappai.2025.110144}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110144}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Improving predictive maintenance: Evaluating the impact of preprocessing and model complexity on the effectiveness of eXplainable artificial intelligence methods}, volume = {144}, year = {2025}, }
• (2025). Adaptable, incremental, and explainable network intrusion detection systems for internet of things. EAAI, 144, 110143. (www)

  @article{ , abstract = {The growing adoption of Internet of Things (IoT) devices expands the cybersecurity landscape and complicates the protection of IoT environments. Therefore, Network Intrusion Detection Systems (NIDSs) have become essential. They are increasingly using Machine and Deep Learning (ML and DL) techniques for detecting and mitigating sophisticated cyber threats. However, the black-box nature of these systems hinders adoption, emphasizing the need for eXplainable Artificial Intelligence (XAI) to clarify decision-making. Additionally, IoT networks require adaptable NIDSs integrating new traffic types without retraining. This study integrates XAI with Class Incremental Learning (CIL) and Domain Incremental Learning (DIL) to improve NIDS transparency and adaptability. This work focuses on training NIDSs with traffic from a source network and extending it to a target network. For the sake of generalization, three recent and publicly available IoT security datasets are leveraged. Each dataset is collected in a different network setup and includes different attacks and benign profiles. Key findings include: ( i ) NIDSs perform effectively within the source network ( > 79 % F1 score) but poorly in the target one (33% F1 score at least); ( i i ) adapting NIDSs incrementally is highly dependent on the source network traffic, with richer traffic complicating the adaptation. Incremental techniques help in adapting NIDSs ( > 71 % F1 score), with Fine-Tuning with Memory ( F T - M e m ) excelling for complex source networks and Bias Correction ( B i C ) for simpler ones; ( i i i ) in terms of XAI, traffic characteristics significantly influence classification outcomes, and NIDS decisions are not based on minimal-distance logic.}, archive = {J_EAAI}, author = {Francesco Cerasuolo and Giampaolo Bovenzi and Domenico Ciuonzo and Antonio Pescapè}, doi = {10.1016/j.engappai.2025.110143}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110143}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptable, incremental, and explainable network intrusion detection systems for internet of things}, volume = {144}, year = {2025}, }
• (2025). Interpretable and multimodal fusion methodology to predict severe hypoglycemia in adults with type 1 diabetes. EAAI, 144, 110142. (www)

  @article{ , abstract = {Type 1 diabetes (T1D) causes insulin deficiency and exogenous therapy is required for maintaining targeted glucose levels. Hypoglycemia is the most frequent side effect of insulin, being severe hypoglycemia (SH) one of the most critical hazards with a range of life-threatening consequences. Artificial intelligence (AI) and multimodal fusion have boosted predictive performance in different domains. This study aims to evaluate the effectiveness of early fusion (EF) and late fusion (LF) approaches for predicting SH, to create a methodology capable of achieving robust results in datasets with a low number of samples for predicting SH and to characterize the risk factors involved in the SH onset using explainable AI (XAI). Data from a case-control study comprising adults over 60 years with T1D and with diabetes duration of 20 years were used and three types of modalities were considered: (1) continuous glucose monitoring data (time series); (2) clinical codes (text); and (3) surveys related to fear, unawareness, depression, and cognitive tests (tabular data). The results revealed that EF outperformed models trained with single-modality data by 5.8%, with an area under the receiver operating characteristic curve of 0.779. XAI techniques helped to discover that features related to fear and unawareness are mainly associated with SH. Our study introduced an interpretable and multimodal methodology capable of predicting the occurrence of SH in adults with T1D in the next year. Our interpretable methodology contributes to predicting SH and identifying related key factors, thus preventing SH complications and improving patient’s quality of life.}, archive = {J_EAAI}, author = {Francisco J. Lara-Abelenda and David Chushig-Muzo and Ana M. Wägner and Maryam Tayefi and Cristina Soguero-Ruiz}, doi = {10.1016/j.engappai.2025.110142}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110142}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpretable and multimodal fusion methodology to predict severe hypoglycemia in adults with type 1 diabetes}, volume = {144}, year = {2025}, }
• (2025). Multi-modal explainable artificial intelligence for neural network-based tool wear detection in machining. EAAI, 144, 110141. (www)

  @article{ , abstract = {As data in manufacturing becomes more accessible and larger in scope, solutions are required to explain the underlying reasoning behind predictive models that use data from various modalities. This research aims to develop multi-modal Explainable Artificial Intelligence approaches to facilitate human-interpretable solutions for tool wear prediction in machining. Initially, a multi-modal neural network was used as the supervised Machine Learning classifier for training and binary classification by applying a set of features comprised of different modalities. The data consisted of two sets of image data representing the Flank and the Rake views of the tools accompanied by time series data including acceleration, acoustics, temperature, spindle speed, and feed rate during the orthogonal tube turning process. The classifier was used to predict the condition of the tool after the cutting process. After the training process and the network performance evaluation were completed, two multi-modal neural network explainability approaches were investigated. The Full Multi-Modal Explainable Artificial Intelligence approach provided a general explainability of the feature importance based on latent space representations of the input data while the Decomposed Multi-Modal Explainable Artificial Intelligence approach produced explainability results of input feature importance by decomposing the multi-modal model into its constituent single modal sub-models. The developed explainability methods were demonstrated to provide sufficient information regarding explaining the multi-modal network performance and the decision-making processes. Hence, the presented methods will enable end-users to understand the underlying logic for advanced neural network-based models that incorporate data from different modalities through the selection of multiple algorithms.}, archive = {J_EAAI}, author = {Saleh Valizadeh Sotubadi and Shyam Sasi Pallissery and Vinh Nguyen}, doi = {10.1016/j.engappai.2025.110141}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110141}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modal explainable artificial intelligence for neural network-based tool wear detection in machining}, volume = {144}, year = {2025}, }
• (2025). A hybrid error correction approach for prediction of credit approval: An explainable artificial intelligence approach. EAAI, 144, 110140. (www)

  @article{ , abstract = {One of the main issues faced in classification problems is the lack of prediction accuracy. This can be due to the use of unsuitable features for certain types of data. It is important to establish a framework that can identify the most important features, even if that means reducing the number of features. However, reducing the number of features can also lead to more prediction errors, so it is crucial to develop an approach that can select features and correct errors simultaneously. This is especially important in processing financial data and making financial decisions, as any errors can have major consequences. The paper introduces a heuristic approach that utilizes eXplainable Artificial Intelligence (XAI) in order to address this issue. Specifically, the paper uses the SHapley Additive exPlanations (SHAP) method, which is designed to identify and predict problems with credit scoring and approval. The main objective of this approach is to correct errors that may occur in conventional classification methods. The results show that, despite strict parameter tuning and selecting primary models with the highest accuracy, the proposed method has effectively corrected classification errors. The findings also demonstrate that this approach can improve error correction rates and model accuracy when applied to datasets such as the German Statlog and Australian Credit Approval datasets.}, archive = {J_EAAI}, author = {Elham Darvish and Mustafa Jahangoshai Rezaee and Mohsen Abbaspour Onari}, doi = {10.1016/j.engappai.2025.110140}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110140}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid error correction approach for prediction of credit approval: An explainable artificial intelligence approach}, volume = {144}, year = {2025}, }
• (2025). Deep reinforcement learning explanation-assisted integer variable reduction method for security-constrained unit commitment. EAAI, 144, 110139. (www)

  @article{ , abstract = {The large-scale security-constrained unit commitment (SCUC) is pivotal for ensuring the secure and economical operation of modern power systems. Formulated as a mixed-integer nonlinear programming problem, mathematical model-based methods struggle to balance computation efficiency and solution accuracy. While artificial intelligence methods offer promising potential, they face several obstacles, including limited interpretability and generalizability constraints. In light of these challenges, this paper proposes an interpretation method for deep reinforcement learning models that is used to reduce integer variables for large-scale SCUC problem. This method employs a Gaussian Mixture Model to cluster the decision outcomes of the agents and utilizes an improved decision tree to interpret the clustering results. We analyze the physical implications behind the phenomenon of unit output distributions exhibiting multiple independent Gaussian distributions. Then, these interpretations are applied to identify active integer variables, thereby simplifying the complexity of the SCUC problem and enhancing solution efficiency. Furthermore, an improved Markov decision process model with domain knowledge pertinent of power systems is constructed to enhance the interpretability and reliability of the agents. A distinctive feature of this model is the incorporation of a bidirectional mapping of unsafe and safe actions. The case studies on the SG-126 system demonstrate that the proposed method achieves a significant increase in solution speed without loss of accuracy. The identified active integer variables are proven to be accurate and effective, contributing to improve computation efficiency of unit commitment. The proposed method also provides a novel explainable artificial intelligence-assisted method for complex decision-making problems in other fields.}, archive = {J_EAAI}, author = {Yuchen Dai and Wei Xu and Minghui Yan and Feng Xue and Jianfeng Zhao}, doi = {10.1016/j.engappai.2025.110139}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110139}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep reinforcement learning explanation-assisted integer variable reduction method for security-constrained unit commitment}, volume = {144}, year = {2025}, }
• (2025). An explainable artificial intelligence and internet of things framework for monitoring and predicting cardiovascular disease. EAAI, 144, 110138. (www)

  @article{ , abstract = {Cardiovascular disease (CVD) is a leading cause of death globally. The unpredictability and severity of CVDs, such as sudden cardiac arrests, necessitate real-time monitoring and prediction using the Internet of Things (IoT) and artificial intelligence (AI) for timely intervention. Existing AI models and IoT frameworks for CVD prediction often lack integration of diverse data and fail to provide transparency in predictions, reducing user confidence and treatment effectiveness. We propose an explainable-IoT framework leveraging eXplainable AI (XAI) in the cloud layer and a mobile application to swiftly communicate predictions and explainability to patients, healthcare providers, and other users, facilitating proactive management and informed decisions. The framework integrates sensor data with cloud-based medical records to improve cardiovascular care and build user trust. Using support vector machine (SVM), random forest (RF), k-nearest neighbor (KNN), and deep neural network (DNN), we develop and evaluate CVD prediction models on two datasets: a heart disease dataset (D1) and the Cleveland dataset (D3). Additionally, we use a synthetic dataset (D2) for comparative analysis. The models are evaluated based on accuracy, precision, recall, area under the curve, time, and F1 score. For D3, SVM achieved the best accuracy (84.62%), while RF performed best on D1 and D2 (92.44% and 98.06%, respectively), comparable to state-of-the-art works. Our results highlight the importance of underrepresented physiological features in CVD datasets and the need for comprehensive datasets to enhance CVD model development. Furthermore, while synthetic data (D2) is effective for initial modeling, it requires validation with real-world data for reliable CVD prediction.}, archive = {J_EAAI}, author = {Mubarak Albarka Umar and Najah AbuAli and Khaled Shuaib and Ali Ismail Awad}, doi = {10.1016/j.engappai.2025.110138}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110138}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An explainable artificial intelligence and internet of things framework for monitoring and predicting cardiovascular disease}, volume = {144}, year = {2025}, }
• (2025). A novel hybrid group method of data handling and levenberg marquardt model for estimating total organic carbon in source rocks with explainable artificial intelligence. EAAI, 144, 110137. (www)

  @article{ , abstract = {The estimation of Total Organic Carbon (TOC) in source rocks is critical for assessing hydrocarbon potential and source rock quality in petroleum exploration and production. This study presents a novel hybrid model of the Group Method of Data Handling and Levenberg Marquardt (GMDH-LM), which integrates mineralogical and well-log data to improve TOC prediction accuracy. The GMDH-LM model employs non-parametric regression, offering superior adaptability and accuracy in solving non-linear problems, even with small datasets. The results show that the GMDH-LM model achieved a higher coefficient of determination (R 2 ) of 0.9784, lower Root Mean Square Error (RMSE) of 0.1162, Mean Absolute Percentage Error (MAPE) of 0.3722 and less computation time of 0.73 s. Furthermore, the integration of mineral constituents into the model significantly improved the accuracy to R 2 = 0.9983 as compared to where only well log data is used. Additionally, the application of Shapley Additive Explanation (SHAP) analysis presents a key innovation in this work as a novel method in TOC estimation, offering unprecedented transparency and interpretability of the GMDH-LM model by identifying gamma ray (GR), calcite, K-feldspar and deep lateral resistivity (LLD) as the most impactful input variables, thus contributing to the model's trustworthiness, reliability, and overall performance. This interpretability enhances model reliability and supports informed decision-making. The proposed GMDH-LM model represents a significant advancement in hybrid machine learning applications, offering a robust and efficient tool for TOC estimation. Its scalability and precision make it valuable for petroleum reservoir characterization and exploration, with broader implications for optimizing resource management in the oil and gas industry.}, archive = {J_EAAI}, author = {Christopher N. Mkono and Chuanbo Shen and Alvin K. Mulashani and Emmanuel John M. Carranza and Daud C. Kalibwami and Magori J. Nyangi}, doi = {10.1016/j.engappai.2025.110137}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110137}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel hybrid group method of data handling and levenberg marquardt model for estimating total organic carbon in source rocks with explainable artificial intelligence}, volume = {144}, year = {2025}, }
• (2025). Insights into the application of explainable artificial intelligence for biological wastewater treatment plants: Updates and perspectives. EAAI, 144, 110132. (www)

  @article{ , abstract = {Explainable artificial intelligence (XAI) is an interactive platform that assists users in comprehending the decisions and predictions made by machine learning (ML) models. This allows users to enhance their knowledge of ML models and their functioning, which not only helps in mitigating bias and errors but also aids in improving user decision-making confidence. XAI, due to its ability to increase the model output interpretation, has gained significant attention in biological wastewater treatment plants (WWTPs). This is owing, in particular, to the fact that it facilitates the experts in steering knowledge about the predictions and decisions made by ML, thus guaranteeing that the model decisions are fair and unbiased. ML has made amazing advances in recent years, thanks to its exponential growth in possessing the power to process massive volumes of data, allowing it to be widely embraced in WWTPs. This review seeks to illustrate the potential of XAI for WWTP applications such as process modeling and control, soft sensing, fusion of data, and the internet of things, and fill the knowledge gap by thoroughly introducing XAI techniques and their use in smart wastewater engineering. Overall, the features of XAI can aid in establishing reliable and efficient water resource management, which is quintessential to achieving environmental sustainability. It is envisioned that the prospects offered would spark new lines of study, helping to reduce the current skepticism and apprehension about ML adoption and integration in WWTP.}, archive = {J_EAAI}, author = {Abdul Gaffar Sheik and Arvind Kumar and Chandra Sainadh Srungavarapu and Mohammad Azari and Seshagiri Rao Ambati and Faizal Bux and Ameer Khan Patan}, doi = {10.1016/j.engappai.2025.110132}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110132}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Insights into the application of explainable artificial intelligence for biological wastewater treatment plants: Updates and perspectives}, volume = {144}, year = {2025}, }
• (2025). Enhanced audio classification leveraging pre-trained deep visual models. EAAI, 144, 110130. (www)

  @article{ , abstract = {The differentiation between speech and music poses a prevalent issue in audio analytic, specifically in dividing audio streams into segments and accurately labeling them as either speech or music. This problem holds significant value across various applications, such as automatic speech recognition and monitoring radio broadcasts. This paper addresses the limitations of existing deep learning-based feature extractors in tackling this problem. Specifically, it investigates the classification performance of five pre-trained deep learning models (SqueezeNet, Alexnet, GoogleNet, Efficient-b0, and Darknet) when utilizing visual features such as Spectrogram and Chromagram extracted from audio signals for speech/music classification. The pre-trained networks, originally trained on millions of images, were fine-tuned on a small subset of data generated from the audio files. We thoroughly tested our system using Scheirer and Slaney (S&S) and George Tzanetakis(GTZAN) datasets to assess its performance. Multiple experiments were conducted to understand the variation of performance of the models for different hyper-parameters. The aim was to filter the best-performing models from the five pre-trained networks for speech/music classification. The best-performing models(GoogleNet, Darkenet-19) were thoroughly investigated under different training scenarios. Best classification accuracies of 99.81% and 99.72% were observed for S&S and GTZAN datasets, respectively, for hybrid visual features. Additionally, the shortlisted networks were also tested on unseen audio files gathered from recordings of Indian radio stations, and satisfactory performance was observed. We also studied the performance of the best-performing models on three different types of noise (white, pink, and babble) over different Signal to Ratio (SNR) conditions. The results obtained were also compared with those of different deep-learning techniques. Our findings show that the proposed methodology can perform much better across all datasets than the current state-of-the-art.}, archive = {J_EAAI}, author = {Arvind Kumar and Rampravesh Kumar and Mahesh Chandra}, doi = {10.1016/j.engappai.2025.110130}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110130}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced audio classification leveraging pre-trained deep visual models}, volume = {144}, year = {2025}, }
• (2025). Data imputation in the frequency domain using echo state networks. EAAI, 144, 110129. (www)

  @article{ , abstract = {This study addresses the problem of reconstructing time series data with only partial spectral information, specifically the power spectral density (PSD), without corresponding phase details. Such situations are common in industrial settings where sensors measure the energy across various frequency bands but do not record the temporal signal. Existing phase-imputation methods often fail to produce reliable time-domain signals suitable for system identification algorithms, such as those used for calculating vibration modes and structural health monitoring. To address this issue, a solution is proposed using a physics-informed Echo State Network (ESN) designed to impute a time signal whose PSD matches the observed spectral data. The approach uses fuzzy sets to represent our limited knowledge about the inputs and extends the ESN definition to handle fuzzy-valued inputs, integrating physical insights directly into the frequency domain within the loss function. This enables the simultaneous learning of a set of weak constraints on the amplitude and autocorrelation of the unknown excitation to the physical system, alongside a recurrent neural network that models the functional dependence between these inputs and the partially observed outputs. This methodology is validated through two empirical analyses: one using a synthetic dataset designed to simulate real-world scenarios for benchmarking, and another through a practical case study focused on diagnosing the condition of industrial fans using vibration data. Empirical validation demonstrates that the proposed method successfully identified shifts in vibration modes, ranging from 8% to 17%, following structural changes.}, archive = {J_EAAI}, author = {Luciano Sánchez and Nahuel Costa and Inés Couso}, doi = {10.1016/j.engappai.2025.110129}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110129}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data imputation in the frequency domain using echo state networks}, volume = {144}, year = {2025}, }
• (2025). An end-to-end autonomous learning approach for waterway images analysis of inland river. EAAI, 144, 110128. (www)

  @article{ , abstract = {With the ongoing trend of implementing intelligent and remote traffic control on inland waterways, relying solely on Automatic Identification System as a data source is insufficient to meet diverse demands. Surveillance Video (SV) has attracted attention because of its cheap and real-time information compensation characteristics. However, existing SV process models still face many challenges, including the lack of exclusive data sets, the traditional detection/classification single-function model scarcity providing the whole surface waterway elements, and the model’s limited adaptability to diverse engineering scenarios. To tackle these challenges, this paper proposes a comprehensive solution for engineering applications, from data set establishment to multi-functional model design as well as model self-updating. Specifically, we propose a data augmentation strategy based on Auto-Augment to reduce the data annotation cost. Furthermore, we improve a multi-function model based on Transformer architecture by simultaneously achieving target detection and panoramic segmentation to obtain complete surface waterway elements. Moreover, inspired by knowledge distillation and generative adversarial ideas, we propose an automatic SV data labeling method and an asynchronous end-to-end autonomous learning strategy, ensuring that the model has sustainable learning ability when transferring to new application scenarios. In a regular test, our multi-functional model’s panoramic segmentation and target detection capability can achieve 0.7675 Intersection over Union and 0.9412 mean Average Precision (mAP), and the target detection task can achieve 0.6530 mAP in the deplorable vision case. This result reaches the engineering level that can be deployed in practice.}, archive = {J_EAAI}, author = {Zechen Li and Bingjie Chen and Tong Liu and Zhonglin Zuo and Shan Liang}, doi = {10.1016/j.engappai.2025.110128}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110128}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An end-to-end autonomous learning approach for waterway images analysis of inland river}, volume = {144}, year = {2025}, }
• (2025). Robot path planning based on three-dimensional artificial potential field. EAAI, 144, 110127. (www)

  @article{ , abstract = {Planning robot trajectories automatically using cameras or lasers is often challenging in numerous robotic manipulative tasks, as these technologies struggle to detect collisions between the tool tip and obstacles in narrow spaces. This paper proposes an innovative three-dimensional path planning method based on an artificial potential field in robotic manipulative tasks. This method senses the slight contact force from the obstacle through a force sensor installed at the end of the robotic arm that estimates the position of contact to avoid collisions. Different from ultrasonic and laser sensors, the force sensor can only obtain the force information of an obstacle point after contacting the obstacle. With such limited force information, the problem of falling into a local minimum of the artificial potential field tends to occur. To solve this problem and improve the three-dimensional space planning based on artificial potential field using force sensor, this paper proposes a three-dimensional artificial potential field. The direction of the rotating repulsive force field considers the forward directions of the tool and the path, thus the robot is able to successfully pass through narrow channels with overlapped repulsive force fields. Experiments have verified that the robotic arm that moves according to the path planning based on the three-dimensional artificial potential field can avoid the local minimum problem, as well as pass through a narrow channel smoothly. In addition, the robotic arm can obtain a smooth path after learning multiple artificial potential field paths.}, archive = {J_EAAI}, author = {Hsien-I Lin and Muhammad Ahsan Fatwaddin Shodiq and Ming Feng Hsieh}, doi = {10.1016/j.engappai.2025.110127}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110127}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robot path planning based on three-dimensional artificial potential field}, volume = {144}, year = {2025}, }
• (2025). A physics-informed neural network for nonlinear deflection prediction of ionic polymer-metal composite based on kolmogorov-arnold networks. EAAI, 144, 110126. (www)

  @article{ , abstract = {To characterize the highly nonlinear state response of the Ionic Polymer-Metal Composite (IPMC), coupled with inherent response uncertainties, we reformulate differential equations that encapsulate the highly nonlinear deflection of IPMC, accounting for uncertainties. Subsequently, we propose a data-driven approach utilizing a physics-informed neural network (PINN) to effectively solve this differential equation and predict the nonlinear deflection of IPMC actuators. To enhance the capacity for accurately capturing the intricate nonlinear behavior of IPMC, this study introduces an improved data-driven methodology that integrates Kolmogorov-Arnold Networks into the framework of PINNs (KAN-PINN). Through a series of experimental analyses and simulation studies, we demonstrate that KAN-PINN significantly outperforms traditional multilayer perceptron-based PINN (MLP-PINN), achieving a prediction accuracy improvement of 51.85% for Ag-IPMC and 27.54% for Pt-IPMC, along with a reduction in the convergence cycle. This research underscores the advantages of employing a data-driven, PINN-based approach in addressing the complexities associated with the nonlinear deflection of IPMC actuators. Moreover, it provides a substantial reference for future investigations and applications of IPMC.}, archive = {J_EAAI}, author = {Lin Zhang and Lei Chen and Fuxiang An and Zixuan Peng and Yuhang Yang and Tingting Peng and Yongshi Song and Yanzheng Zhao}, doi = {10.1016/j.engappai.2025.110126}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110126}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A physics-informed neural network for nonlinear deflection prediction of ionic polymer-metal composite based on kolmogorov-arnold networks}, volume = {144}, year = {2025}, }
• (2025). Heterogeneous hypergraph transformer network with cross-modal future interaction for multi-agent trajectory prediction. EAAI, 144, 110125. (www)

  @article{ , abstract = {Predicting the future trajectories of traffic agents is crucial for ensuring the safety and efficiency of self-driving vehicles and navigation robots in complex environments. Accurate trajectory prediction enhances decision-making and collision avoidance, which are fundamental to the deployment of autonomous systems in real-world scenarios such as urban traffic, delivery logistics, and pedestrian navigation. However, this task remains challenging due to agents' diverse interactions and personalized behavior based on their categories. Moreover, most trajectory prediction models neglect the interaction between generated multimodal trajectories, limiting their practical applicability. This paper proposes a heterogeneous hypergraph transformer network with cross-modal future interaction (HHT-CFI) for multi-agent trajectory prediction to address these challenges. Specifically, the trajectory coordinates are first fed into a temporal model using transformer encoders to capture temporal features. Next, a heterogeneous hypergraph transformer (HHT) is designed to model social influence between agents of different categories, capturing historical interaction features. A cross-modal future interaction (CFI) module is then developed to enable agents to perceive multiple possible future paths of their neighbors. Finally, multimodal trajectories are generated twice, where the second predictions refine the first by utilizing cross-modal future features. Extensive experiments on four datasets demonstrate the superior performance of our model over state-of-the-art methods, showcasing its potential for real-world applications in autonomous navigation and traffic management. The code of the proposed method will be publicly available at https://github.com/zhou00NJUST/HHT-CFI .}, archive = {J_EAAI}, author = {Xiangzheng Zhou and Xiaobo Chen and Jian Yang}, doi = {10.1016/j.engappai.2025.110125}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110125}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Heterogeneous hypergraph transformer network with cross-modal future interaction for multi-agent trajectory prediction}, volume = {144}, year = {2025}, }
• (2025). Enhanced differential evolution with hierarchical selection mutation and distance-based selection strategy. EAAI, 144, 110124. (www)

  @article{ , abstract = {Differential Evolution (DE) is an evolutionary algorithm that utilizes differential information among individuals within a population, which is simple yet effective. Since its inception, numerous variants have been proposed. A key factor in DE's performance is the mutation strategy, as it largely dictates the algorithm's ability to explore and exploit the search space. Traditional DE algorithms typically rely on a single mutation strategy, which can result in high randomness during individual selection and insufficient detection of stagnation, ultimately leading to weak performance. This paper introduces a new hierarchical selection mutation strategy and applies different mutation strategies at different stages of evolution to enhance the algorithm's performance. Additionally, a new scaling factor control strategy is introduced to better identify individual stagnation. To better assist individuals in escaping local optima, this paper designs a new distance-based probabilistic selection strategy. Combining these strategies, this paper further proposes an improved DE with Hierarchical Selection Mutation Strategy and Distance-Based Probabilistic Selection Strategy (HDDE). To thoroughly evaluate the performance of HDDE, extensive experiments are conducted on the Congress on Evolutionary Computation (CEC) 2017 and 2022 test suites, and the comparison results between HDDE and six other advanced algorithms demonstrate HDDE's competitive performance. Furthermore, HDDE has been validated on multiple real-world problems, and the results demonstrate its good performance.}, archive = {J_EAAI}, author = {Zhenyong Luo and Xuezhong Qian and Wei Song}, doi = {10.1016/j.engappai.2025.110124}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110124}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced differential evolution with hierarchical selection mutation and distance-based selection strategy}, volume = {144}, year = {2025}, }
• (2025). Unpacking predictive relationships in graphene oxide-reinforced cementitious nanocomposites: An explainable ensemble learning approach for augmented data. EAAI, 144, 110123. (www)

  @article{ , abstract = {This study utilizes ensemble learning models to predict the compressive strength of nanocomposites reinforced with three types of graphene oxide (GO): ordinary graphene oxide, reduced graphene oxide, and edge-oxidized graphene oxide, in both paste and mortar forms. A comprehensive dataset of 279 samples was collected, containing information regarding binder ingredients, GO characteristics, manufacturing details, and the compressive strength of the control sample (C0). An initial mean absolute error (MAE) of 1.88 MPa (MPa) and R-squared (R 2 ) of 0.98 were achieved when the control sample (C0) was included. To enhance the model's applicability for engineering applications, as it is a dependent variable, and to mitigate the risk of target leakage, C0 was excluded, and the model was retrained using augmented data generated by three generative adversarial network (GAN)-based methods. The eXtreme Gradient Boosting (XGB) model, optimized through a multi-objective Bayesian approach, yielded an MAE of 3.75 MPa and an R2 of 0.93. Four model interpretation methods were adopted, identifying water-to-cement ratio (W/C), GO content, GO functional group addition level, nano particles' thickness, and surfactant content as the most influential variables. SHapley Additive exPlanations (SHAP) results indicate that lower W/C ratios, moderate GO content, lower thickness, and sonication are beneficial for enhancing strength. The contribution of this study lies in the effective combination of machine learning, data processing, and synthetic data generation, addressing the complexities of predictive relationships in GO-reinforced nanocomposites. The developed models enable efficient mixture design optimization tasks by identifying optimal binder design and nanoparticle characteristics. To facilitate practical application, a graphical user interface will be made available to provide user-friendly predictions.}, archive = {J_EAAI}, author = {Hossein Adel and Majid Ilchi Ghazaan and Asghar Habibnejad Korayem}, doi = {10.1016/j.engappai.2025.110123}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110123}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unpacking predictive relationships in graphene oxide-reinforced cementitious nanocomposites: An explainable ensemble learning approach for augmented data}, volume = {144}, year = {2025}, }
• (2025). A prompting multi-task learning-based veracity dissemination consistency reasoning augmentation for few-shot fake news detection. EAAI, 144, 110122. (www)

  @article{ , abstract = {With the rise of social media, traditional news channels are being supplanted, leading to increased prevalence of fake news. Existing Fake News Detection (FEND) methods, while effective in fully-supervised settings, struggle with data scarcity and lack robustness in few-shot scenarios. Social pattern analysis shows that users tend to share information aligning with their existing beliefs, a behavior known as ”Confirmation Bias”. This bias influences how news spreads within social groups, where consistent sharing patterns among closely connected users often serve as credible indicators of news authenticity. We refer to this phenomenon as the ”user spreading news engagement bias”. Inspired by this, we propose a novel P rompting M ulti- T ask L earning framework guided by news veracity Dis semination Co nsistency reasoning augmentation for few-shot FEND, abbreviated as PMTL-DisCo , which leverages advanced Artificial Intelligence (AI) techniques, such as multi-task learning (MTL), prompt-based tuning, and masked language model (MLM), for feature extraction and optimization. Unlike prior works that primarily rely on pre-trained language models (PLMs) for feature extraction, PMTL-DisCo integrates the following innovations: (1) an auxiliary task, ”news distributed representation optimization,” that leverages dissemination consistency signals from neighboring news cases to enhance feature learning. (2) a high-quality, expanded label words-based verbalizer engineering, which improves prompt-tuning performance through adaptive multi-label mappings. (3) a multi-neighbor reasoning enhancement strategy, which refines predictions by incorporating the veracity features of socially connected news cases. These innovations address the limitations of existing FEND methods, particularly their inability to exploit cross-news dissemination patterns effectively. Extensive experiments and ablation studies on three widely used benchmarks demonstrate that PMTL-DisCo outperforms state-of-the-art few-shot FEND models, showcasing its superior generalization and robustness in low-resource settings.}, archive = {J_EAAI}, author = {Weiqiang Jin and Ningwei Wang and Tao Tao and Mengying Jiang and Yebei Xing and Biao Zhao and Hao Wu and Haibin Duan and Guang Yang}, doi = {10.1016/j.engappai.2025.110122}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110122}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A prompting multi-task learning-based veracity dissemination consistency reasoning augmentation for few-shot fake news detection}, volume = {144}, year = {2025}, }
• (2025). Frequency-domain multi-scale kolmogorov-arnold representation attention network for mixed-type wafer defect recognition. EAAI, 144, 110121. (www)

  @article{ , abstract = {Wafer defects are often complex, diverse, and frequently contaminated by noise, making their recognition challenging for advancing semiconductor manufacturing processes. Developing a high-precision wafer defect recognition method based on deep learning is particularly important, as it can effectively manage noise interference and improve accuracy. In this study, we propose frequency-domain multi-scale Kolmogorov-Arnold representation attention network (FMKA-Net) for wafer defect recognition. FMKA-Net employs two-dimensional discrete wavelet transform (DWT) to decompose frequency, and performs down-sampling and up-sampling on the images through multi-dimensional DWT layers and transposed convolutional blocks to achieve multi-scale feature extraction. Additionally, a fusion feature attention module is designed following Kolmogorov-Arnold network (KAN) to establish an information exchange channel between shallow and deep information layers, enabling detailed exploration of wafer map information. The fused features are finalized by the spatial pyramid pooling module, enhancing the model's representation capability and robustness. The discrete wavelet transform facilitates clear extraction of multi-scale information and enables learning across multiple scales, improving the model's adaptability to image scale variations. The attention module aids in selecting frequency location information captured by the DWT layer, further enhancing the model's performance. Our network achieved 99.03% accuracy on the Mixed38WM wafer dataset, outperforming existing methods under both noise-added and noise-free conditions. To demonstrate generalizability, our model was also tested on the WM-811K wafer dataset and the NEU-CLS steel surface defect dataset.}, archive = {J_EAAI}, author = {Qihang Huang and Fan Zhang and Yuqian Zhao and Ji'an Duan}, doi = {10.1016/j.engappai.2025.110121}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110121}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Frequency-domain multi-scale kolmogorov-arnold representation attention network for mixed-type wafer defect recognition}, volume = {144}, year = {2025}, }
• (2025). Intelligent path planning for autonomous ground vehicles in dynamic environments utilizing adaptive neuro-fuzzy control. EAAI, 144, 110119. (www)

  @article{ , abstract = {Autonomous Ground Vehicles (AGVs) are increasingly deployed across diverse industries, where enhancing their operational efficiency is critical, particularly in dynamic environments. This study proposes a hybrid control strategy that integrates an improved A∗ algorithm for path planning with a Proportional-Integral-Derivative (PID) controller adaptively tuned by an Adaptive Neuro-Fuzzy Inference System (ANFIS) for path correction. The enhanced A∗ algorithm, combined with the Dynamic Window Approach (DWA), significantly reduces computational overhead while improving pathfinding speed by incorporating the vehicle's kinematic and dynamic constraints. To address non-linearities in AGV movement, the ANFIS framework continuously fine-tunes PID parameters in real-time based on sensor feedback, improving the system's ability to correct path deviations in complex terrains. Experimental results demonstrate the efficacy of the proposed method. The enhanced A∗ algorithm achieves an average path search time of 2.52 s, significantly faster than the traditional A∗ algorithm's 5.56 s. It also reduces the average search grid size from 160 to 100, yielding a shorter path length of 27.44 m compared to 32.25 m, reflecting a more efficient path search process. Additionally, the ANFIS-PID control algorithm achieves a convergence time of 0.038 s, ensuring smooth path correction with robust stability under varying load conditions. Comparisons with state-of-the-art techniques, including Rapidly-Exploring Random Trees (RRT) and Probabilistic Roadmap Algorithm, highlight the enhanced A∗ algorithm's competitive performance, particularly in resource-constrained, real-time applications. The integration of ANFIS with PID control enhances AGV navigation by enabling adaptive, real-time path correction, improving performance in dynamic environments across agricultural, industrial, logistics, and autonomous transportation applications.}, archive = {J_EAAI}, author = {Ambuj and Rajendra Machavaram}, doi = {10.1016/j.engappai.2025.110119}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110119}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent path planning for autonomous ground vehicles in dynamic environments utilizing adaptive neuro-fuzzy control}, volume = {144}, year = {2025}, }
• (2025). A novel deep learning-based approach for video quality enhancement. EAAI, 144, 110118. (www)

  @article{ , abstract = {Video content has experienced a surge in popularity, asserting its dominance over internet traffic and Internet of Things (IoT) networks. Video compression has long been regarded as the primary means of efficiently managing the substantial multimedia traffic generated by video-capturing devices. Nevertheless, video compression algorithms entail significant computational demands in order to achieve substantial compression ratios. This complexity presents a formidable challenge when implementing efficient video coding standards in resource-constrained embedded systems, such as IoT edge node cameras. To tackle this challenge, this paper introduces an innovative deep-learning model specifically designed to mitigate compression artifacts stemming from lossy compression codecs. This enhancement significantly elevates the perceptible quality of low-bit-rate videos. By employing our proposed deep-learning model, the video encoder within the video-capturing node can reduce output quality, thereby generating low-bit-rate videos and effectively curtailing both computation and bandwidth requirements at the edge. On the decoder side, which is typically less encumbered by resource limitations, our suggested deep-learning model is applied after the video decoder to compensate for artifacts and approximate the quality of the original video. Experimental results affirm the efficacy of the proposed model in enhancing the perceptible quality of videos, especially those streamed at low bit rates.}, archive = {J_EAAI}, author = {Parham Zilouchian Moghaddam and Mehdi Modarressi and Mohammad Amin Sadeghi}, doi = {10.1016/j.engappai.2025.110118}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110118}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel deep learning-based approach for video quality enhancement}, volume = {144}, year = {2025}, }
• (2025). Resource constraint crop damage classification using depth channel shuffling. EAAI, 144, 110117. (www)

  @article{ , abstract = {Accurate crop damage classification is crucial for timely interventions, loss reduction, and resource optimization in agriculture. However, datasets and models for binary classification of damaged versus non-damaged crops remain scarce. To address this, we conducted an extensive study on crop damage classification using deep learning, focusing on the challenges posed by imbalanced datasets common in agriculture. We began by preprocessing the “Consultative Group for International Agricultural Research (CGIAR)” dataset to enhance data quality and balance class distributions. We created the new “Crop Damage Classification (CDC)” dataset tailored for binary classification of “Damaged” versus “Non-damaged” crops, serving as an effective training medium for deep learning models. Using the CDC dataset, we benchmarked the state-of-the-art models to evaluate their effectiveness in classifying crop damage. Leveraging the depth channel shuffling technique of ShuffleNetV2, we proposed a lightweight model “Light Crop Damage Classifier (LightCDC)” , reducing the parameters from 1.40 million to 1.13 million while achieving an accuracy of 89.44%. LightCDC outperformed existing classification and ensemble models in terms of model size, parameter count, inference time, and accuracy. Furthermore, we tested LightCDC under adverse conditions like blur, low light, and fog, validating its robustness for real-world scenarios. Thus, our contributions include a refined dataset and an efficient model tailored for crop damage classification, which is essential for timely interventions and improved crop management in resource-constrained precision agriculture. To ensure reproducibility, we released the code and dataset on GitHub .}, archive = {J_EAAI}, author = {Md Tanvir Islam and Safkat Shahrier Swapnil and Md. Masum Billal and Asif Karim and Niusha Shafiabady and Md. Mehedi Hassan}, doi = {10.1016/j.engappai.2025.110117}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110117}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Resource constraint crop damage classification using depth channel shuffling}, volume = {144}, year = {2025}, }
• (2025). Context-aware resemblance detection for data deduplication with neural network. EAAI, 144, 110116. (www)

  @article{ , abstract = {As the prevalence of cloud storage increases, many individuals and companies prefer outsourcing their data for backup and management. However, this has led to a significant increase in redundancy, decreasing storage utilization and wasting network bandwidth. While conventional resemblance detection methods remove redundancy among similar data by comparing the features extracted from each chunk’s content. However, we observed that small changes between similar data chunks may cause false dissimilarity detection by conventional resemblance detection techniques. This is because features derived solely from the chunk content are highly susceptible to various modification patterns. Fortunately, we have discovered that two chunks are likely to be similar if their surrounding chunks are also similar, a concept we refer to as “chunk-context”. Therefore, we propose a novel chunk-context aware resemblance detection method, called CARD, which includes a network-based chunk-context aware model and an N-sub-chunk shingles-based initial feature extraction strategy. By leveraging the Neural network, it can discover the complex patterns between the chunk-context and chunk content itself. A high-level understanding of the contextual information with chunk content can be synthesized into the representation of a chunk. The primary difference compared with others is that our design can significantly improves the accuracy or efficiency of resemblance detection by considering the chunk-context with chunk content itself. Furthermore, we implemented a CARD prototype and conducted extensive experiments using real workload, demonstrating that CARD can detect up to 75.03% more redundant data and accelerate the resemblance detection operations by 5.6 × to 86.7 × faster than state-of-the-art work.}, archive = {J_EAAI}, author = {Xuming Ye and Wenlong Tian and Yaping Wan and Ruixuan Li and Weijun Xiao and Zhiyong Xu}, doi = {10.1016/j.engappai.2025.110116}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110116}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Context-aware resemblance detection for data deduplication with neural network}, volume = {144}, year = {2025}, }
• (2025). An integrated outranking approach for selecting the best waste water treatment process with spherical fuzzy rough numbers. EAAI, 144, 110115. (www)

  @article{ , abstract = {Water is the essence of life, sustaining ecosystems, agriculture, industries, and human well-being. The water waste or contaminated by human activity, such as domestic, industrial or agricultural use make it unsafe for consumption, threatens human health, disrupts ecosystems and can lead to the depletion of clean water resources. It contains harmful pollutants and requires treatment to remove toxins before being safely released back into the environment. Proper management of the waste water is crucial for protecting water resources and public health. This paper tackles this issue by using spherical fuzzy rough numbers. Spherical fuzzy rough numbers deal with additional uncertain degree of neutral opinion along with degrees of association and disassociation by manipulating the subjective evaluations. The Criteria Importance Through Inter Criteria Correlation (CRITIC) approach is used for criteria weights objectively by analyzing their variability and inter-correlation. It assigns higher weights to criteria with greater variation and less correlation, ensuring a balance of significance and independence. By fusing the CRITIC method with the spherical fuzzy rough numbers-based REGIME technique, this study seeks to create a hybrid model that addresses ambiguity in a more flexible way. The main principle of the REGIME technique is to rank alternatives by analyzing their pairwise performance across multiple criteria, using a preference function to capture the strength of dominance. It integrates both qualitative and quantitative data, avoiding the need for predefined weights while maintaining consistency in decision-making. The spherical fuzzy rough CRITIC-REGIME technique is illustrated with a flowchart for ease of understanding. In order to assess the efficacy of the suggested technique, a comprehensive case study is solved using the spherical fuzzy rough CRITIC-REGIME technique to determine the best waste water treatment process. Following that, the suggested method is compared with the existing methods. Sensitivity analysis is conducted at the end to verify the accuracy of computations of proposed method.}, archive = {J_EAAI}, author = {Muhammad Akram and Safeena Azam and Ghulam Mustafa and Dragan Pamucar}, doi = {10.1016/j.engappai.2025.110115}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110115}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated outranking approach for selecting the best waste water treatment process with spherical fuzzy rough numbers}, volume = {144}, year = {2025}, }
• (2025). Decision support model under the fractional orthotriple fuzzy information. EAAI, 144, 110114. (www)

  @article{ , abstract = {Fractional orthotriple fuzzy set is a revolutionary and a generalized framework designed to contain more ambiguity than the picture fuzzy set and Spherical fuzzy set. Thus, its implementations are too comprehensive. The concept proposed in this paper, eliminates the limitations of the existing frameworks and allows decision-makers to choose grades without any restrictions. In this paper the generalized concept of orthotriple fuzzy set is presented and establish their basic operational laws. We provide a formulation for the composition of fractional orthotriple fuzzy numbers, build algebraic aggregation operators utilizing fundamental operational rules, and thoroughly examine the characteristics of aggregation operators (AOs). To demonstrate the practicality and applicability of the proposed model, a case study is conducted to evaluate the performance of project team members across multiple criteria. Finally, the reliability and effectiveness of the decision-making framework are validated through sensitivity analysis and a comparative study with existing aggregation methods, underscoring its superiority and practical relevance.}, archive = {J_EAAI}, author = {Muhammad Qiyas and Neelam Khan and Darjan Karabasevic and Muhammad Luqman Alvi}, doi = {10.1016/j.engappai.2025.110114}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110114}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Decision support model under the fractional orthotriple fuzzy information}, volume = {144}, year = {2025}, }
• (2025). More consistent learnt relationships in regression neural networks using fit to median error measure. EAAI, 144, 110113. (www)

  @article{ , abstract = {Machine learning is increasingly used to optimise complex systems. The quality of this optimisation is dependent on the models having a reasonable physical representation of the system. However, standard error metrics are pointwise, providing a reasonable physical representation only under certain constraints. As an example, data models of ship powering have low errors values, < 2%, but fail to consistently approximate the input–output relationships and cannot be used to optimise performance. This paper illustrates that the Fit to Median error measure can be used to assess how well the ground truth is modelled. It provides more consistent learnt relationships and improves the extrapolation accuracy of neural networks. This is illustrated on real-world data used for ship power prediction.}, archive = {J_EAAI}, author = {Amy I. Parkes and Adam J. Sobey and Dominic A. Hudson}, doi = {10.1016/j.engappai.2025.110113}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110113}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {More consistent learnt relationships in regression neural networks using fit to median error measure}, volume = {144}, year = {2025}, }
• (2025). A lightweight method for detecting turned white belly fish in ponds using unmanned aerial vehicle imagery. EAAI, 144, 110111. (www)

  @article{ , abstract = {Rapid and accurate detecting of turned white belly fish in pond environments is essential for monitoring fish growth conditions, conducting ecological assessments, and facilitating insurance compensation. However, fixed monitoring equipment struggles to capture complete images from extensive aquatic environments, and the high memory requirements and poor real-time performance of the detection model hinder deployment on edge devices. This study proposes a method for detecting turned white belly fish in ponds using Unmanned Aerial Vehicle (UAV) imagery and lightweight You Only Look Once Version 8 (YOLOv8). Initially, the Adaptive As-Natural-As-Possible (AANAP) algorithm is utilized to stitch the UAV images. Image processing techniques are applied to identify regions of interest, and these regions are uniformly cropped to obtain local pond images. Subsequently, a lightweight model is developed to detect the positions of belly-up fish in the local images. The number of belly-up fish in the original pond image is obtained through a merging process. The method is validated on a real dataset, achieving a weighted harmonic mean of precision and recall (F1-score) of 98.58%, a mean Average Precision (mAP) of 99.40%, a model size of 8.62 Mega Byte (MB), and a Floating-point Operations (FLOPs) of 2.28 × 10 10 . The Mean Relative Error (MRE) of counting is 2.23%. Compared with other methods, this approach shows higher detection accuracy and speed, with lower computational cost and a smaller model size. The results demonstrate that the proposed method can accurately and quickly estimate the number of belly-up fish in ponds, providing theoretical support for practical aquaculture.}, archive = {J_EAAI}, author = {Mingrui Kong and Yiran Liu and Beibei Li and Qingling Duan}, doi = {10.1016/j.engappai.2025.110111}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110111}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight method for detecting turned white belly fish in ponds using unmanned aerial vehicle imagery}, volume = {144}, year = {2025}, }
• (2025). Detection of novel prohibited item categories for real-world security inspection. EAAI, 144, 110110. (www)

  @article{ , abstract = {The application of prohibited item detection in intelligent security inspection systems is essential for protecting public safety and categories of prohibited item are continuously increasing in accordance with the latest laws. Previous works all focus on closed-set scenarios, which can only recognize known categories used for training and often require time-consuming as well as labor-intensive annotations when learning novel categories, resulting in limited real-world applications. Although the success of vision-language models such us Contrastive Language-Image Pretraining (CLIP) provides a new perspective for X-ray prohibited item detection in open scenarios, directly applying CLIP to X-ray domain leads to a sharp performance drop due to domain shift between X-ray data and general data used for pre-training CLIP. To address aforementioned challenges, in this paper, we introduce distillation-based open-vocabulary object detection (OVOD) task into X-ray security inspection domain by extending CLIP to learn visual representations in specific X-ray domain, aiming to detect novel prohibited item categories beyond base categories on which the detector is trained. Specifically, we propose X-ray feature adapter and apply it to CLIP within OVOD framework to develop Open-Vocabulary X-ray prohibited item Detection (OVXD) model. Extensive experiments conducted on three mainstream X-ray security inspection datasets demonstrate that OVXD performs favorably against baseline methods in X-ray scenarios. Furthermore, OVXD can be directly transferred to different X-ray datasets without fine-tuning, verifying its generalization ability in real-world applications.}, archive = {J_EAAI}, author = {Shuyang Lin and Tong Jia and Hao Wang and Bowen Ma and Mingyuan Li and Dongyue Chen}, doi = {10.1016/j.engappai.2025.110110}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110110}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detection of novel prohibited item categories for real-world security inspection}, volume = {144}, year = {2025}, }
• (2025). BV-NORM: A neural operator learning framework for parametric boundary value problems on complex geometric domains in engineering. EAAI, 144, 110109. (www)

  @article{ , abstract = {Boundary Value Problems (BVPs) are extensively employed in engineering for process modelling and optimisation. These problems frequently involve complex geometries and require the massive solution of BVPs under different boundary conditions (BCs). Neural operators (NOs), capable of learning mappings between infinite-dimensional functions, present a potential solution for solving parametric BVPs. However, existing NOs are typically designed for scenarios where the input and output functions share the same domain, thus not applicable to BVPs in which the BC and solution functions are defined over different complex domains. Therefore, this study presents a novel deep learning framework called Boundary Value Neural Operator on Riemannian Manifolds (BV-NORM) for solving parametric BVPs involving complex geometric domains. BV-NORM introduces two sub-networks, Geometry-NET (Geo-NET) and Boundary condition-NET (BC-NET), to encode geometric information and boundary conditions. Consequently, the geometric information of the output domain can then be incorporated into the learning process. Furthermore, the Laplace kernel integration module of NORM is employed to construct the sub-networks, thereby enhancing the capacity to analyse complex geometric domains. The performance of the proposed method is evaluated in four benchmark cases, including toy Partial differential equation (PDE) cases and engineering applications, through comparisons with existing baseline neural operators. The experimental results validate that BV-NORM effectively addresses BVPs across various engineering scenarios.}, archive = {J_EAAI}, author = {Zhiliang Deng and Qinglu Meng and Yingguang Li and Xu Liu and Gengxiang Chen and Lu Chen and Changqing Liu and Xiaozhong Hao}, doi = {10.1016/j.engappai.2025.110109}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110109}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {BV-NORM: A neural operator learning framework for parametric boundary value problems on complex geometric domains in engineering}, volume = {144}, year = {2025}, }
• (2025). Graph neural networks informed locally by thermodynamics. EAAI, 144, 110108. (www)

  @article{ , abstract = {Thermodynamics-informed neural networks employ inductive biases for the enforcement of the first and second principles of thermodynamics. To construct these biases, a metriplectic evolution of the physical system under study is assumed. This provides excellent results, when compared to uninformed, black box networks. While the degree of accuracy can be increased in one or two orders of magnitude, in the case of graph networks, this requires assembling global Poisson and dissipation matrices, which breaks the local structure of such networks. In order to avoid this drawback, a local version of the metriplectic biases has been developed in this work, which avoids the aforementioned matrix assembly, thus preserving the node-by-node structure of the graph networks. We apply this framework for examples in the fields of solid and fluid mechanics. Our approach demonstrates significant computational efficiency and strong generalization capabilities, accurately making inferences on examples significantly different from those encountered during training.}, archive = {J_EAAI}, author = {Alicia Tierz and Icíar Alfaro and David González and Francisco Chinesta and Elías Cueto}, doi = {10.1016/j.engappai.2025.110108}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110108}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph neural networks informed locally by thermodynamics}, volume = {144}, year = {2025}, }
• (2025). Modeling biodiesel properties by preference learning: Case study of cetane number. EAAI, 144, 110107. (www)

  @article{ , abstract = {The properties and performance of biodiesels depend on the composition of the fatty acid methyl esters (FAMEs) present. Obtaining new biodiesels is neither a cheap nor a simple task. Therefore, it is important to model their properties in advance in order to know what composition would be needed to fulfill a given requirement. This paper presents a multitask approach that models the behavior of biodiesels using preference learning and also obtains representations of those biodiesels in a feature space by means of an autoencoder. The performance of the system in different configurations is also analyzed, showing its robustness. A case study was conducted to analyze the cetane number, which is an important property in biodiesel. Correlations between the cetane number and the distribution of FAMEs have been identified. These correlations are more clearly observed when the FAMEs are grouped according to common characteristics, such as the percentage of saturated fatty acids (SFA), monounsaturated fatty acids (MUFA), and polyunsaturated fatty acids (PUFA) in the biodiesel, or the average number of double bonds (DB). Based on these groupings, it can be observed that the cetane number decreases with increasing unsaturation. Additionally, it was determined that the cetane number value has a minimal or no correlation with the average molecular weight (Mw) of the biodiesel.}, archive = {J_EAAI}, author = {Guillermo Díez-Valbuena and Alejandro García-Tuero and Eduardo Rodríguez and Antolin Hernández Battez and Jorge Díez}, doi = {10.1016/j.engappai.2025.110107}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110107}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Modeling biodiesel properties by preference learning: Case study of cetane number}, volume = {144}, year = {2025}, }
• (2025). A hybrid time series and physics-informed machine learning framework to predict soil water content. EAAI, 144, 110105. (www)

  @article{ , abstract = {This study investigates a predictive hybrid framework for modeling soil water content that leverages the strengths of both data-driven machine learning and simple hydrological models through two distinct approaches: hybrid uncertainty quantification (UQ) and physics-informed neural networks (PINN). A modified antecedent precipitation index model served as the core domain knowledge in both methods. In the hybrid UQ approach, deep learning models—such as support vector regression, random forest, feedforward neural networks, and long short-term memory networks (LSTM)—were trained to learn the residuals between the predictions of the hydrological model and observed values collected from the Kansas Mesonet environmental monitoring network. The residuals were then integrated with the hydrological model to form the predictive hybrid model. In the PINN approach, the LSTM-based sequence-to-sequence and feedforward neural network models were trained using a cost function incorporating the hydrological model and observed data. Our simulations showed that the two investigated frameworks improved predictions compared to purely using data-driven or the hydrological model, demonstrating the potential of hybrid machine learning frameworks for accurately modeling soil moisture dynamics for guiding agricultural and hydrological applications.}, archive = {J_EAAI}, author = {Amirsalar Bagheri and Andres Patrignani and Behzad Ghanbarian and Davood B. Pourkargar}, doi = {10.1016/j.engappai.2025.110105}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110105}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid time series and physics-informed machine learning framework to predict soil water content}, volume = {144}, year = {2025}, }
• (2025). Corrigendum to “A novel multi-modal neural network approach for dynamic and generic sports video summarization” [Eng. appl. artif. intell. 126 part b (2023) 106964]. EAAI, 144, 110104. (www)

  @article{ , archive = {J_EAAI}, author = {Pulkit Narwal and Neelam Duhan and Komal Kumar Bhatia}, doi = {10.1016/j.engappai.2025.110104}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110104}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Corrigendum to “A novel multi-modal neural network approach for dynamic and generic sports video summarization” [Eng. appl. artif. intell. 126 part b (2023) 106964]}, volume = {144}, year = {2025}, }
• (2025). A self-prompt based dual-domain network for nighttime flare removal. EAAI, 144, 110103. (www)

  @article{ , abstract = {Existing nighttime flare removal work regards flare as a single degradation factor in spatial domain. However, flare in complex scenes consists of multiple flare types, and it is difficult to distinguish them from the background, leading to distorted results and incomplete perception. In this paper, we propose a self-prompt based dual-domain network named SPDDNet for nighttime flare removal, which encodes the data distributions of different flare types to generate prompt features and facilitates the interaction of the prompt features with the decoder to guide network for flare removal. In addition, we introduce Fast Fourier Transform and parallel attention in traditional convolutional neural network that is designed to extract global frequency features and location-dependent local information to accurately perceive the flare region. Finally, to adequately integrate spatial details, contextual information, prompt and image features, we propose a feature fusion module that generates a set of learned dynamic weights to adaptively guide the information fusion across channels. Extensive experiments on real-world and synthetic datasets strongly demonstrate the effectiveness of our proposed SPDDNet and its superior performance compared to state-of-the-art methods. Moreover, as an essential pre-processing step, the potential advantages of our method for other computer vision applications, including object detection and semantic segmentation, are demonstrated.}, archive = {J_EAAI}, author = {Kejing Qi and Bo Wang and Yun Liu}, doi = {10.1016/j.engappai.2025.110103}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110103}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A self-prompt based dual-domain network for nighttime flare removal}, volume = {144}, year = {2025}, }
• (2025). Do they like your game? early-stage churn prediction using a two-phase neural network system. EAAI, 144, 110102. (www)

  @article{ , abstract = {In this study, we propose a system named uitChurnPred that enables early-stage churn prediction by using deep learning neural networks and machine learning models, focusing on minimal feature sets that are available in the immediate post-tutorial phase of user onboarding. Utilizing a dataset comprising over 713.884 user records, we introduce a two-phase system for early-stage churn prediction in mobile gaming, develop and evaluate fours Artificial Neural Network models that vary in the number of parameters and fours Machine Learning models: Logistic Regression, Random Forest, Gradient Boosting, eXtreme Gradient Boosting. The models are trained on minimal feature sets post-tutorial such as device information, user demographics, user acquisition information, user interaction during tutorial steps, and behavior data in the first session, with some preprocessing steps to handle high cardinality and missing values. Despite the challenges posed by feature overlap, our first phase model achieved an accuracy of up to 68.77% and a remarkable recall rate for Churned users of 80%. The second phase adds first-session user behavior data, enhancing accuracy to 73% and recall to 81%. These metrics underscore the potential efficacy of our approach for early stage churn detection with minimal data requirements. Real-world split testing substantiated the efficacy of the model, indicating a 20% increase in Day 1 retention (D1R).}, archive = {J_EAAI}, author = {Ha Dang Hoang and Nguyen Tan Cam}, doi = {10.1016/j.engappai.2025.110102}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110102}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Do they like your game? early-stage churn prediction using a two-phase neural network system}, volume = {144}, year = {2025}, }
• (2025). Neural networks for estimating surface solar irradiation from satellite images. EAAI, 144, 110101. (www)

  @article{ , abstract = {In this work, we present a MultiLayer Perceptron (MLP) model to estimate ground solar radiation, in terms of Global Horizontal Irradiance (GHI), over a given site and for a specific time. The MLP model generates GHI estimates from the Meteosat 12-channel satellite images centered over the target location, and GHI values in clear-sky conditions over the same location. The dataset includes two years of data, covering 2016 and 2017, with a temporal granularity of 15 min, relative to a set of 16 test sites distributed across Europe, Africa and South America. We populate the training- and test-sets with all available data for the 15 and remaining station, respectively. We test all possible combinations of stations to define training and test sets, demonstrating the generalizability of the presented MLP model over potentially any location included in the Meteosat full-disk image. The estimated GHI values are compared to ground-measured GHI data achieving an overall Root Mean Square Error (RMSE) and Coefficient of determination (R 2 ) of 77.682 W/m 2 and 0.929, respectively, across all locations. Finally, the GHI estimates are set against those generated by the Heliosat-4 method, our benchmark, yielding an overall RMSE improvement of 3 W/m 2 . The experiments show that neural networks produce competitive results with fewer and accessible inputs compared to complex physical models for estimating solar radiation. Furthermore, historical and near real-time GHI estimation enabled by the proposed methodology would help photovoltaic (PV) planners determine the irradiance profile of a site where the deployment of a weather station is precluded.}, archive = {J_EAAI}, author = {Raimondo Gallo and Marco Castangia and Alberto Macii and Edoardo Patti and Alessandro Aliberti}, doi = {10.1016/j.engappai.2025.110101}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110101}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural networks for estimating surface solar irradiation from satellite images}, volume = {144}, year = {2025}, }
• (2025). Moving infrared dim and small target detection by mixed spatio-temporal encoding. EAAI, 144, 110100. (www)

  @article{ , abstract = {Infrared dim and small target detection has become a challenging and hot topic in recent years. Due to the extremely small pixels and low intensity of targets, the existing methods based on single frame seldom consider the temporal relationship between frames, thus often resulting in poor detection performance. Moreover, unlike general objects, the detection task suffers from an overwhelming amount of background information. Usually, directly employing inter-frame attention mechanisms or Convolutional Long Short Term Memory would introduce excessive interfering information and computational load. To advance the detection of infrared dim and small targets and enhance the application of artificial intelligence techniques in this field, this paper explores an artificial intelligence-based method for spatio-temporal feature fusion, which is prevalent in the video domain. Simply but effectively, we propose an infrared target detection network based on spatio-temporal feature fusion principles. It consists of two parallel feature extraction branches. One is designed to extract temporal features across entire video clip, and the other is designed to isolate spatial features from a selected key frame. Following this, a self-refining module comes into play, adaptively adjusting the extracted temporal features and fusing spatio-temporal features at different scales, enhancing deep feature representation. The extensive experiments on two public datasets validate our method simple yet effective. It could reduce floating point operations per second (FLOPs) obviously, promote inference speed, and achieve superior detection accuracy in comparison to current state-of-the-art methods. Our code is publicly accessible at https://github.com/UESTC-nnLab/STME .}, archive = {J_EAAI}, author = {Shuang Peng and Luping Ji and Shengjia Chen and Weiwei Duan and Sicheng Zhu}, doi = {10.1016/j.engappai.2025.110100}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110100}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Moving infrared dim and small target detection by mixed spatio-temporal encoding}, volume = {144}, year = {2025}, }
• (2025). A hybrid prognosis method based on health indicator and wiener process: The case of multi-sensor monitored aero-engine. EAAI, 144, 110099. (www)

  @article{ , abstract = {The stochastic nature of the multi-sensor monitored aero-engine degradation process limits the accuracy of its remaining useful life (RUL) prediction. To address this problem, a hybrid prognosis method for multi-sensor monitored aero-engine based on health indicators and the Wiener process is proposed. The method constructs composite health indicators (CHI) for aero-engine. The one-dimensional sensor-monitored signals are converted to two-dimensional features through the Gramian Angle Field (GAF), thereby preserving potential feature relationships in the time series data. A hybrid convolutional neural network (CNN) and Transformer are constructed, and the feature maps obtained from the GAF are used as inputs to fuse multi-sensor data to obtain the CHI. The Wiener process is used to model the CHI degradation. The initial parameters for the degradation model are obtained using the maximum likelihood estimation. The Bayesian and expectation maximization methods adaptively update the parameters online based on real-time data. On this basis, the probability density function of the RUL under the first-hitting-time is derived. To enhance adaptability, CHI is combined with the Wiener process for joint training. The effectiveness of the hybrid model is experimentally verified on the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset, and the generalization ability is further verified on the New Commercial Modular Aero-Propulsion System Simulation (N-CMAPSS) dataset. The results show that the proposed method improves the root mean square error by 2.07 and the prediction accuracy by 18% on average, which can meet the real-time and storage requirements in engineering applications and provide technical support for actual maintenance decision-making.}, archive = {J_EAAI}, author = {Xueqi Yang and Xinqin Gao and Haiyang Zheng and Mingshun Yang and Yong Liu}, doi = {10.1016/j.engappai.2025.110099}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110099}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid prognosis method based on health indicator and wiener process: The case of multi-sensor monitored aero-engine}, volume = {144}, year = {2025}, }
• (2025). A three-stage adaptive memetic algorithm for multi-objective optimization of flexible assembly job-shop scheduling problem. EAAI, 144, 110098. (www)

  @article{ , abstract = {The flexible assembly job-shop scheduling problem (FAJSP) widely arises in the manufacturing industry. Various approaches have been designed in recent years to address this problem. However, existing methods have rarely considered assembly process constraints and task assembly wait time. For this reason, this paper proposes a three-stage adaptive memetic algorithm (TA-MA) to solve the FAJSP with process route constraints. Specifically, the proposed algorithm combines memetic algorithms and reinforcement learning. The optimization objectives are completion time, equipment load, and assembly operation waiting time. Moreover, a two-layer integer coding method is proposed to encode the problem, and a reinforcement learning method is introduced to assist the solution search of the memetic algorithm. Further, a three-stage search framework is designed to reasonably equilibrium TA-MA’s exploration and mining capabilities as iterations advance. Finally, the effectiveness of the proposed algorithm is assessed through a series of experiments. The outcomes demonstrate that the proposed algorithm is effective and outperforms existing algorithms.}, archive = {J_EAAI}, author = {Chenlu Zhang and Jiamei Feng and Mingchuan Zhang and Lei Yang and Lei Zhang and Lin Wang and Junlong Zhu and Qingtao Wu}, doi = {10.1016/j.engappai.2025.110098}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110098}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A three-stage adaptive memetic algorithm for multi-objective optimization of flexible assembly job-shop scheduling problem}, volume = {144}, year = {2025}, }
• (2025). Physics-informed deep operator networks with stiffness-based loss functions for structural response prediction. EAAI, 144, 110097. (www)

  @article{ , abstract = {Finite element (FE) modeling is a powerful tool for structural analysis, but it often involves extensive pre-processing, significant analysis efforts, and time-consuming computations, especially for complex structures. To overcome these challenges, this study presents an innovative approach for real-time prediction of structural static responses using Deep Operator Networks (DeepONet). This method leverages a novel, physics-informed network guided by structural balance laws, enabling accurate predictions for various load scenarios. The trained DeepONet can generate solutions for the entire domain across every mesh point in a fraction of a second, eliminating the need for repetitive FE modeling for each new case. The proposed method is applied to two structures: a simple beam and a real-life model of the KW-51 bridge. To predict multiple variables, two strategies are employed: a split branch/trunk approach and multiple DeepONets combined into one. A parametric study optimizes the network’s design, considering factors like neurons, layers, batch size, and aspect ratio, ensuring high accuracy while avoiding underfitting and overfitting. Beyond data-driven (DD) training, the study introduces novel physics-informed training methods that utilize structural stiffness matrices to enforce equilibrium and energy conservation principles. These methods lead to two new loss functions: energy conservation (EC) and static equilibrium using the Schur complement (SE-S). By combining these loss functions, the model achieves less than 5% error with significantly reduced training time. This study shows that DeepONet, enhanced with hybrid loss functions, can efficiently and accurately predict structural responses at each mesh point, with minimal training time.}, archive = {J_EAAI}, author = {Bilal Ahmed and Yuqing Qiu and Diab W. Abueidda and Waleed El-Sekelly and Borja García de Soto and Tarek Abdoun and Mostafa E. Mobasher}, doi = {10.1016/j.engappai.2025.110097}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110097}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed deep operator networks with stiffness-based loss functions for structural response prediction}, volume = {144}, year = {2025}, }
• (2025). Reading comprehension powered semantic fusion network for identification of N-ary drug combinations. EAAI, 144, 110096. (www)

  @article{ , abstract = {The concurrent use of multiple medications to treat one or more diseases is prevalent. Identifying N-ary drug combinations from biomedical texts aids in uncovering significant pharmacological effects triggered by drug-drug interactions. Previous methods for this emerging task have primarily concentrated on representing drug entities using pre-trained language models, overlooking the comprehensive extraction of contextual and task-specific semantic information. To address these limitations, we develop a semantic fusion method grounded in machine reading comprehension (MRC) framework. Our model, termed Reading Comprehension powered semantic Fusion network for Identification of N-ary Drug combinations (RCFIND), first constructs relevant contexts and queries for each individual drug combination. Then, diverse information sources, including task-specific semantics, drug entity representations and contextual details, are fused by using a simplified Capsule network as well as incorporating contrastive learning. We assess RCFIND, achieving F1 scores ranging from 72.0% to 83.3% across four types of evaluations. Experimental results demonstrate significant performance enhancements over existing baselines, with at least a 5% F1 score improvement. Ablation studies and further analysis confirm the efficacy of the MRC framework and contrastive learning in accurately identifying N-ary drug combinations.}, archive = {J_EAAI}, author = {Hua Zhang and Peiqian Zhan and Cheng Yang and Yongjian Yan and Zijing Cai and Guogen Shan and Bo Jiang and Bi Chen and Qing Gu and Qingqing Zhou}, doi = {10.1016/j.engappai.2025.110096}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110096}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reading comprehension powered semantic fusion network for identification of N-ary drug combinations}, volume = {144}, year = {2025}, }
• (2025). A context feature enhancement and adaptive weighted fusion network for river floating debris detection. EAAI, 144, 110095. (www)

  @article{ , abstract = {While certain universal detectors have demonstrated commendable performance on public datasets, their effectiveness diminishes when applied to the detection of floating debris on rivers. This deficiency stems from two aspects: most existing public datasets are one-class, failing to represent the multi-class, multi-scene scenarios crucial for water management, and the complex, variable features inherent to floating objects pose significant challenges for accurate feature extraction. This paper introduces a new detector, named context feature e nhancement and a daptive weighted f usion network (EAFDet), designed for river floating object detection with higher accuracy and less testing time. A context feature enhancement module (EM) strengthens the context feature via the fusion of global context features and multi-scale local context features, and an adaptive weighted fusion module (AFM) solves the category ambiguity problem caused by occlusion. The AFM learns the spatial weight of feature map and fuses the semantic information of features at different levels, thereby improving the representation of effective features. Our network achieves an impressive detection accuracy of 86.7%, with a recall of 93.1% at a speed of real-time, surpassing other currently common object detection methods. Lastly, our detector can be embedded in to cameras and integrated in the surveillance system to achieve smart inspections.}, archive = {J_EAAI}, author = {Lili Zhang and Zizun Wei and Yehong Shao and Zhe Chen and Zhonghua Luo and Yingyan Dou}, doi = {10.1016/j.engappai.2025.110095}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110095}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A context feature enhancement and adaptive weighted fusion network for river floating debris detection}, volume = {144}, year = {2025}, }
• (2025). Effective fuzzing testcase generation based on variational auto-encoder generative adversarial network. EAAI, 144, 110094. (www)

  @article{ , abstract = {Fuzzing is an effective way to detect vulnerability, yet the quality of initial testcases greatly affects it. Recently generative adversarial network(GAN) has been applied to the automatic generation of testcases, which are expected between testcases that fully conform to the syntax format and the random testcases that are probably rejected. These testcases can cover new edges of programs to find bugs. However, there are still some problems in the current approaches based on GAN, such as unstable training process, single generation method, and generating testcases that are difficult to find new edges. In this paper, a novel fuzzing testcase generation technique based on variational auto-encoder generative adversarial network(VAE-GAN) is proposed, which introduces the representation learning process of the variational auto-encoder(VAE) for traditional GANs. We innovate two generation methods that the generator can learn and generate testcases according to the feature information of the training set, which can improve the stability of training and generates more diversified testcases. We use generated testcases as the initial testcases for fuzzing tools such as AFL++ to execute common Linux programs. The experiments show that the testcases generated by VAE-GAN outperform other GANs in 7/9 of target programs in terms of the unique tuples triggered; compared with the training set generated by AFL++ mutation, testcases generated by VAE-GAN can increase 57.11% edge discovery, 57.14% unique crash discovery and 85.26% unique hang discovery for target programs at most; the initial testcases generated by VAE-GAN also optimizes the performance of Fairfuzz and Angora, reflecting the scalability of the VAE-GAN model.}, archive = {J_EAAI}, author = {Zhongyuan Qin and Jiarong Fan and Xujian Liu and Zeru Li and Xin Sun}, doi = {10.1016/j.engappai.2025.110094}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110094}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Effective fuzzing testcase generation based on variational auto-encoder generative adversarial network}, volume = {144}, year = {2025}, }
• (2025). Forward and inverse adversarial model applying to well-logging. EAAI, 144, 110090. (www)

  @article{ , abstract = {Geophysical logging is critical for reservoir characterization but is limited by the small sample problem in machine learning. To address this, we propose the Forward and Inverse Adversarial Model (FIAM), which borrows the training method from generative adversarial networks and applies to geophysical logging. The FIAM includes a forward model for generating synthetic logging data and an inverse model for predicting reservoir parameters. These models are trained using an adversarial process, enabling continuous improvement without large labeled datasets. In the case study with exploration logging data, the FIAM enhances reservoir parameter prediction effect and maps the relationship between logging curves and reservoir parameters by pure data-driven training. The inverse model is pre-trained by measured data and then guides the forward model to generate logging data based on virtual reservoir parameters. Both models are trained alternately until no further improvement is achieved. Experimental results on oilfield datasets show that the FIAM improves reservoir parameter predictions by more than 30% when facing the small sample problem. The FIAM demonstrates significant potential for improving reservoir parameters prediction with small sample dataset, advancing both artificial intelligence methodologies and practical engineering applications.}, archive = {J_EAAI}, author = {Jun Zhou and Juan Zhang and Rongbo Shao and Lizhi Xiao and Guangzhi Liao}, doi = {10.1016/j.engappai.2025.110090}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110090}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forward and inverse adversarial model applying to well-logging}, volume = {144}, year = {2025}, }
• (2025). Jekyll institute or mrs hyde? gender identification with machine learning. EAAI, 144, 110087. (www)

  @article{ , abstract = {Social media platforms offer an invaluable wealth of data to understand what is taking place in our society. However, social media data hides demographic biases related to characteristics such as gender or age. Therefore, considering social media data as representative of the population can lead to fallacious interpretations. For instance, in France in 2021, women represent 51.6% of the population 1 , whereas on Twitter they represent only 33.5% of French users 2 . With such a significant difference between social network user demographics and the actual population, detecting the gender or age before delving into a deeper analysis of social phenomena becomes a priority. In this paper, we tackle the gender detection problem on Twitter. We introduce miniAM2, which is an assemblage model of an enriched distillation with weak-supervised learning. Our contributions are threefold: (i) a novel multilingual model that outperforms existing models in both accuracy and speed, allowing for real-time gender detection and organization status on Twitter based on their name , screen_name , and description , making it lighter and faster than state-of-the-art; (ii) an innovative assemblage multi-language strategy that enriches a distillation process with weak-supervised learning using minimal annotated data, and (iii) a unique method to adapt the model to similar languages without requiring annotated data in the target language, which provides significant advancements in handling resource-poor languages in gender detection tasks. We provide our model on demand so social scientists can use it for their analysis.}, archive = {J_EAAI}, author = {Arnault Gombert and Borja Sánchez-López and Jesus Cerquides}, doi = {10.1016/j.engappai.2025.110087}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110087}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Jekyll institute or mrs hyde? gender identification with machine learning}, volume = {144}, year = {2025}, }
• (2025). A multi-attribute quantum group consensus model considering psychological preference. EAAI, 144, 110086. (www)

  @article{ , abstract = {In response to the issues of information uncertainty and opinion interference in meteorological disaster emergency rescue, this paper proposes a multi-attribute quantum group decision-making model. Firstly, the paper employs the linguistic Z-numbers to represent the evaluation information of decision-makers in complex uncertain decision-making environments. Secondly, a group consensus model based on minimal adjustments is constructed, considering the psychological state and adjustment willingness of decision-makers when facing group preference pressure. On this basis, the weights of decision-makers are determined by incorporating their marginal contribution levels. Meanwhile, an optimization model is developed based on the marginal contributions from Shapley to determine the optimal attributes’ weights. Furthermore, a quantum-like Bayesian network is constructed to compute the quantum probabilities of alternatives and rank them, while utilizing Deng entropy measure to address the interference terms in the quantum probabilities. Finally, through the analysis of simulation data, the applicability of the model in evaluating emergency alternatives for meteorological disasters is verified, and its stability and practicality are further demonstrated through sensitivity analysis, comparative analysis and simulation analysis. The result shows that combining quantum computing with artificial intelligence achieves efficient consensus in intelligent group decision-making processes.}, archive = {J_EAAI}, author = {Yizhao Xu and Shuli Yan and Yangxue Li}, doi = {10.1016/j.engappai.2025.110086}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110086}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-attribute quantum group consensus model considering psychological preference}, volume = {144}, year = {2025}, }
• (2025). Multi-view collaborative signal fusion and representation property optimization for recommendation. EAAI, 144, 110085. (www)

  @article{ , abstract = {Modern recommender systems frequently grapple with challenges like data sparsity, data noise, and long-tail distribution, which limit the efficacy of these models. In the past, numerous studies have attempted to solve these problems through graph neural networks or other advanced techniques, achieving considerable performance. However, there still remains a significant lack in the application of self-supervised and contrastive learning approaches, leading to inadequate data representation capabilities, poor model generalization, and lower recommendation accuracy. To address these difficulties, we propose Multi-View Collaborative Signal Fusion and Representation Property Optimization for Recommendation (MCSF-RPORec), customized for item recommendations. The model comprises a preference encoding strategy, multi-view collaborative signal fusion, contrastive learning and representation optimization. Our model’s multi-view collaborative fusion approach effectively addresses the data sparsity issue by leveraging diverse sources of information, while contrastive learning and representation property optimization enhance representation robustness against noise. The advancement of our model is demonstrated through its application in item recommendations on Yelp and Epinions datasets. Our code is at https://github.com/CQURec/-MCSF-RPORec .}, archive = {J_EAAI}, author = {Pengfan Chen and Wei Zhou and Yao Chang and Jun Zeng and Junhao Wen}, doi = {10.1016/j.engappai.2025.110085}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110085}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-view collaborative signal fusion and representation property optimization for recommendation}, volume = {144}, year = {2025}, }
• (2025). MSPFNet: Multi-scale perceptual focusing network for scrap steel segmentation and classification. EAAI, 144, 110084. (www)

  @article{ , abstract = {Scrap steel recycling is one of the important measures to reduce carbon emissions and promote sustainable development. However, due to scrap steel data has the characteristics of a large number and variety of targets in the same image, disorderly arrangement, varying sizes, and mutual backgrounds, traditional manual rating methods are inefficient and highly influenced by subjective factors. Moreover, many efficient vision transformers methods currently face challenges in scrap steel segmentation and classification. To address these problems, we propose multi-scale perceptual focusing network for scrap steel segmentation and classification (MSPFNet). MSPFNet utilizes multi-scale feature extraction and perception mechanisms to focus on the key regions of scrap steel images, thereby improving the accuracy of scrap steel segmentation and classification. We apply MSPFNet to both public datasets (ImageNet-1K, ADE20K and Cityscapes) and scrap steel dataset in real-world application scenarios, experimental results show that MSPFNet achieves better performance than other existing methods with the similar parameter magnitude. MSPFNet-B (The base model of MSPFNet, MSPFNet is divided into three levels according to model size: tiny, small, and base) achieves mIoU (mean Intersection over Union) of 69.8% on the scrap steel dataset, surpassing other state-of-the-art segmentation models with the similar parameter magnitude, such as Segformer-B3 (The B3 model of Segformer, Segformer is divided into six levels according to model size: B0, B1, B2, B3, B4 and B5) and SegNeXt-B (The base model of SegNeXt) by 2.1% and 4.8%, respectively. Furthermore, extensive experiments were conducted on public datasets (ImageNet-1K, ADE20K and Cityscapes), showing that MSPFNet outperforms other existing methods with the similar parameter magnitude. We tested MSPFNet in a real-life scenario at a steel plant, and the model accurately segmented different levels of scrap steel. This research provides a new solution for the scrap steel recycling and processing industry, with the potential to improve the efficiency and quality of scrap steel recycling and achieve automation in the process.}, archive = {J_EAAI}, author = {Yunfeng Xu and Changda Liu and Jiakui Zhong and Wei Mei and Yan Zhang}, doi = {10.1016/j.engappai.2025.110084}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110084}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MSPFNet: Multi-scale perceptual focusing network for scrap steel segmentation and classification}, volume = {144}, year = {2025}, }
• (2025). Dual replay memory reinforcement learning framework for minority attack detection. EAAI, 144, 110083. (www)

  @article{ , abstract = {Cyberattacks are becoming more frequent as the number of systems connected to the Internet has grown significantly. Intrusion detection systems based on machine learning have become essential to cyber security because they can evaluate and find patterns in huge amounts of data. A branch of machine learning called reinforcement learning has shown promise in identifying new types of cyberattacks, but the class imbalance problem remains unresolved. Addressing this gap, this work presents a novel offline reinforcement learning framework that significantly enhances the capabilities of traditional Deep Q-network based reinforcement learning models for network intrusion detection. The innovation lies in using dual replay memory, which prioritizes learning minority attack classes. Experiments with and without dual-replay memory on the same offline reinforcement learning framework highlighted the significant improvements in evaluation metrics like accuracy, precision, recall and F1-Score. The research investigated the influence of various ratios of replay memories on learning through the grid search method. It evaluated their performance for each class using receiver operating characteristics and area under the curve. Experimental results using the one-vs-rest and aggregated approaches demonstrated the effectiveness of handling the class imbalance problem with high accuracy. Compared to existing methods, the technique performs better in detecting attacks and reducing false positive rates. The proposed model represents significant progress in intrusion detection, offering a robust and highly effective solution for detecting common and minority network intrusions.}, archive = {J_EAAI}, author = {Ankit Sharma and Manjeet Singh}, doi = {10.1016/j.engappai.2025.110083}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110083}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual replay memory reinforcement learning framework for minority attack detection}, volume = {144}, year = {2025}, }
• (2025). Competitive dual-students using bi-level contrastive learning for semi-supervised medical image segmentation. EAAI, 144, 110082. (www)

  @article{ , abstract = {Semi-supervised image segmentation aims to train the neural network with a small number of labeled images and a large number of unlabeled images, which helps to alleviate the burden of having less manually labeled medical data. However, the Mean-Teacher (MT) model, a benchmark method for semi-supervised medical segmentation, leads to a performance bottleneck as its student model eventually converges to the teacher model. In addition, existing segmentation methods treat all pixels equally and underestimate the importance of indistinguishable and underrepresented pixels, failing to mine the potential information in these regions effectively. To address the above issues, this paper proposes a Competitive Dual-Student (CDS) incorporating bi-level contrastive learning. First, an additional competitive dual-student model is added to the MT model and promoting knowledge sharing and complementarity among networks. Competitive instruction by the teacher through feature information exchange and positive comparisons reduces the accumulation of biased knowledge in the model. It stimulates the potential for further optimization of the model as a whole. Furthermore, a bi-level contrastive learning is designed. The high-level contrastive learning encourages competitive dual students to learn high-quality features from each other by constructing reliability constraints. The low-level contrastive achieved deep mining and accurate processing of local edge features by introducing class prototypes of high-quality features for teacher networks. Finally, the comprehensive experimental results on left atrium, brain tumor segmentation 2019 and automated cardiac diagnosis challenge datasets indicate that the segmentation performance of the proposed CDS outperforms the state-of-the-art compared methods. Code is released at https://github.com/FengZhao2001/CDS .}, archive = {J_EAAI}, author = {Gang Hu and Feng Zhao and Essam H. Houssein}, doi = {10.1016/j.engappai.2025.110082}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110082}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Competitive dual-students using bi-level contrastive learning for semi-supervised medical image segmentation}, volume = {144}, year = {2025}, }
• (2025). Efficient mining of closed high-utility itemsets in dynamic and incremental databases. EAAI, 144, 110081. (www)

  @article{ , abstract = {Closed high-utility itemsets (CHUIs) offer a concise overview of high-utility itemsets (HUIs). In this study, we focus on applying CHUI mining specifically in dynamic and incremental databases, where both item utility and transaction volumes can fluctuate. This paper investigates the mining of closed high-utility itemsets under these conditions by proposing a tree hierarchy with an updated vertical header list. This list is generated in a single database pass and updated with each new transaction. Two strategies are employed for pruning: transaction-weighted utilization ( twu ) and subtree utility ( su ). Additionally, a dedicated table facilitates the updating and insertion of new closed high-utility itemsets. This approach serves as a crucial preprocessing step in artificial intelligence (AI) applications, where high-quality, concise data patterns are essential for enhancing decision-making processes and significantly improving the quality and efficiency of data analysis. Experimental results indicate that the proposed algorithm significantly outperforms state-of-the-art algorithms for closed high-utility itemset mining in terms of runtime and memory usage while effectively scaling with increasing transaction volumes. In summary, this algorithm showcases high efficiency.}, archive = {J_EAAI}, author = {Mahnaz Naderi Vlashejerdi and Negin Daneshpour}, doi = {10.1016/j.engappai.2025.110081}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110081}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient mining of closed high-utility itemsets in dynamic and incremental databases}, volume = {144}, year = {2025}, }
• (2025). On the spatial prediction of the turbulent flow behind an array of cylinders via echo state networks. EAAI, 144, 110079. (www)

  @article{ , abstract = {This study investigates the capabilities of the spatial prediction of the turbulent flow behind an array of seven cylinders via Echo State Networks (ESN). The flow is dominated by five distinct unsteady vortex streets, each with its own characteristics, that interact with each other. The goal is to reconstruct the entire section of the flow field by training the ESN to predict the values based on the available low resolution experimental data from the rest of the flow field. Four scenarios – forward, backward, vertical, and central – of data availability to the ESN have been explored. The ESN is capable of reconstructing realistic flow fields in terms of the shape of the packs of up and downward vortices alongside their correct arrangements behind each cylinder. Furthermore, the variable phase synchronization of the neighboring cylinders has been well preserved. Nevertheless, there are preferable regions that result in better predictions of the rest of the flow when they are available to the network. Prediction of the downstream region is more challenging due to the increased entropy of the flow as the vortices move downstream. Furthermore, the most compact area of the information seems to be the central vortex street, as it produces better predictions with the lowest amount of input grid points. The spatial predictions remain stable over time, unaffected by divergence, and are largely insensitive to variations in network hyperparameters, except in cases of very small leaking rates.}, archive = {J_EAAI}, author = {M. Sharifi Ghazijahani and C. Cierpka}, doi = {10.1016/j.engappai.2025.110079}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110079}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {On the spatial prediction of the turbulent flow behind an array of cylinders via echo state networks}, volume = {144}, year = {2025}, }
• (2025). DFTQuake: Tripartite fourier attention and dendrite network for real-time early prediction of earthquake magnitude and peak ground acceleration. EAAI, 144, 110077. (www)

  @article{ , abstract = {Earthquakes are extremely destructive natural disasters, causing ground shaking, tsunamis, fissures, and landslides that result in loss of life. Early prediction of earthquake magnitude and peak ground acceleration (PGA) at a specific station using real-time recording can offer vital lead time, which can be utilized to save lives. In this paper, a novel deep-learning model, namely, Dendrite Fourier Transform Earthquake prediction (DFTQuake), has been designed specifically for earthquake magnitude and PGA prediction. The proposed model highlights the tripartite Fourier transform attention mechanism and dendrite connection. Fourier transform in feature maps is applied in the model to obtain periodic information, and the conventionally utilized fully connected layer at the end of the deep learning pipeline is replaced by a dendrite neural network. The DFTQuake model can predict the magnitude of an earthquake by utilizing the early three seconds of the primary waveform. A large number of strong motion records with train, test, and validation datasets of 17081, 6438, and 2760 records have been selected for this work. DFTQuake model outperforms the existing state-of-the-art artificial intelligence models by 13.16% and 34.21%, with a mean absolute error of 0.66 in Japan meteorological agency unit and 0.25 gal for early magnitude and PGA prediction. The model’s effectiveness is demonstrated through real-time predictions of magnitude and PGA at various recording sites for seven distinct earthquake events, including the damaging 2024 Noto earthquake. The model’s efficacy is further established by testing the model in India and nearby countries earthquake data.}, archive = {J_EAAI}, author = {Anushka Joshi and Nithya Reddy Vedium and Balasubramanian Raman}, doi = {10.1016/j.engappai.2025.110077}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110077}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DFTQuake: Tripartite fourier attention and dendrite network for real-time early prediction of earthquake magnitude and peak ground acceleration}, volume = {144}, year = {2025}, }
• (2025). A gaze-driven manufacturing assembly assistant system with integrated step recognition, repetition analysis, and real-time feedback. EAAI, 144, 110076. (www)

  @article{ , abstract = {Modern manufacturing faces significant challenges, including efficiency bottlenecks and high error rates in manual assembly operations. To address these challenges, we implement artificial intelligence (AI) and propose a gaze-driven assembly assistant system that leverages artificial intelligence for human-centered smart manufacturing. Our system processes video inputs of assembly activities using a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network for assembly step recognition, a Transformer network for repetitive action counting, and a gaze tracker for eye gaze estimation. The application of AI integrates the outputs of these tasks to deliver real-time visual assistance through a software interface that displays relevant tools, parts, and procedural instructions based on recognized steps and gaze data. Experimental results demonstrate the system’s high performance, achieving 98.36% accuracy in assembly step recognition, a mean absolute error (MAE) of 4.37%, and an off-by-one accuracy (OBOA) of 95.88% in action counting. Compared to existing solutions, our gaze-driven assistant offers superior precision and efficiency, providing a scalable and adaptable framework suitable for complex and large-scale manufacturing environments.}, archive = {J_EAAI}, author = {Haodong Chen and Niloofar Zendehdel and Ming C. Leu and Zhaozheng Yin}, doi = {10.1016/j.engappai.2025.110076}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110076}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A gaze-driven manufacturing assembly assistant system with integrated step recognition, repetition analysis, and real-time feedback}, volume = {144}, year = {2025}, }
• (2025). An adaptive weighted stacking ensemble framework for photovoltaic power generation forecasting with joint optimization of features and hyperparameters. EAAI, 144, 110075. (www)

  @article{ , abstract = {Accurate photovoltaic (PV) power forecasting is crucial for the efficient operation of power systems. However, the limited availability of historical data often hinders prediction accuracy. To address this challenge, an adaptive weighted stacking (AWS) ensemble framework is proposed, combining artificial intelligence ensemble models with joint optimization (JO) of features and hyperparameters to enhance prediction performance in small sample (SS) scenarios. In the framework, a base learner selection strategy based on kernel hierarchical clustering (KHC) is used to balance model performance and diversity. Multi-initiated parallel bayesian optimization with early stopping (MI-PBOES) algorithm is employed for the JO of features and hyperparameters to obtain the optimal subset of features and the corresponding parameter configuration. Additionally, an AWS strategy is employed to dynamically adjust ensemble weights, further improving accuracy and robustness. To evaluate its effectiveness in engineering applications, experiments were conducted using data from the desert knowledge australia (DKA) solar centre. The results indicate that the proposed framework reduces root mean square error ( RMSE ) by 51.84% and mean absolute error ( MAE ) by 56.42% compared to traditional machine learning models, while RMSE decreases by 129.89% and MAE by 124.82% compared to deep learning models. Under noisy conditions, RMSE and MAE are reduced by 39.46% and 37.37%, respectively, showing remarkable prediction performance and robustness.}, archive = {J_EAAI}, author = {Shaolong Zheng and Danyun Li and Yidong Li and Zihan Zeng and Yuxuan Zhou and Changsi Li and Kun Song}, doi = {10.1016/j.engappai.2025.110075}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110075}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive weighted stacking ensemble framework for photovoltaic power generation forecasting with joint optimization of features and hyperparameters}, volume = {144}, year = {2025}, }
• (2025). A graph convolutional neural network that dynamically adapts its architecture to input data at inference time to predict the short-term states of traffic. EAAI, 144, 110074. (www)

  @article{ , abstract = {Every deep learning model that is designed to forecast traffic states, takes a fixed form in its structure. However, the limitation of such a fixed model architecture is that it will be neither accurate nor efficient in accommodating varying traffic conditions when forecasting traffic states. When a traffic state is stable with no congestion, a simple model architecture should be sufficient for predicting the state of short-term traffic. Under interrupted traffic conditions, however, a deeper model is necessary if the extensive spatio-temporal interactions, both in and between traffic states, is to be recognized. A sufficiently deep neural network can cover various traffic conditions once it has been trained on a large scale. Evaluating such a model at inference time is expensive, while its employment can be redundant unless traffic conditions are always congested. To address these issues, we devised a neural network that dynamically adapts its architecture to input traffic conditions at inference time. Thus, once a full-size model with the maximum number of layers is trained in advance, the sub-portion of the model is autonomously selected based on input traffic states and used for inference, which allows it to efficiently predict the traffic state. It was subsequently found that the proposed model was competitive against “state-of-the-art” deep learning models in forecasting traffic speeds on a network-wide scale. The computation time for inference also dwindled considerably in comparisons with reference models.}, archive = {J_EAAI}, author = {Chaewon Kwon and Gyeongjun Kim and Keemin Sohn}, doi = {10.1016/j.engappai.2025.110074}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110074}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A graph convolutional neural network that dynamically adapts its architecture to input data at inference time to predict the short-term states of traffic}, volume = {144}, year = {2025}, }
• (2025). Hybrid cross-modality fusion network for medical image segmentation with contrastive learning. EAAI, 144, 110073. (www)

  @article{ , abstract = {Medical image segmentation has been widely adopted in artificial intelligence-based clinical applications. The integration of medical texts into image segmentation models has significantly improved the segmentation performance. It is crucial to design an effective fusion manner to integrate the paired image and text features. Existing multi-modal medical image segmentation methods fuse the paired image and text features through a non-local attention mechanism, which lacks local interaction. Besides, they lack a mechanism to enhance the relevance of the paired features and keep the discriminability of unpaired features in the training process, which limits the segmentation performance. To solve the above problem, we propose a hybrid cross-modality fusion network (HCFNet) based on contrastive learning for medical image segmentation. The key designs of our proposed method are a multi-stage cross-modality contrastive loss and a hybrid cross-modality feature decoder. The multi-stage cross-modality contrastive loss is utilized to enhance the discriminability of the paired features and separate the unpaired features. Furthermore, the hybrid cross-modality feature decoder conducts local and non-local cross-modality feature interaction by a local cross-modality fusion module and a non-local cross-modality fusion module, respectively. Experimental results show that our method achieved state-of-the-art results on two public medical image segmentation datasets.}, archive = {J_EAAI}, author = {Xichuan Zhou and Qianqian Song and Jing Nie and Yujie Feng and Haijun Liu and Fu Liang and Lihui Chen and Jin Xie}, doi = {10.1016/j.engappai.2025.110073}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110073}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid cross-modality fusion network for medical image segmentation with contrastive learning}, volume = {144}, year = {2025}, }
• (2025). A unified multi-criteria equity metric for data-driven decision-making: Application to tract-level electric vehicle charging station planning. EAAI, 144, 110070. (www)

  @article{ , abstract = {This paper introduces a unified, data-driven multi-criteria equity metric designed to address inequities in Electric Vehicle Charging Station (EVCS) accessibility and Electric Vehicle (EV) adoption. Existing studies predominantly focus on macro-level factors and rely on single-criterion or context-specific approaches, limiting their applicability to dynamic, multi-dimensional challenges. Additionally, current equity evaluations often involve inherent subjectivity, complicating decision-making processes. Focusing on the Dallas-Fort Worth area, this study advances the field by transitioning from city-level to census tract-level analysis. The proposed framework utilizes dimension reduction techniques to construct a unified equity metric and high-resolution clustering analysis to identify EV-related socio-demographic typologies. These typologies—EV developed, EV developing, and EV underdeveloped/disadvantaged—are validated against disadvantaged communities (DACs) defined by the Climate and Economic Justice Screening Tool (CEJST), mapped EV/EVCS distributions, and their corresponding socio-demographic radar plots. The data-driven equity metric has advantages in unity, objectivity, adaptability, and robustness compared to existing metrics. By identifying micro-level inequities, the framework provides tailored policy recommendations for each typology, supporting equitable EV infrastructure planning.}, archive = {J_EAAI}, author = {Zhi Li and Janille Smith-Colin and Jianhui Wang}, doi = {10.1016/j.engappai.2025.110070}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110070}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A unified multi-criteria equity metric for data-driven decision-making: Application to tract-level electric vehicle charging station planning}, volume = {144}, year = {2025}, }
• (2025). A transformer-based multi-platform sequential estimation fusion. EAAI, 144, 110069. (www)

  @article{ , abstract = {This paper considers estimation fusion problem in the case of unknown correlations among local estimates, motivated by multi-sensor target tracking with correlated measurement noises. A Transformer-based sequential multi-platform fusion method is put forward by learning data features of historical local tracks, instead of numerical optimization in existing weighting fusion. Firstly, a neural network-based sequential fusion framework is proposed, where it owns a hierarchical structure and sequential training process to adapt to different numbers of local tracks without changing network parameters and retraining. Secondly, the Taylor expansion-based positional encoding in Transformer network is constructed, by using a third-order Taylor expansion to approximately replace original sin and cos functions to better extract aperiodic variation features of input sequence. Thirdly, by arranging different local estimates of input sequence in time order, a max–min normalization-based data pre-processing and its inverse process are presented, to prevent precision truncation and retain data diversity. An example of target tracking with multiple sensors show that the proposed method owns superior fusion precision than that of the sequential filter, simple convex combination, covariance intersection and Long Short-Term Memory-based sequential fusion methods, in terms of different correlation coefficients. And its fusion precision is also improved with the increasing of sensor numbers.}, archive = {J_EAAI}, author = {Xupeng Zhai and Yanbo Yang and Zhunga Liu}, doi = {10.1016/j.engappai.2025.110069}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110069}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A transformer-based multi-platform sequential estimation fusion}, volume = {144}, year = {2025}, }
• (2025). Nonlinear time-varying system response modeling via a real-time updated runge-kutta physics-informed neural network. EAAI, 144, 110067. (www)

  @article{ , abstract = {Accurate structural response modeling advances the understanding of complex dynamic systems and enables effective structural design, control and monitoring. Due to damage, engineering structures will exhibit nonlinear and time-varying characteristics, which makes the structural response highly complex and challenging to be accurately modeled by traditional methods. This study proposes a Runge-Kutta-based real-time updated physics-informed neural network (RTU-PINN) to model structural responses of complex dynamic systems. The Recurrent Neural Network (RNN) is enhanced with Runge-Kutta method and neural Ordinary Differential Equations (neural ODEs) to model the system responses. The unknown structural parameters and nonlinear restoring force can be identified from the neural network. To capture the nonlinear and time-varying characteristics caused by structural damage or hysteretic behaviors, a Real-Time Updating (RTU) strategy is proposed to update the time-varying parameters and nonlinear restoring force with a sliding time window to minimize the discrepancy between predicted result and testing data. In addition, the proposed method can be applied to high-dimensional time-varying structures. Performance of the proposed method is examined via numerical and laboratory case studies. It is found that the RTU-PINN models the dynamic response with Relative Root Mean Square Error (RRMSE) values less than 0.001 in the target nonlinear time-varying dynamic system. Research results show that the proposed RTU-PINN method can accurately model the complex dynamic responses of the numerical and experimental structures with time-varying and nonlinear characteristics. The proposed method has the potential to address modeling uncertainties/errors and is applicable for system identification of complex systems with nonlinear and time-varying features.}, archive = {J_EAAI}, author = {Huaguan Li and Chao Sun}, doi = {10.1016/j.engappai.2025.110067}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110067}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Nonlinear time-varying system response modeling via a real-time updated runge-kutta physics-informed neural network}, volume = {144}, year = {2025}, }
• (2025). A transfer reinforcement learning and digital-twin based task allocation method for human-robot collaboration assembly. EAAI, 144, 110064. (www)

  @article{ , abstract = {Currently, human-robot collaboration systems are considered to have great application potential in complex and flexible assembly tasks. In human-robot collaborative assembly systems, the roles and tasks allocation between human and robot are the vital stage to exert and integrate the strength of both. Current many researches of task allocation mainly train decision-making model depending on a large number of predefined standard assembly data such as assembly time and assembly difficulty. However, due to individual differences and frequent product updates, the dynamic assembly conditions lack enough data for model training, and the constructed model based on historical data may not adapt to the new condition. To solve this gap, a transfer reinforcement learning and digital-twin based task allocation method is proposed to achieve the accurate and efficient multi-agent human-robot collaboration task allocation policy learning. Firstly, based on the digital twin of human-robot collaboration environment, the augmented reality is leveraged to simulate assembly process and collect execution data of workers and robots before the physical assembly. Secondly, a multi-agent reinforcement learning method is introduced to use domain randomization strategy to pre-train decision policy before the accurate model training. Thirdly, the knowledge distillation is leveraged to develop the transfer reinforcement learning framework to learn the accurate decision policy by reusing the pre-trained policy and utilizing the collected assembly simulation data. Finally, a case study is performed to demonstrate the effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Jingfei Wang and Yan Yan and Yaoguang Hu and Xiaonan Yang and Lixiang Zhang}, doi = {10.1016/j.engappai.2025.110064}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110064}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A transfer reinforcement learning and digital-twin based task allocation method for human-robot collaboration assembly}, volume = {144}, year = {2025}, }
• (2025). Anti-damage ability of biological plausible spiking neural network with synaptic time delay based on speech recognition under random attack. EAAI, 144, 110061. (www)

  @article{ , abstract = {The complex electromagnetic environment can reduce the performance of neuromorphic hardware. The robustness of a brain-inspired model is beneficial for maintaining its performance under exterior attack. However, the synaptic plasticity in a brain-inspired model still lacks biological plausibility. The objective of this paper is to enhance the anti-damage ability of brain-inspired model under random attack by improving the biological plausibility of its synaptic plasticity. In this study, we propose a new spiking neural network (SNN) as a brain-inspired model, in which the topology is constrained by functional magnetic resonance imaging (fMRI) data on the human brain, its nodes are Izhikevich neuron models, and its edges are synaptic plasticity models with a random time delay that conforms to the range of the biological synaptic time delay (STD), and called it as fMRI-SNN with synaptic random time delay (SRTD). Then, taking speech recognition (SR) as a case study, we certify the recognition performance of fMRI-SNN with SRTD. To evaluate the anti-damage ability of fMRI-SNN with SRTD, we compare its SR accuracy before and after random attack. To elucidate the anti-damage mechanism, we discuss the neuroelectric characteristics, adaptive regulation of synaptic plasticity, and dynamic topological characteristics of fMRI-SNN with SRTD under random attack. The results indicate that our approach enhances the anti-damage ability of the brain-inspired model, and our discussion elucidates its anti-damage mechanism. Our results prompt that the brain-inspired model with biological plausibility can enhance its information processing ability.}, archive = {J_EAAI}, author = {Lei Guo and Weihang Ding and Youxi Wu and Menghua Man and Miaomiao Guo}, doi = {10.1016/j.engappai.2025.110061}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110061}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Anti-damage ability of biological plausible spiking neural network with synaptic time delay based on speech recognition under random attack}, volume = {144}, year = {2025}, }
• (2025). Sparse temporal aware capsule network for robust speech emotion recognition. EAAI, 144, 110060. (www)

  @article{ , abstract = {Recent advancements in deep learning have significantly improved speech emotion recognition systems, particularly through the integration of attention mechanisms and convolutional architectures. However, challenges such as generalization, robustness, and reproducibility persist, especially when working with diverse, real-world datasets. To address these issues, we introduce the Sparse Temporal-Aware Capsule Network (STACN), a novel architecture designed to enhance the accuracy and reliability of speech emotion recognition systems. The model captures emotional nuances in speech through five key components: sparsity-enhanced feature mapping, a multi-head attention mechanism, a temporal-aware block, a residual module, and a capsule module. The temporal-aware block preserves long-range dependencies in the speech signal, while the multi-head attention mechanism effectively captures diverse emotional cues. The capsule module, combined with dynamic routing, enhances the preservation of spatial hierarchies and improves feature interpretability. Additionally, a sparse dropout mechanism helps mitigate overfitting, thereby improving the model's generalizability. To ensure consistent performance across various implementations and datasets, we propose novel reproducibility metrics. STACN is rigorously evaluated across six benchmark datasets, demonstrating robust performance with accuracies of 94.81% on EMODB, 78.12% on SAVEE, 89.17% on CASIA, 91.70% on BodEMODB, 68.82% on IEMOCAP, and 95.92% on ESD. These results highlight the model's versatility and robustness, achieving high accuracy across datasets with varying complexities—from well-annotated emotional speech to spontaneous and overlapping expressions. With its comprehensive design and rigorous evaluation, STACN establishes itself as a reliable and scalable solution for real-world speech emotion recognition applications.}, archive = {J_EAAI}, author = {Huiyun Zhang and Heming Huang and Puyang Zhao and Zhenbao Yu}, doi = {10.1016/j.engappai.2025.110060}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110060}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sparse temporal aware capsule network for robust speech emotion recognition}, volume = {144}, year = {2025}, }
• (2025). Experimental validation of a semi-active fuzzy control strategy based on deep reinforcement learning for a piezoelectric smart isolation system. EAAI, 144, 110058. (www)

  @article{ , abstract = {The high-tech industry, commanding a substantial portion of the global market, remains highly susceptible to disruptions caused by earthquakes because its products rely heavily on vibration-sensitive equipment. Moreover, the characteristics of seismic ground motions are usually difficult to predict; therefore, seismic protection of high-tech equipment is a challenging task. In this context, this study leverages the robustness and adaptability of artificial intelligence techniques, particularly deep reinforcement learning (DRL) and a piezo-electric smart isolation system (PSIS) to protect high-tech equipment. The control objective of the PSIS is to dynamically optimize the isolation performance of the PSIS in real time, accommodating diverse ground motion characteristics. The proposed control strategy for PSIS involves a DRL framework integrated with a fuzzy inference system (FIS) and non-striking friction (NSF) control technique. In the DRL framework, a Deep Deterministic Policy Gradient (DDPG) algorithm is utilized to train the DRL model, employing tailored reward functions to achieve diverse control objectives under varying seismic conditions. The efficacy of the proposed PSIS control strategy is evaluated through numerical simulations and experimental validation. The proposed DRL control module demonstrated superior isolation performance compared to other control algorithms, particularly in reducing isolation displacement by approximately 24% during near-fault earthquakes and superstructure acceleration of 39% reduction under far-field earthquakes. The research outcomes underscore the effectiveness of the proposed approach, positioning it as a promising solution for fortifying the seismic resilience of vibration-sensitive equipment and a strong support of advancing the current nano-scale manufacturing process in the high-tech industry.}, archive = {J_EAAI}, author = {Tzu-Kang Lin and Chandrasekhara Tappiti and Lyan-Ywan Lu and Ting-Kuan Lin}, doi = {10.1016/j.engappai.2025.110058}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110058}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Experimental validation of a semi-active fuzzy control strategy based on deep reinforcement learning for a piezoelectric smart isolation system}, volume = {144}, year = {2025}, }
• (2025). Exploring the synergies of hybrid convolutional neural network and vision transformer architectures for computer vision: A survey. EAAI, 144, 110057. (www)

  @article{ , abstract = {The hybrid of Convolutional Neural Network (CNN) and Vision Transformer (ViT) architecture has emerged as a groundbreaking approach, pushing the boundaries of Computer Vision (CV), significantly advancing CV tasks such as image classification, object detection, and segmentation. This comprehensive review provides a thorough examination of the literature on state-of-the-art hybrid CNN-ViT architectures, exploring the synergies between these two approaches. The main content of this survey includes: (1) a background on the vanilla CNN and ViT, (2) systematic review of various taxonomic hybrid designs to explore the synergy achieved through merging CNN and ViT models, (3) comparative analysis, task-specific synergy and real-world application among various hybrid architectures, (4) challenges and future directions for hybrid models, (5) lastly, the survey concludes with a summary of key findings and recommendations. Through this exploration, the survey aims to serve as a guiding resource, enhancing understanding of the dynamics between CNN and ViT and their impact on future developments in CV.}, archive = {J_EAAI}, author = {Yunusa Haruna and Shiyin Qin and Abdulrahman Hamman Adama Chukkol and Abdulganiyu Abdu Yusuf and Isah Bello and Adamu Lawan}, doi = {10.1016/j.engappai.2025.110057}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110057}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploring the synergies of hybrid convolutional neural network and vision transformer architectures for computer vision: A survey}, volume = {144}, year = {2025}, }
• (2025). Multi-objective optimization of continuous monitoring scheduling for moving targets by earth observation satellites. EAAI, 144, 110056. (www)

  @article{ , abstract = {Continuous monitoring scheduling for moving targets by earth observation satellites is a crucial optimization problem in the field of artificial intelligence. In this scenario, for moving targets, extending the observation duration and increasing the capture times are important. However, few studies consider optimizing these two objectives simultaneously. In this paper, we present a novel methodology for this multi-objective optimization scenario. Firstly, a multi-objective optimization model is established. To address the complex Pareto front of the problem, we propose a decomposition-based, weight-adjusted multi-objective evolutionary algorithm that demonstrates strong convergence on the baseline model while further enhancing diversity through weight-adjustment techniques. Experimental results demonstrate that: 1) The proposed method achieves a trade-off solution set that simultaneously balances two objectives, and 2) in comparison to existing multi-objective optimization methods, the proposed algorithm outperforms the existing algorithms in terms of convergence and diversity.}, archive = {J_EAAI}, author = {Tao Chao and Xiaofeng Han and Xiang Li and Ming Yang}, doi = {10.1016/j.engappai.2025.110056}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110056}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective optimization of continuous monitoring scheduling for moving targets by earth observation satellites}, volume = {144}, year = {2025}, }
• (2025). Towards robust shielded reinforcement learning through adaptive constraints and exploration: The fear field framework. EAAI, 144, 110055. (www)

  @article{ , abstract = {Machine Learning (ML) techniques, including Reinforcement Learning (RL), demonstrate potential as decision-making controllers. However, enhancing the robustness required for real-world deployment remains imperative. Within the realm of Safe RL, Shielded RL emerges as a solution, employing shields to block actions leading to unsafe states and offering safe alternatives through known policies. Yet, many Shielded RL methods rely on dynamic environment models, which may inaccurately predict future states, compromising controller robustness. We introduce the Fear Field framework to mitigate this issue for discrete Markov Decision Process-based (MDP) shields with strictly connected unsafe state spaces and fully observable states, which adjusts safe operation constraints based on disparities between model predictions and actual environmental dynamics. We employ parallel learning and Curriculum Learning (CL) strategies to mitigate lengthy training times in high state-space size environments. Additionally, an adaptive exploration algorithm enhances convergence rates amidst significant environmental dynamic shifts. In our case study, integrating CL and the adaptive exploration algorithm with the Fear Field framework reduces unsafe state occurrences by two orders of magnitude while enhancing convergence time following sudden environmental changes. The Fear Field framework significantly reduces unsafe states in the Frozen Lake Gridworld environment at low computational expense when model predictions deviate from reality, with negligible costs otherwise.}, archive = {J_EAAI}, author = {Haritz Odriozola-Olalde and Maider Zamalloa and Nestor Arana-Arexolaleiba and Jon Perez-Cerrolaza}, doi = {10.1016/j.engappai.2025.110055}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110055}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards robust shielded reinforcement learning through adaptive constraints and exploration: The fear field framework}, volume = {144}, year = {2025}, }
• (2025). Multiple scales fusion and query matching stabilization for detection with transformer. EAAI, 144, 110047. (www)

  @article{ , abstract = {Recent advances in object detection with Transformer-based models like Detection with Transformer (DETR) have improved performance, but challenges remain with multi-scale fused features. These features introduce redundant tokens and bias toward larger objects, slowing down training. To overcome these issues, we propose two novel encoders: the Similarity-based Deduplication Encoder (SDE) and the Hybrid Multi-object Encoder (HMoE). HMoE employs an offset-based attention window to enhance local attention for objects of varying sizes across feature maps, while SDE reduces redundancy by calculating attention scores across multiple scales. Additionally, we introduce a One-to-many Positive Matching (OmPM) strategy to improve query stability. OmPM generates query vectors from multiple positive samples, resulting in more diverse and semantically meaningful queries. Our model demonstrates substantial performance improvements. On the Visual Object Classes Challenge 2007 dataset, it achieves a +5.04 mean Average Precision ( mAP ) and +5.1 Average Precision for small objects ( APs ) for small objects in just 24 epochs . On the Microsoft Common Objects in Context (COCO) dataset, the model reaches 50.1 mAP and 34.2 APs in only 8 epochs , and 52.4 mAP and 35.6 APs in 24 epochs . This significantly accelerates convergence, reducing training time by 66% compared to benchmarks while maintaining or exceeding detection accuracy. Furthermore, our model achieves 27 Frames Per Second ( FPS ) on the COCO dataset, setting a new record among DETR-like methods with high detection accuracies.}, archive = {J_EAAI}, author = {Shenyu Du and Xijun Liang and Kun Wu and Ye Tian and Yue Liu and Ling Jian}, doi = {10.1016/j.engappai.2025.110047}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110047}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multiple scales fusion and query matching stabilization for detection with transformer}, volume = {144}, year = {2025}, }
• (2025). Optimizing optic cup and optic disc delineation: Introducing the efficient feature preservation segmentation network. EAAI, 144, 110038. (www)

  @article{ , abstract = {Recently, computer vision and healthcare had significantly enhanced glaucoma diagnosis, necessitating automated solutions due to the rising prevalence and subjective nature of current diagnostic methods. Optic cup (OC) and optic disc (OD) segmentation in retinal fundus imaging is crucial, yet challenging due to non-distinctive OC boundaries and image property variations. Ongoing efforts to develop automated diagnostic systems aim to enhance glaucoma detection. Accurate OC and OD segmentation has great potential for providing valuable clinical insights and validating glaucoma screening effectiveness. Retinal fundus images exhibit wide variations in dimensions, shape, color, and background, posing segmentation challenges. Consequently, we propose the efficient feature preservation segmentation network (EFPS-Net) to accurately segment OC and OD. EFPS-Net incorporates an atrous residual block for capturing features and facilitating computational efficiency on the encoder side, along with a recurrent residual block on the decoder side to prevent information loss. Additionally, intrinsic feature fusion and extrinsic feature aggregation were incorporated to improve the model's learning ability and efficient feature preservation. The proposed network achieves high segmentation performance even with only 2.6 million parameters. Our model was evaluated on five publicly available retinal fundus image datasets and outperformed state-of-the-art models. These results highlight EFPS-Net's advancements in empowering healthcare professionals in glaucoma diagnosis and treatment by leveraging advanced artificial intelligence techniques with engineering innovations.}, archive = {J_EAAI}, author = {Zain Ul Abidin and Rizwan Ali Naqvi and Hyung Seok Kim and Hak Seob Kim and Daesik Jeong and Seung-Won Lee}, doi = {10.1016/j.engappai.2025.110038}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110038}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing optic cup and optic disc delineation: Introducing the efficient feature preservation segmentation network}, volume = {144}, year = {2025}, }
• (2025). Software maintainability prediction based on change metric using neural network models. EAAI, 144, 110032. (www)

  @article{ , abstract = {The increased software system complexity requires better approaches to advise developers in the process of changing parts of the software during the maintenance and evolution process. The aim of this paper is twofold: (1) to provide an in-depth analysis and comparison of various existing and newly developed maintainability prediction models, and (2) to investigate and obtain the best set of metrics for software maintainability prediction. Various code metrics were employed to explore and discover the best model, three previously developed configuration metrics as dependent variables, i.e. with 10 metrics, 2 metrics, and 3 metrics. The new proposed model considers the correlation between code metrics and the maintenance change metric (four and five metrics were considered in the two used projects). The overall framework of our Neural Network Change Prediction (NNCP) approach uses an artificial neural network algorithm with various input layers. Different validation methods were employed: k-fold cross-validation and hold-out validation. Statistical analysis performed using the Wilcoxon signed rank test between the four models shows that the new proposed model can better predict the maintainability change metric, having one metric in common with the compared model but considering other three distinct ones.}, archive = {J_EAAI}, author = {Andreea Vescan and Daniel Barac-Antonescu}, doi = {10.1016/j.engappai.2025.110032}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110032}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Software maintainability prediction based on change metric using neural network models}, volume = {144}, year = {2025}, }
• (2025). Rectified self-supervised monocular depth estimation loss for nighttime and dynamic scenes. EAAI, 144, 110026. (www)

  @article{ , abstract = {Self-supervised monocular depth estimation has attracted much attention in computer vision recently. However, most existing methods assume that the scenes are static and the photometric is consistent, so that their performances tend to degrade significantly in nighttime and dynamic scenes. To address this issue, this paper proposes a self-supervised monocular depth estimation model to tackle two challenges. One is the drastic photometric changes problem due to underexposure of distant areas in nighttime scenes, the other is the moving objects problem in dynamic scenes. In the proposed model, an Effective Area Photometric Loss function (EAPL) is designed which is gated by the Effective Area Mask (EAM) and Potentially Moving Objects Mask (PMOM). Then, a motion flow network is introduced to estimate the motion of moving objects, and a Motion Flow Loss function (MFL) is proposed based on three facts, i.e., the motion flow of static objects should be zero, most moving objects in autonomous driving scenarios are approximately rigid objects, and the relative motion flows between consecutive frames should be mutually inverse. Finally, a decoupled training approach is provided to facilitate the optimization process of the model. Experimental results show that our model achieves state-of-the-art or second best performance on the nuTonomy Scenes (nuScenes) and Dense Depth for Autonomous Driving (DDAD) dataset which contains many nighttime or dynamic scenes, and also achieves competitive performance on the Karlsruhe Institute of Technology and Toyota Technological Institute at Chicago (KITTI) dataset which is dominated by daytime and static scenes. Codes are available at https://github.com/pandaswfas/effdepth .}, archive = {J_EAAI}, author = {Xiaofei Qin and Lin Wang and Yongchao Zhu and Fan Mao and Xuedian Zhang and Changxiang He and Qiulei Dong}, doi = {10.1016/j.engappai.2025.110026}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110026}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rectified self-supervised monocular depth estimation loss for nighttime and dynamic scenes}, volume = {144}, year = {2025}, }
• (2025). Tiny network for faults recognition in freight cars. EAAI, 144, 110018. (www)

  @article{ , abstract = {The Trouble of Moving Freight Car Detection System (TFDS) plays a critical role in railway maintenance, but existing fault detection methods often struggle to accurately distinguish specific fault types, limiting their practical application and maintenance guidance. These methods typically require extensive fault samples and significant computational resources. The current TFDS still relies on manual inspection for fault detection, which can be time-consuming and prone to human error. To address these limitations, this study introduces Tiny-Classify, a lightweight fault recognition network that integrates both shallow and deep features to effectively identify fault types. By employing a lightweight backbone and incorporating attention mechanisms, our approach improves accuracy while significantly reducing model size and computational resources. To overcome the scarcity of TFDS samples, we propose a novel data augmentation method tailored to simulate real-world operational conditions, generating diverse fault instances to enrich the dataset and enhance the model's robustness and generalization capabilities. Evaluation on a real TFDS dataset including specific fault types—brake shoe slag, lost brake shoe pin, lost brake shoe pin ring, and brake shoe overrun—demonstrates that Tiny-Classify achieves a classification accuracy of 99.2% with a compact model size of 0.8 Megabytes. These results underscore its potential for practical deployment in railway freight car maintenance, providing a versatile and efficient framework for real-time fault classification while contributing to enhanced railway safety and operational efficiency.}, archive = {J_EAAI}, author = {Yuan Liu and Hongmei Shi and Ji Qiu and Zujun Yu}, doi = {10.1016/j.engappai.2025.110018}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110018}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Tiny network for faults recognition in freight cars}, volume = {144}, year = {2025}, }
• (2025). Early fault detection with health stage division: An adversarial learning method with model update mechanism for condition monitoring. EAAI, 144, 110015. (www)

  @article{ , abstract = {Early fault detection (EFD) presents significant challenges due to the subtle nature of abnormal signals during this phase. Traditional EFD methods primarily focus on identifying the onset time of failures. In this paper, we propose a novel EFD approach that simultaneously achieves early fault detection and health stage classification. We begin by developing a threshold-free fault detector utilizing adversarial learning, enabling the network to perform supervised classification and unsupervised fault detection concurrently. To further enhance the stability of the adversarial network, we introduce a straightforward ensemble learning technique. A pivotal aspect of this study is the conceptualization of the EFD problem as fault detection with continuously emerging new classes (FDENC), wherein different stages of a fault are treated as new classes. To support this framework, we investigate a self-model update mechanism based on the designed network. Unlike traditional EFD paradigms, our proposed method not only identifies the onset of early failures but also facilitates health stage classification, thereby indicating the progression of the fault. We evaluate the effectiveness of our approach through two run-to-failure experiments designed to induce sub-surface cracks. The experimental results demonstrate the method's sensitivity to the emergence and progression of micro-cracks in bearings. Additionally, we employed two benchmark datasets to assess the method's performance, further highlighting its superiority in both fault detection and health stage classification.}, archive = {J_EAAI}, author = {Yu Wang and Qingbo Wang and Szymon Bernat and Alexey Vinogradov}, doi = {10.1016/j.engappai.2025.110015}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110015}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Early fault detection with health stage division: An adversarial learning method with model update mechanism for condition monitoring}, volume = {144}, year = {2025}, }
• (2025). Multitask genetic programming for automated design of heuristics for the container relocation problem. EAAI, 144, 110001. (www)

  @article{ , abstract = {Automated design of heuristics with genetic programming (GP) received significant attention for various optimisation problems. One of these is the container relocation problem (CRP), encountered in shipping terminals and warehouses. Although GP can design high quality heuristics, it has certain limitations. One is that numerous variants of each optimisation problem exist, meaning that appropriate heuristics have to be designed for each. Most commonly these heuristics are evolved independently from each other, which can be regarded as wasteful since they should exhibit a certain kind of similarity because they solve the same problem. This encouraged the development of knowledge transfer methods, such as multitask learning, that share the knowledge between solutions evolved for different problem types. In multitask learning several problems are solved simultaneously, with solutions of different problems shared between the populations that solve each problem. Until now, in the context of the automated design of heuristics, multitask learning has been primarily utilised to solve scheduling problems with a limited number of different problem types. This motivates us to investigate multitask learning with two main objectives. First, to apply multitask learning to evolve heuristics for a different problem, namely CRP. Second, to investigate how problem types that are optimised simultaneously influence its ability to obtain efficient heuristics. The obtained results demonstrate that multitask learning improves the convergence of GP, although the overall performance of the obtained heuristics remains similar. Furthermore, the effectiveness of multitask learning significantly depends on the problem types optimised simultaneously, outlining that the choice of the problems is critical.}, archive = {J_EAAI}, author = {Marko Đurasević and Mateja Đumić and Francisco Javier Gil Gala}, doi = {10.1016/j.engappai.2025.110001}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110001}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multitask genetic programming for automated design of heuristics for the container relocation problem}, volume = {144}, year = {2025}, }
• (2025). Masticatory organ dysfunction recognition based on multimodal information. EAAI, 144, 109998. (www)

  @article{ , abstract = {This paper describes research on designing a system for initial masticatory dysfunction recognition, a study with significant implications for patient treatment. The input contains three sources of information: text from questionnaires, videos, and face images acquired using mobile applications. We collected a dataset of 86 patients. Because of the relatively small dataset, we used binary classification in this research. We applied three classic machine learning models: Regularized Logistic Regression (RLR), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGB). We designed several features acquired from video and images and selected some using different selection methods. In the experimental study, we assessed which source of information is the most discriminative for this classification. We achieved the highest F1 measure (macro) using textual data. Classification results based on dynamic features from video achieved 0.643 F1 measure (macro). Features from static images represent the worst predictive value, and the best result is 0.5694. This result does not support our hypothesis that facial asymmetry is essential in diagnosing masticatory disorders. However, the identified asymmetry is the basis for further tests to correct posture problems and additional tests by a physiotherapist, which underscores the practical implications of our research on the patient’s treatment.}, archive = {J_EAAI}, author = {Urszula Markowska-Kaczmar and Michał Karolewski and Michał Wawrzyniak and Karol Szymończyk and Joanna Szołomicka and Jakub Pyka and Karol Kulawiec and Aneta Demidaś}, doi = {10.1016/j.engappai.2024.109998}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109998}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Masticatory organ dysfunction recognition based on multimodal information}, volume = {144}, year = {2025}, }
• (2025). Towards real-time adaptive prediction of rotary kiln processes: An enhanced framework combining parallel temporal convolution and long short-term memory networks. EAAI, 144, 109993. (www)

  @article{ , abstract = {Accurate modeling of critical quality variables is crucial for stable control and effective optimization of the rotary kiln process. However, offline trained models often fail during online application due to performance degradation and process dynamics. The inherent physical and chemical reactions further complicate traditional models, preventing them from accurately capture the dynamic behavior of the rotary kiln. To address these challenges, we propose a deep neural network that integrates a parallel temporal convolutional network with a long short-term memory unit. This model accurately predicts rotary kiln tail temperature (RKTT) by capturing multi-scale and long-term dependencies within the data, and extracting global and dynamic features from time series data. Additionally, we employ transfer learning techniques to reuse learned knowledge and facilitate rapid model updates. Specifically, we propose four transfer learning strategies and a model evaluation function to dynamically select the most appropriate model update strategies. Then, by combining online learning with sliding window technology, we develop an online adaptive update strategy to enhance the accuracy and responsiveness of model predictions. A data set from a zinc rotary kiln field site is collected for validation. The model demonstrates high accuracy in RKTT prediction, with mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R 2 ) values of 0.054, 0.150, and 0.9999, respectively. The results confirm the effectiveness of the proposed method for establishing an accurate rotary kiln process model, and for rapid and accurate model updates.}, archive = {J_EAAI}, author = {Xiaoming Wang and Xiaojun Liang and Chaobo Zhang and Chunhua Yang and Weihua Gui and Yiqi Liu}, doi = {10.1016/j.engappai.2024.109993}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109993}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards real-time adaptive prediction of rotary kiln processes: An enhanced framework combining parallel temporal convolution and long short-term memory networks}, volume = {144}, year = {2025}, }
• (2025). Adaptive temporal fusion network with depth supervision and modulation for robust three-dimensional object detection in complex scenes. EAAI, 144, 109988. (www)

  @article{ , abstract = {Autonomous driving perception relies on cameras and Light Detection and Ranging (LiDAR) sensors. Existing methods for LiDAR-camera fusion are primarily based on the Lift-Splat (LS) framework, which serves as a foundation for multi-modal fusion. However, these methods still face challenges such as unreliable depth information, insufficient dynamic perception, and limited robustness. This paper proposes a novel multi-modal three-dimensional (3D) detection method that optimizes depth by fully leveraging image and point cloud data and employs spatiotemporal deformable attention for adaptive fusion across frames. Specifically, we generate optimized depth maps through point clouds for depth supervision, refine the depth using Conditional Random Fields (CRF), and improve the fusion features by optimizing the depth estimation range. Additionally, we propose a dual-alignment method with spatiotemporal adaptive attention to acquire high-quality temporal features, allowing the model to learn beneficial information from adjacent frames. The proposed method achieves leading mean Average Precision (mAP) on mainstream 3D object detection datasets. Extensive experiments on multiple datasets demonstrate the superiority of the proposed method. Notably, our method remains effective even when a sensor fails, highlighting its potential to improve the robustness of autonomous perception in real-world scenarios.}, archive = {J_EAAI}, author = {Yifan Liu and Yong Zhang and Rukai Lan and Xiaopeng Cui and Linbo Xie and Zhaolong Wu}, doi = {10.1016/j.engappai.2024.109988}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109988}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive temporal fusion network with depth supervision and modulation for robust three-dimensional object detection in complex scenes}, volume = {144}, year = {2025}, }
• (2025). Quantifying algorithmic discrimination: A two-dimensional approach to fairness in artificial intelligence. EAAI, 144, 109979. (www)

  @article{ , abstract = {Fairness is a foundational pillar in the development of ethical and responsible artificial intelligence. One of the most pressing issues in this context is discrimination, which occurs when an algorithm displays unequal treatment towards data from different groups without an objective justification for such disparity. Artificial intelligence is increasingly taking on decision-making roles in society at large and in engineering fields in particular. In domains such as autonomous vehicle control systems, where unbiased decision-making can impact safety and trust, and in smart grid management, where equitable energy distribution is crucial, fairness must be a primary consideration. This study introduces a novel metric to measure fairness, consisting of a two-dimensional vector: Equality and Equity. When applied to benchmark datasets, this metric demonstrated superior informativeness by effectively distinguishing equality-related issues from equity-related challenges, surpassing traditional methods like Disparate Impact. Contributions of this work include (1) a pioneering metric for measuring equity; (2) a pure measure of fairness definition that takes into account equity and equality; (3) a vector to guide the mitigation algorithms; and, (4) a Fairness curve where the disparities between groups can be interpreted and explained.}, archive = {J_EAAI}, author = {Rubén González-Sendino and Emilio Serrano and Javier Bajo}, doi = {10.1016/j.engappai.2024.109979}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109979}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Quantifying algorithmic discrimination: A two-dimensional approach to fairness in artificial intelligence}, volume = {144}, year = {2025}, }
• (2025). Review of state-of-the-art surface defect detection on wind turbine blades through aerial imagery: Challenges and recommendations. EAAI, 144, 109970. (www)

  @article{ , abstract = {Maintenance is a critical aspect of wind power generation as it not only ensures the efficient operation of wind turbines, but also their continuous availability and functionality. For wind turbine blades, regular maintenance is essential to optimise power output and minimise operational downtime. While various maintenance strategies are well-documented, such as predictive approaches using Machine Learning and traditional visual inspections, there is limited research on leveraging aerial imagery for detecting defects on turbine blades. The objective of this review paper is to address this by focusing on the challenges and requirements for effective surface defect detection in wind turbine blades through aerial imagery. The task of inspecting surface defects on wind turbine blades is particularly difficult due to data scarcity, substantial computational requirements, and the geometric difficulties in accurately localising defects. By addressing these issues, we aim to identify and propose promising future directions to overcome these challenges at hand, thereby ensuring a progression of research and development in this field.}, archive = {J_EAAI}, author = {Imad Gohar and Weng Kean Yew and Abderrahim Halimi and John See}, doi = {10.1016/j.engappai.2024.109970}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109970}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Review of state-of-the-art surface defect detection on wind turbine blades through aerial imagery: Challenges and recommendations}, volume = {144}, year = {2025}, }
• (2025). Stability analysis of multi-criteria decision making techniques for relief centre location selection: A safety-centric relocation model with cost considerations. EAAI, 144, 109932. (www)

  @article{ , abstract = {The persistent threat of large-scale disasters necessitates effective emergency management strategies to mitigate their extensive physical and psychological impacts. This research introduces a safety-centric relocation model that addresses the safest relocation of affected populations within disaster relief operations. The model incorporates specific cost considerations, such as emergency services, route analysis, transit points, en-route assistance, and establishing secure healthcare facilities to enhance the safety and efficiency of relocation processes. In the aftermath of a disaster, establishing a suitable location for a relief centre is paramount for effective disaster response and recovery efforts. The study compares and implements six distinct multi-criteria decision-making (MCDM) techniques with five distinct weight methods to optimize the selection of relief centre sites utilizing ten criteria. Stability analysis identifies combined compromise solution (COCOSO) λ = 0 . 5 as the most reliable MCDM technique. The study further highlights the importance of implementing these MCDM techniques for society’s benefit. The unpredictability and vagueness of input data in post-disaster scenarios are addressed using single-valued trapezoidal neutrosophic numbers. The research employs optimization techniques using the linear interactive and general optimizer optimizing (LINGO) solver to obtain the optimal solution for the proposed multi-objective solid transportation model. The minimum value of the cost objective function is obtained using goal programming while the minimum value for the time objective function is obtained using fuzzy goal programming. The global criterion method provides the intermediate cost and time objective function results. The model’s application is demonstrated through a case study on the 2021 Uttarakhand flood.}, archive = {J_EAAI}, author = {Mayank Singh Bhakuni and Pooja Bhakuni and Amrit Das}, doi = {10.1016/j.engappai.2024.109932}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109932}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Stability analysis of multi-criteria decision making techniques for relief centre location selection: A safety-centric relocation model with cost considerations}, volume = {144}, year = {2025}, }
• (2025). Temporal-constrained parallel graph neural networks for recognizing motion patterns and gait phases in class-imbalanced scenarios. EAAI, 143, 110106. (www)

  @article{ , abstract = {Exoskeletons have emerged as a promising technology in the field of motor rehabilitation, particularly for individuals with lower limb motor dysfunction. Related research on human motion intention recognition based on wearable sensors and deep learning is garnering increasing attention from scholars. This study proposes a novel two-stage framework that first performs Motion Pattern Recognition (MPR) and then achieves Gait Phase Recognition (GPR) using surface electromyography and kinematic data from subjects' lower limbs. In view of the short-range dependency of convolutional neural networks, this study exploits Graph Neural Networks (GNNs) to adaptively fuse multimodal signals, and designs several graph generation methods to tackle the issue of lacking original topology in multimodal signals. On this basis, a Parallel GNNs (PGNNs) architecture is constructed to fuse graph-level and domain knowledge features, and a dynamically decaying weighted cross-entropy loss is presented to enhance the recognition performance of PGNNs in class-imbalanced scenarios. Additionally, a categorical hidden Markov model is established using the prior temporal constraints among subject's gait phases to reduce the “phase mutation” errors in PGNNs' predictions without additional training costs. The effectiveness of the proposed framework is verified on a benchmark dataset containing five common lower limb motions of ten healthy subjects. It achieves an average accuracy of 99.97% in the MPR task, and average GPR accuracies of 94.06%, 95.27%, and 95.43% in level walking, upstairs and downstairs motions, respectively, outperforming existing state-of-the-art methods. Experimental results indicate potential support for the controller design of human-robot highly-coupled rehabilitation exoskeletons.}, archive = {J_EAAI}, author = {Changhe Zhang and Zidong Yu and Xiaoyun Wang and Ze-Jian Chen and Chao Deng}, doi = {10.1016/j.engappai.2025.110106}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110106}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Temporal-constrained parallel graph neural networks for recognizing motion patterns and gait phases in class-imbalanced scenarios}, volume = {143}, year = {2025}, }
• (2025). A critical review of safe reinforcement learning strategies in power and energy systems. EAAI, 143, 110091. (www)

  @article{ , abstract = {The high penetration of distributed energy resources (DERs) in modern smart power systems introduces unforeseen uncertainties for the electricity sector, leading to increased complexity and difficulty in the operation and control of power systems. As a cutting-edge machine learning technology, deep reinforcement learning (DRL) has been widely implemented in recent years to handle the uncertainty in power systems. However, in critical infrastructures such as power systems, safety issues always receive top priority, while DRL may not always meet the safety requirements of power system operators. The concept of safe reinforcement learning (safe RL) is emerging as a potential solution to overcome the shortcomings of conventional DRL in the operation and control of power systems. This study provides a rigorous review of the latest research efforts focused on safe RL to derive power system control policies while accounting for the unique safety requirements of power grids. Furthermore, this study highlights various safe RL algorithms applied in diverse applications within the power and energy system sector, from single grid-connected power converters, residential smart homes, and buildings to large power distribution networks. For all methods outlined, a discussion on their bottlenecks, research challenges, and potential opportunities in the operation and control of power system applications is also presented. This review aims to support research in the area of safe RL algorithms, embracing smart power system operation with safety constraints amid high uncertainty from DERs.}, archive = {J_EAAI}, author = {Van-Hai Bui and Sina Mohammadi and Srijita Das and Akhtar Hussain and Guilherme Vieira Hollweg and Wencong Su}, doi = {10.1016/j.engappai.2025.110091}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110091}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A critical review of safe reinforcement learning strategies in power and energy systems}, volume = {143}, year = {2025}, }
• (2025). A new approach for failure mode and effect analysis based on fermatean fuzzy Z-number weighted muirhead mean operator. EAAI, 143, 110080. (www)

  @article{ , abstract = {Failure Mode and Effects Analysis (FMEA), as a risk assessment tool for identifying types of failures, their causes and consequences in product design or manufacturing processes, is widely used in various areas to improve the reliability and safety of systems. However, some obvious shortcomings in the traditional FMEA methodology have been exposed in practical applications, such as the neglect of ambiguity and uncertainty in the assessment information, the problem of the same risk priority number (RPN), the lack of relative weights among risk factors, and the neglect of correlation among risk factors. In order to improve these shortcomings, this paper defines the Fermatean fuzzy Z-number (FFZ) by integrating the Fermatean fuzzy set (FFS) and Z-number to express the evaluation information more completely. Meanwhile, a new Fermatean fuzzy Z-number weighted Muirhead mean (FFZWMM) operator is obtained by introducing the Muirhead mean (MM) operator, which is used to aggregate the evaluation information. On this basis, a new FMEA framework based on FFZWMM operator is proposed. Finally, the validity and superiority of the method are verified by applying the method to two numerical cases and conducting a comparative analysis.}, archive = {J_EAAI}, author = {Chuanhai Chen and Ruiliang Zhang and Jinyan Guo and Zhifeng Liu and Chunlei Hua and Haoming Yan and Baobao Qi and Tongtong Jin}, doi = {10.1016/j.engappai.2025.110080}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110080}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new approach for failure mode and effect analysis based on fermatean fuzzy Z-number weighted muirhead mean operator}, volume = {143}, year = {2025}, }
• (2025). Privacy-preserving federated learning compatible with robust aggregators. EAAI, 143, 110078. (www)

  @article{ , abstract = {Federated learning has emerged as a promising paradigm for collaborative machine learning across decentralized devices, offering the benefit of model training without centralized data aggregation. However, it also brings unique challenges related to security and privacy. Ensuring security and privacy is a critical challenge in federated learning. On one hand, sharing gradients or parameters of user models poses privacy risks; on the other hand, malicious users sending incorrect parameters can disrupt the global model on the server. Robust aggregators play a crucial role in handling such malicious users by scrutinizing transmitted vectors for anomalies. However, privacy-preserving methods often add noise or encryption to user vectors, which potentially affects the effectiveness of robust aggregators. In this paper, we have proposed a privacy-preserving method that maintains compatibility with robust aggregators by not directly altering the vectors of user parameters. Instead, noise is introduced to the training data or during model training. Our findings demonstrate that the proposed method not only reduces the accuracy of the membership inference attack to random levels but also maintains compatibility with robust aggregators without compromising the model's accuracy, thereby maintaining a balance between privacy, security, and efficiency.}, archive = {J_EAAI}, author = {Zeinab Alebouyeh and Amir Jalaly Bidgoly}, doi = {10.1016/j.engappai.2025.110078}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110078}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Privacy-preserving federated learning compatible with robust aggregators}, volume = {143}, year = {2025}, }
• (2025). Physics-informed deep learning framework for explainable remaining useful life prediction. EAAI, 143, 110072. (www)

  @article{ , abstract = {This study proposes a comprehensive framework of explainable remaining useful life (RUL) prediction. The proposed framework aims to not only enhance the accuracy and robustness of RUL estimation but also improve the interpretability of the estimated RUL. The proposed framework features three characteristics. First, raw measurements are transformed into informative features by accounting for the physics of degradation. This transformation results in five types of physics-informed features. The proposed method attenuates the adverse effect of distinct operational conditions while disclosing hidden information in the raw measurements. Second, a novel architecture of multi-scale deep convolutional neural network is addressed to extract temporal patterns from disparate time scales of transformed features, enabling accurate but robust prediction of RUL. Third, a novel interpretation method traces the root causes of the degradation. Specifically, layer-wise relevance propagation is deployed to create a relevance map, which provides fast but accurate identification of the failure modes regardless of the number of features. A systematic analysis on experiments with the commercial modular aero-propulsion system simulation dataset demonstrates the accuracy and interpretability of the proposed framework. The comparative study also confirms that the proposed framework would be effective in predicting and interpreting the RUL of systems in real-world applications, where disparate operational conditions and failure modes are complexly intertwined.}, archive = {J_EAAI}, author = {Minjae Kim and Sihyun Yoo and Seho Son and Sung Yong Chang and Ki-Yong Oh}, doi = {10.1016/j.engappai.2025.110072}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110072}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed deep learning framework for explainable remaining useful life prediction}, volume = {143}, year = {2025}, }
• (2025). A novel two-stage variables contribution analysis method toward explainable graph convolutional network-based industrial fault diagnosis. EAAI, 143, 110071. (www)

  @article{ , abstract = {In industrial systems, fault diagnosis methods play a crucial role for ensuring their safe and stable operations. Among these methods, graph convolutional network (GCN)-based fault diagnosis stands out for its ability to capture complex relationships within data. However, conventional GCN-based methods regard fault diagnosis solely as a classification problem, terminating the diagnostic process once classification decisions are made. This approach overlooks the analysis of the contributions of various relevant variables to the diagnostic results, leading to a lack of explainability and reliability, which hinders their industrial adoption. To address this challenge, the paper proposes a two-stage variables contribution analysis method called graph self-reproductive non-dominated sorting genetic algorithm (Graph-SRGA). The method sequentially conducts neighbor-stage and sensor-stage contribution analysis for the explainability of GCN-based methods. Specifically, two objective functions are formulated as evaluation indicators, and the selection process is encoded into a coding string. The proposed SRGA algorithm, integrating a self-reproductive strategy into the conventional non-dominated sorting genetic algorithm, identifies the required influential set. Experimental validation from various perspectives, using application cases in the high-speed train traction system and the Tennessee Eastman process, demonstrates that the proposed method offers a distinctive solution for the contribution analysis of GCN-based fault diagnosis methods.}, archive = {J_EAAI}, author = {Jiamin Xu and Siwen Mo and Zhiwen Chen and Haobin Ke and Zhaohui Jiang}, doi = {10.1016/j.engappai.2025.110071}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110071}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel two-stage variables contribution analysis method toward explainable graph convolutional network-based industrial fault diagnosis}, volume = {143}, year = {2025}, }
• (2025). Dynamic context-aware high-resolution network for semi-supervised semantic segmentation. EAAI, 143, 110068. (www)

  @article{ , abstract = {Real-time semi-supervised semantic segmentation provides a detailed understanding of dynamic urban situations for Intelligent transportation systems. However, state-of-the-art models following different segmentation tasks fail to incorporate High-Resolution Networks (HRNet) for real-time situations due to the inability to adjust in dynamic environments and high computational complexity. The study aims to examine the efficacy of student-teacher networks in semi-supervised semantic segmentation, mainly focusing on intelligent transportation. We present a Dynamic Context-aware High-Resolution Network (DC-HRNet) to enhance semantic segmentation in a dynamic urban environment incorporating a Student-Teacher knowledge distillation mechanism. We train teacher networks utilizing high-resolution networks to enhance robustness and precision in autonomous driving. Additionally, we employ Multi-path Blocks (MPBs) with HRNet to train our student network to address the challenges of the scarcity of labeled datasets in real-world scenarios. Utilization of MPBs for downsampling helps avoid low-resolution loss and generate pseudo labels by leveraging feature maps. We integrate a Dynamic Context-Aware Segmentation Network (DCSNet) to reduce the computational cost further and increase segmentation accuracy. Integrating DCSNet across various resolutions of HRNet with diverse dilation rates has enhanced contextual data utilization. This study contributes to intelligent transportation systems by improving segmentation accuracy, model generalization, and overall resilience, expanding the research scope in this area. Extensive investigations on the publicly available datasets show that our model can handle high-resolution images in real-time with state-of-the-art performance.}, archive = {J_EAAI}, author = {Khawaja Iftekhar Rashid and Chenhui Yang and Chenxi Huang}, doi = {10.1016/j.engappai.2025.110068}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110068}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic context-aware high-resolution network for semi-supervised semantic segmentation}, volume = {143}, year = {2025}, }
• (2025). A rapid detection and quantification method for levee leakage outlets using drone infrared thermography and semantic segmentation. EAAI, 143, 110066. (www)

  @article{ , abstract = {Leaking erosion is one of the most harmful factors causing levee breaches. To ensure the safety of levees, rapid leakage detection and evaluation is imperative. In current practice, manual operation—which is risky and inefficient—remains the most adopted method and lacks modern intervention. To bridge this gap, a novel fully automated method utilizing drones, infrared thermography, and deep learning is proposed. Drones equipped with infrared and visible imagers are utilized to perform fast sensing of distributed levees, and the sensing data is communicated to a server where pre-trained models are deployed for leakage exit recognition and quantification. To build such a service system, three different thermal imagers are employed to collect leakage images in a variety of leaking scenarios, including laboratory models and field actual levees. A total of 4686 raw infrared images containing diverse leakage objects are selected out and carefully annotated by hand at the pixel level. Afterwards, 14 models are trained using 5-fold cross-validation, taking into consideration different backbones and class imbalance. The highest performance model achieved a mean intersection over union value of 82.73% on the test set, despite the test set including many challenging samples. To assess the practicality of the proposed approach, a full-scale levee section is constructed. Several field tests demonstrated the effective of the proposed detection and quantification strategy. The well-trained detector demonstrated excellent performance in terms of practicability and generalization. The influencing factors and limitations of the proposed strategy are thoroughly examined.}, archive = {J_EAAI}, author = {Renlian Zhou and Monjee K. Almustafa and Zhiping Wen and Moncef L. Nehdi and Libing Zhang and Guang Yang and Huaizhi Su}, doi = {10.1016/j.engappai.2025.110066}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110066}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A rapid detection and quantification method for levee leakage outlets using drone infrared thermography and semantic segmentation}, volume = {143}, year = {2025}, }
• (2025). Research on fresh supply chain risk evaluation of community group purchase based on hybrid multi-attribute decision-making method. EAAI, 143, 110065. (www)

  @article{ , abstract = {The community group purchase (CGP) is rapidly evolving as a novel e-commerce mode that combines modern e-commerce with community integration in China. However, the core products of CGP are mainly fresh goods, where the fresh supply chain demands higher quality, faster processing, and greater freshness compared to the traditional product supply chain. Therefore, a risk evaluation of the CGP's fresh supply chain is crucial to ensure supplier stability, logistics efficiency, and storage conditions. This study presents a comprehensive risk evaluation model of CGP's fresh supply chain to enhance risk management and maintain product and freshness. Firstly, the social network analysis (SNA) method and the grounded theory are used to extract key indicators from existing literature and textual data. Secondly, the weight of each indicator is determined from the subjective and objective dimensions by using the order relationship analysis (G1) method and the Criteria Importance Through Intercriteria Correlation (CRITIC) method, and the decision matrix is constructed by the Fermatean hesitant fuzzy set. Then, considering that decision-maker (DM) may not be in a completely rational state during the evaluation process, the regret theory is introduced to quantify the DM's regret psychology. In addition, after obtaining the relative dominance through the TOmada de Decisão Iterativa Multicritério (TODIM) method, the risk evaluation is completed by using the VIseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR) method. Finally, sensitivity analysis validates the model's validity and applicability. This study offers a dynamic fresh supply chain risk evaluation model for the CGP industry, contributing to stable CGP enterprise operations.}, archive = {J_EAAI}, author = {Nian Zhang and Anling Wang and Yifan Zhou and Shuo Zheng}, doi = {10.1016/j.engappai.2025.110065}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110065}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Research on fresh supply chain risk evaluation of community group purchase based on hybrid multi-attribute decision-making method}, volume = {143}, year = {2025}, }
• (2025). Remaining useful life prediction of machinery using federated public feature representation in edge-cloud collaboration architecture. EAAI, 143, 110059. (www)

  @article{ , abstract = {Significant progress has been made in the prediction methods of the remaining useful life (RUL) of machinery. Nevertheless, two major challenges still exist for the large-scale practical application of these methods. Firstly, most prediction methods necessitate operation on high compute capability devices, which inevitably leads to a substantial increase in costs. Secondly, there are issues related to data silos and privacy protection. To address these challenges, we propose a new RUL prediction method by using federated public feature representation in edge-cloud collaboration (ECC) architecture. The training process of the proposed method contains two stages. In the first stage, the proposed Federated Feature Separation Autoencoder (FedFSAE) extracts shallow public feature representations from data located at different clients, with similar distributions. FedFSAE only requires training a shallow autoencoder, it does not require high compute capability on the clients. Furthermore, by aggregating only the public encoders, it achieves low communication costs and enhances privacy protection. In the second stage, clients upload the public features and corresponding labels to a trusted server with high compute capability, then, a global RUL prediction model, the proposed long-term fully convolutional network (LTFCN), is trained. This fully utilizes the high compute capability of the server to obtain a high-performance prediction model. Finally, the comparative analysis was conducted on milling cutter and bearing datasets to validate the effectiveness and superiority of the proposed method.}, archive = {J_EAAI}, author = {Liang Chen and Hongli Gao and Liang Guo and Junhua Liang and Lin Peng}, doi = {10.1016/j.engappai.2025.110059}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110059}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Remaining useful life prediction of machinery using federated public feature representation in edge-cloud collaboration architecture}, volume = {143}, year = {2025}, }
• (2025). NFIoT-GATE-DTL IDS: Genetic algorithm-tuned ensemble of deep transfer learning for NetFlow-based intrusion detection system for internet of things. EAAI, 143, 110046. (www)

  @article{ , abstract = {Industry 5.0 requires robust Internet of Things (IoT) networks, which are constantly vulnerable to cyber threats. Despite contributions in intrusion detection systems (IDS), creating generalized attack classification models remains a challenge. Conventional machine learning or deep learning-driven IDSs struggle to retain learned knowledge and to keep up with rapidly increasing IoT threats, whereas transfer learning-based models may lose the resilience required for model generalization. Thus, this study proposes a Genetic Algorithm-Tuned Ensemble of Deep Transfer Learning for NetFlow-Based Intrusion Detection System for Internet of Things (NFIoT-GATE-DTL IDS) to fill the gap. Two public NetFlow IoT datasets are preprocessed and transformed into three-dimensional images for convolutional neural networks (CNNs). Six pre-trained CNNs, including Xception, Inception, MobileNet, MobileNetV2, DenseNet121, and EfficientNetB0, undergo hyperparameter optimization using a Genetic Algorithm (GA). The top-five models are then combined using a soft voting ensemble to boost detection robustness across diverse attack types. Validation methods are employed, including assessing the impact of GA optimization, comparing it to optimizers like the covariance matrix adaptation evolution strategy and the coyote optimization algorithm, and comparing it to cutting-edge studies. The proposed framework consistently achieves 100% accuracy across 15 attack classes, including three highly minority threats like backdoors, ransomware, and theft in IoT networks. Furthermore, the NFIoT-GATE-DTL IDS outperforms recent methodologies, achieving a 5–7% multi-classification higher accuracy on average. This research significantly contributes to a robust IDS with a high detection rate for NetFlow-based IoT networks.}, archive = {J_EAAI}, author = {Jing Li and Hewan Chen and Mohd Shahizan Othman and Naomie Salim and Lizawati Mi Yusuf and Shamini Raja Kumaran}, doi = {10.1016/j.engappai.2025.110046}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110046}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {NFIoT-GATE-DTL IDS: Genetic algorithm-tuned ensemble of deep transfer learning for NetFlow-based intrusion detection system for internet of things}, volume = {143}, year = {2025}, }
• (2025). Topology-joint curvilinear segmentation network using confidence-based bezier topological representation. EAAI, 143, 110045. (www)

  @article{ , abstract = {Segmenting continuous curve structures is crucial in medical lesion analysis, crack detection, and remote sensing image processing. Current semantic segmentation methods often neglect pixel-wise topological relationships, lack topological feature extraction, and lose fine curve details during upsampling, resulting in potential discontinuities. The discontinuity of the curve significantly reduces its analysis and application. Thus, this paper presents the Topology-joint Curvilinear Segmentation Network (TCSN), comprising Confidence-based Bezier Topological Representation (CBTR), Topology Enhanced Block (TEB), and Topology Supplement Block (TSB) to extract more continuous curve structure. CBTR establishes differentiable multiscale bezier representations, defining a topological loss function to guide network learning. TEB utilizes dilated convolutions to map features to bezier space, introducing topological encoding to enhance topology perception. TSB introduces continuous skeleton topological ground truth, using residual connections for topological repair and promoting segmentation of more continuous curves. Experiments on three datasets from medical analysis, crack detection, and remote sensing image processing domains yield topological segmentation accuracy(Quality) of 35.21%, 23.53%, and 59.19%, respectively. These results surpass the recent state-of-the-art methods and demonstrate that TCSN possesses excellent topological preservation and cross-field application capabilities. To the best of our knowledge, this is the first study of introducing bezier into curve structure segmentation, demonstrating potential applications for other domains of curvilinear structures.}, archive = {J_EAAI}, author = {Jianwei Li and Yuchun Huang and Xi Ye and He Yang}, doi = {10.1016/j.engappai.2025.110045}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110045}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Topology-joint curvilinear segmentation network using confidence-based bezier topological representation}, volume = {143}, year = {2025}, }
• (2025). Optimized data-driven method to study the self-healing and durability of ultra-high performance concrete. EAAI, 143, 110043. (www)

  @article{ , abstract = {Machine learning frameworks offer significant potential for predicting the self-healing properties of Ultra-High-Performance Concrete (UHPC) and can also serve in the design of advanced durability-based concrete. Existing machine learning models, while exhibiting good generalization, are limited in prediction accuracy due to training on small datasets with outliers. Moreover, many studies lack comprehensive interpretive analyses, resulting in inadequate extraction of physical insights related to self-healing. This paper introduces an advanced data-driven framework to predict crack closure and chloride diffusion in cracked concrete. The framework has three main components: (1) The Tree-structured Parzen Estimator and 10-fold cross-validation are employed to optimize the hyperparameters of the model. (2) Isolation Forest and forward stepwise selection are used for data cleaning, identifying and removing outliers and irrelevant variables in the dataset. (3) Feature importance analysis, SHapley Additive exPlanations (SHAP), and Partial Dependence Plots (PDP) are utilized to explain and quantify the impact of features on model composition and prediction. This research framework utilizing Light Gradient Boosting Machine (LightGBM) as the base model has been established, characterized by high accuracy and strong generalization in predicting concrete self-healing and durability. Using interpretive tools, the crack width thresholds for healing and chloride transport has been quantified at 60–125 μm and 35–60 μm, respectively.}, archive = {J_EAAI}, author = {Zhewen Huang and Estefanía Cuenca and Liberato Ferrara}, doi = {10.1016/j.engappai.2025.110043}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110043}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimized data-driven method to study the self-healing and durability of ultra-high performance concrete}, volume = {143}, year = {2025}, }
• (2025). Towards open-world recognition: Critical problems and challenges. EAAI, 143, 110042. (www)

  @article{ , abstract = {With the emergence of rich classification models and high computing power, recognition systems are widely used in various fields. Unfortunately, as the scale of open systems increases, the assumption of a closed-world will lead to the failure of fragile models because almost all machine learning based recognition algorithms are evaluated as implicit “closed-set”. Comparing with classical methods, open-world learning can address concerns in dynamic environments where the input data (size, category, etc.) is changing rapidly. Nevertheless, there still lacks a thorough review of recent advances in open-world recognition. Therefore, we provide an in-depth discussion of open-world recognition based on some recent works. First, we propose a learning framework for open-world recognition, and analyze the challenges from three aspects: domain shift,limits on the amount of labeled data and perception scene with dynamic changes. Secondly, we evaluate the current state of the art, summarize the intersection of various methods and find out some existing problems. Finally, we discuss the limitations of current procedures and new technologies as well as future directions in order to make meaningful progress. This article will help researchers understand Open-world learning and the possibilities of extending research into appropriate areas.}, archive = {J_EAAI}, author = {Ke Wang and Zhikang Li and Yang Chen and Wenjie Dong and Junlan Chen}, doi = {10.1016/j.engappai.2025.110042}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110042}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards open-world recognition: Critical problems and challenges}, volume = {143}, year = {2025}, }
• (2025). Unsupervised multi-view feature selection based on weighted low-rank tensor learning and its application in multi-omics datasets. EAAI, 143, 110041. (www)

  @article{ , abstract = {With the explosive growth of unlabeled multi-view data with high dimensionality in various fields such as bioinformatics, unsupervised multi-view feature selection has become a key technique to simultaneously handle the curse of dimensionality in multiple datasets. Most existing methods heavily rely on the pseudo labels obtained from respective views, where they perform feature selections without comprehensively exploiting the high-order connections among views. In this paper, we propose a weighted low-rank tensor-based unsupervised multi-view feature selection framework (WLTL), which integrates multi-view spectral clustering and weighted low-rank tensor to generate high-quality pseudo labels for feature selection. Specifically, we stack the clustering indicator matrices into a three-dimensional tensor and impose the weighted tensor nuclear norm constraint, which captures high-order correlations and consistent pseudo label information while allowing for more refined consideration of the importance of each view. Our method also features an adaptive strategy to automatically assign view weights, optimizing the feature selection process by considering both individual view characteristics and inter-view relationships. We present an efficient optimization algorithm for iteratively refining the proposed framework. Extensive experiments implemented on six machine learning and three multi-omics datasets confirm the superiority of WLTL. Additionally, the case study performed on the cancer dataset serves to affirm the practicality of our model.}, archive = {J_EAAI}, author = {Daoyuan Wang and Lianzhi Wang and Wenlan Chen and Hong Wang and Cheng Liang}, doi = {10.1016/j.engappai.2025.110041}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110041}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised multi-view feature selection based on weighted low-rank tensor learning and its application in multi-omics datasets}, volume = {143}, year = {2025}, }
• (2025). Supervised learning based method for condition monitoring of overhead line insulators using leakage current measurement. EAAI, 143, 110040. (www)

  @article{ , abstract = {As a new practical and economical solution to the aging problem of overhead line (OHL) assets, the technical policies of most power grid companies in the world experienced a gradual transition from scheduled preventive maintenance to a risk-based approach in asset management. Although the accumulation of contamination is predictable to a certain degree, and various methods have been proposed to assess the risk of insulator flashover, it remains significant potential to improve these methods and to develop more reliable and efficient strategies for insulator maintenance planning. This paper presents a novel machine learning (ML) based method for estimating the flashover probability of the cup-and-pin glass insulator string. Utilizing the Extreme Gradient Boosting (XGBoost) supervised ML model, the method incorporates the measurement results of leakage current (LC) features and applied voltage as the input. The proposed method can be divided into two stages. The first stage contains a classification model which achieves over 98% F1-score accuracy in estimating humidity level as a binary representation (dry or wet). In the second stage, the regression model predicts the critical flashover voltage ( U 50 % ) for various OHL insulator designs across different voltage levels. Results indicate that by selecting the optimal input features, the error of the flashover prediction can be minimized to around 1.16% according to the root mean square error (RMSE) metric. Furthermore, the proposed method is able to accurately determine the dielectric strength of the insulator strings and instruct asset management engineers to take appropriate actions.}, archive = {J_EAAI}, author = {Mile Mitrovic and Dmitry Titov and Klim Volkhov and Irina Lukicheva and Andrey Kudryavzev and Petr Vorobev and Qi Li and Vladimir Terzija}, doi = {10.1016/j.engappai.2025.110040}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110040}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Supervised learning based method for condition monitoring of overhead line insulators using leakage current measurement}, volume = {143}, year = {2025}, }
• (2025). Shoreline change prediction along the cijin coastline of taiwan using deep learning and satellite imagery. EAAI, 143, 110039. (www)

  @article{ , abstract = {Coastal zones are vital areas that support industries, the environment, and human activities. However, global warming has exacerbated coastal erosion due to rising sea levels and stronger wave-current dynamics, presenting significant challenges for many countries. Accurately predicting shoreline changes is essential for effective coastal management, sustainable aquaculture, and infrastructure development. This study aims to develop a method for forecasting shoreline changes along the Cijin coastline in Taiwan using deep learning techniques applied to satellite imagery. The approach combines Canny edge detection with a Bidirectional Long Short-Term Memory (Bi-LSTM) network for time-series prediction. The model achieved high accuracy, with a Mean Absolute Percentage Error (MAPE) of 0.6% and a Root Mean Square Error (RMSE) of 2.28. Additionally, the network demonstrated fast performance, making predictions in an average time of 8.411 s. These results highlight the effectiveness of this approach in providing valuable insights into future shoreline changes, which can help governmental organizations assess coastal protection strategies and make informed decisions for sustainable coastal management.}, archive = {J_EAAI}, author = {Li-Hung Tsai and Chih-Hsien Wu and Qin-Sen Zhang and Jen-Chung Shao and Chih-Min Hsieh and Cheng-Hong Yang}, doi = {10.1016/j.engappai.2025.110039}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110039}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Shoreline change prediction along the cijin coastline of taiwan using deep learning and satellite imagery}, volume = {143}, year = {2025}, }
• (2025). Hierarchical intention recognition framework in intelligent human‒computer interactions for helicopter and drone collaborative wildfire rescue missions. EAAI, 143, 110037. (www)

  @article{ , abstract = {Intention recognition is the prerequisite and foundation for achieving intelligent human‒computer interaction (IHCI). In complex task scenarios, human intentions often exhibit hierarchical and context-related characteristics. In this paper, we propose a hierarchical intention recognition framework suitable for complex tasks, aiming to infer multilevel intentions from the lowest-level data such as human behaviour, machine and environmental information. The framework's usability was validated via a case study of helicopter and drone collaborative wildfire rescue missions. First, task analysis was employed to extract the intention structure among high-level task intentions, low-level interaction intentions, and feature data. Flight simulation data from 80 complete trials conducted by 16 participants were subsequently collected. Finally, reliable recognition of interaction intentions and task intentions was achieved via a one-dimensional convolutional neural network (1D-CNN) + bidirectional long short-term memory (Bi-LSTM) neural networks and a dynamic Bayesian network (DBN). The highest recognition accuracies for interaction intentions and task intentions were 94.57% and 97.33%, respectively. The methods and results of this study provide a novel perspective for human intention recognition in complex tasks and offer methodological guidance and support for intelligent adaptive interfaces and IHCI in complex task scenarios.}, archive = {J_EAAI}, author = {Ruisheng Zhang and Xuyi Qiu and Jichen Han and Hang Wu and Minglang Li and Xiaozhou Zhou}, doi = {10.1016/j.engappai.2025.110037}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110037}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hierarchical intention recognition framework in intelligent human‒computer interactions for helicopter and drone collaborative wildfire rescue missions}, volume = {143}, year = {2025}, }
• (2025). Optimization of centrifugal pump performance and excitation force based on machine learning and enhanced non-dominated sorting genetic algorithm III. EAAI, 143, 110036. (www)

  @article{ , abstract = {This study addresses the issue of excessive vibration in centrifugal pumps without compromising their performance by employing six machine learning methods. These include the Backpropagation (BP) Neural Network, Feedforward Neural Network (FNN), Support Vector Machine (SVM), Extreme Learning Machine (ELM), Gaussian Process Regression (GPR), and Tree Ensemble (TE). These methods are used to construct a mapping model that correlates optimization objectives and parameters. The model served as the fitness function in the Adaptive Non-dominated Sorting Genetic Algorithm III (A-NSGA-III). A total of 1000 optimization results generated by the A-NSGA-III were evaluated to determine the final optimization scheme, using a reverse design approach to identify optimal parameters.A comparison of the hydraulic performance before and after optimization revealed minor improvements in the head, efficiency, and shaft power. The radial force decreased by 17.81%, and the axial force decreased by a substantial 98.12%, meeting the set objectives. The significant reduction in axial force was predominantly observed at the balance chamber. Entropy production theory analysis indicated a considerable decrease in the entropy production rate on the impeller's inner wall.Cross-wavelet transform and wavelet coherence analysis showed a significant correlation between the entropy production rate and the energy variation of the axial force. Before optimization, there was a pronounced coherence between the entropy production rate and the resonance period, which was notably reduced after optimization. This method significantly addresses the issue of excessive axial force in the operation of centrifugal pumps without compromising the basic hydraulic performance. This study provides a reference for the structural optimization of centrifugal pumps.}, archive = {J_EAAI}, author = {Jiang Haoqing and Dong Wei and Li Sucheng and He Fan}, doi = {10.1016/j.engappai.2025.110036}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110036}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimization of centrifugal pump performance and excitation force based on machine learning and enhanced non-dominated sorting genetic algorithm III}, volume = {143}, year = {2025}, }
• (2025). Enhancing rate of penetration prediction in drilling operations: A data stream framework approach. EAAI, 143, 110034. (www)

  @article{ , abstract = {In geoenergy and mining industries, drilling operations serve as vital processes for accessing underground resources. Among the myriad factors shaping drilling success, the Rate of Penetration (ROP) through subsurface rock formations emerges as a pivotal parameter. Precise prediction and optimization of ROP assume heightened significance, given their profound impact on cost efficiency, safety protocols, and the overarching aim of minimizing environmental footprints. Machine Learning (ML) algorithms offer enhanced flexibility in model formulation to optimize drilling operations. Yet, existing literature on ROP prediction overlooks the dynamic nature of ROP data, treating real-time data as static during both collection and analysis. We advocate that ROP data presents changes in its underlying distribution over time, primarily due to the diverse lithology properties. These changes can impede the effectiveness of ML algorithms, a phenomenon referred to as concept drift. Therefore, in this paper, we propose addressing ROP prediction under the data stream framework, that accounts for high-speed data, frequent model updating, and effective handling of concept drift. To validate our hypothesis, we compare three data stream algorithms—Adaptive Random Forest (ARF) with the concept drift detectors Adaptive Windowing (AW), (ARF AW ) and Page–Hinkley (PH),(ARF PH ); and Fast Incremental Model Trees with Drift Detection (FIMTDD)—with three traditional ML algorithms: Random Forest (RF), AdaBoost, and Gradient Tree Boosting (GTB). This comparison utilizes seven datasets obtained from actual drilling operations. We observed a 63% decrease in absolute error rate and a substantial 77% reduction in processing time, on average, when employing appropriate data stream algorithms compared to traditional ML approaches.}, archive = {J_EAAI}, author = {João Roberto Bertini Junior and Bahram Lavi}, doi = {10.1016/j.engappai.2025.110034}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110034}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing rate of penetration prediction in drilling operations: A data stream framework approach}, volume = {143}, year = {2025}, }
• (2025). UitPDF-MalDe: Malicious portable document format files detection using multi machine learning models. EAAI, 143, 110031. (www)

  @article{ , abstract = {The widespread use of Portable Document Format (PDF) files has created an ideal environment for the proliferation of malicious PDFs, posing a significant threat to information security. In response to this growing concern, this study conducts a comprehensive survey of current network threat trends and existing methods for detecting and classifying malicious PDFs. A new dataset in CSV (Comma-Separated Values) format is developed from the CIC-Evasive-PDFMal2022 dataset, featuring refined attributes and an additional feature, Uniform Resource Identifier (URI), which significantly improves model classification performance. Among the six popular machine learning (ML) models evaluated, Logistic Regression (LR), Support Vector Machines (SVM), Decision Trees (DT), Random Forests (RF), Naive Bayes (NB), and K Nearest Neighbor (KNN), the RF model achieves the highest accuracy at 99.78%, outperforming related studies, with Precision of 99.66%, Recall of 99.81%, F1-Score of 99.73%, and Area Under the Curve (AUC) of 99.99%. Furthermore, this study proposes a feature selection method that enhances the accuracy of four out of six models, with a notable improvement in the NB model’s accuracy from 45.94% to 95.80% with Precision of 93.96%, Recall of 96.35%, F1-Score of 95.03%, and AUC of 97.10%. The approach’s robustness is further validated using the Contagio dataset, consistently achieving high performance. Various evaluation metrics, including Accuracy, Recall, Precision, F1-Macro Average, ROC-AUC (Receiver Operating Characteristic - Area Under the Curve), and Confusion matrix, are utilized to assess model performance comprehensively, while a correlation coefficient matrix ensures the selected features contribute effectively to model performance without redundancy.}, archive = {J_EAAI}, author = {Nguyen Tan Cam and Tran Quang Hung and Pham Tien Nam}, doi = {10.1016/j.engappai.2025.110031}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110031}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {UitPDF-MalDe: Malicious portable document format files detection using multi machine learning models}, volume = {143}, year = {2025}, }
• (2025). Prediction and rationality analysis of new energy vehicle sales in china with a novel intelligent buffer operator. EAAI, 143, 110030. (www)

  @article{ , abstract = {Despite the industry's booming development, there is a concern that new energy vehicles (NEVs) overcapacity and resource and environmental constraints pose a risk to Chinese new energy vehicle industry. Accordingly, it is essential to achieve accurate prediction of future sales to promote the healthy and expeditious growth of Chinese new energy vehicle industry. On the one hand, starting from the data-oriented perspective, a novel intelligent buffer operator (which is an important data processing method in the field of Artificial Intelligence) that integrates the principle of prioritizing new information and introduces a power exponent is proposed, which addresses the issue of systematic prediction traps resulting from perturbations in the data sets of new energy vehicle sales in China. Then a whitenization grey prediction model has been formulated by optimizing parameters combination, which has high compatibility, adaptability, and modelling capability. The new grey prediction model utilizing an intelligent buffer operator has been employed to simulate the data of sales for Chinese new energy vehicles. The new grey prediction model based on intelligent buffer operator is applied to model new energy vehicle sales, and its comprehensive error is better than that of the models based on the weighted geometric average buffer operator and the full information buffer operator, providing a better description of the data characteristics and development trend of new energy vehicle sales in China. The latest forecast indicates that the sales of new energy vehicles in China will continue to increase from 2023 to 2025, surpassing 10 million units in 2024.}, archive = {J_EAAI}, author = {Bo Zeng and Fengfeng Yin and Jianzhou Wang and Wei Meng}, doi = {10.1016/j.engappai.2025.110030}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110030}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction and rationality analysis of new energy vehicle sales in china with a novel intelligent buffer operator}, volume = {143}, year = {2025}, }
• (2025). Examining the application of strategic management and artificial intelligence, with a focus on artificial neural network modeling to enhance human resource optimization with advertising and brand campaigns. EAAI, 143, 110029. (www)

  @article{ , abstract = {This study employed artificial intelligence with machine learning method and used an artificial neural network to estimate waste management, travel expenses, and paper consumption by manipulating varying water and energy costs, as well as the number of employees. The results show that increased costs and employee numbers have an impact on waste management and travel expenses, while paper consumption remained consistent. Research and analysis show that viral advertising outperforms traditional advertising in terms of brand awareness, attitudes, engagement, and purchase intent. Linear regression analysis confirmed the accuracy of the network's predictions. Recognizing that every organization is part of a larger system, cooperation within this system occurs when subsystems interact with environmental issues. Hence, it becomes possible not only to make organizations environmentally conscious but also to foster environmental consciousness in each employee, underlining the significance of individual and collective contributions to environmental sustainability. Within the context of social care, this study aims to explore the role of strategic management and artificial intelligence in optimizing human resources management to address environmental concerns. By examining the intersection of these fields, this study endeavors to provide insights into how organizations can leverage technology and strategic planning to cultivate sustainable human resource management practices in the realm of social care.}, archive = {J_EAAI}, author = {Cao Ruoxing and Wang Jianning and Ali Basem and Rasha Abed Hussein and Soheil Salahshour and S. Baghaei}, doi = {10.1016/j.engappai.2025.110029}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110029}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Examining the application of strategic management and artificial intelligence, with a focus on artificial neural network modeling to enhance human resource optimization with advertising and brand campaigns}, volume = {143}, year = {2025}, }
• (2025). Causality-guided fault diagnosis under visual interference in fused deposition modeling. EAAI, 143, 110027. (www)

  @article{ , abstract = {Fused deposition modeling (FDM) is one of the additive manufacturing (AM) technologies widely used in various industrial fields. Several factors can affect the manufacturing process, leading to quality issues in produced parts. Fault diagnosis techniques aiming at identifying the cause of quality degradation are critical to managing the FDM process. Image-based methods are a trending topic in fault diagnosis with non-contact sensors. Visual interference may cause an unreliable correlation between the background and the root cause labels. Herein, a causality-guided model is proposed for fault diagnosis in the FDM process, which has practical value for improving the accuracy of the fault diagnosis models and helps provide reliable recommendations to the operators. From a causal inference perspective, a structural causal model (SCM) is constructed to describe the "pixel-feature-label" relationship in fault diagnosis tasks. A causality-guided network is proposed to extract features causally related to the root cause labels by eliminating the background information. It comprises three parts: a variational autoencoder (VAE) for learning the background from raw images, a convolution-based fault learning network for extracting defective features, and a feature intervention unit for implementing intervention strategy in SCM. In the experiments, the proposed method achieved a diagnostic F1-score of 92.80 % in the classification of one normal and six abnormal scenarios. It outperformed state-of-art networks including non-causality and causality-guided ones. Furthermore, Grad-CAM visualizations reveal that the proposed model tends to focus on more compact defect areas, aiding in precise fault localization. This enhances both the interpretability and trustworthiness of the fault diagnosis results.}, archive = {J_EAAI}, author = {Qian Li and Tingting Huang and Jie Liu and Shanggang Wang}, doi = {10.1016/j.engappai.2025.110027}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110027}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Causality-guided fault diagnosis under visual interference in fused deposition modeling}, volume = {143}, year = {2025}, }
• (2025). Efficient intelligent quality detection of pistachios using multi-view deep learning. EAAI, 143, 110025. (www)

  @article{ , abstract = {The market value of pistachios is primarily determined by their external quality. In this study, we propose a novel method for the rapid detection of pistachio external quality. Multi-view imaging was employed to capture comprehensive external features, addressing the limitations of incomplete information in single-view systems. We utilized a convolutional neural network, specifically the Residual Network with 18 layers, as the backbone model for detecting pistachio quality. The model was pruned using first-order Taylor expansion to reduce the number of filters and thus the model's complexity. Under three-view imaging, the pruned network achieved an accuracy of 99.56%, with a per-image detection time of 0.0880 s. With two-view imaging, the accuracy was 99.39%, reducing detection time to 0.0667 s per image. The pruning process reduced the number of model parameters by 96.02% compared to the original network. Finally, the optimized models were integrated into two types of pistachio sorting machines for real-time detection of open and closed pistachios. This method delivers high detection accuracy, faster processing speeds, and significantly reduced model complexity.}, archive = {J_EAAI}, author = {Hongfei Zhu and Huayu Fu and Cong Wang and Zhenlu Hua and Xingyu Liu and Weiming Shi and Ziyan Zong and Yanshen Zhao and Fei Liu and Limiao Deng and Ranbing Yang and Zhongzhi Han}, doi = {10.1016/j.engappai.2025.110025}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110025}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient intelligent quality detection of pistachios using multi-view deep learning}, volume = {143}, year = {2025}, }
• (2025). Multi-modal dataset and fusion network for simultaneous semantic segmentation of on-road dynamic objects. EAAI, 143, 110024. (www)

  @article{ , abstract = {An accurate and robust perception system is essential for autonomous vehicles to interact with various dynamic objects on the road. By applying semantic segmentation techniques to the data from the camera sensor and light detection and ranging sensor, dynamic objects can be classified at pixel and point levels respectively. However, there are challenges when using a single sensor, especially under adverse lighting conditions or with sparse point densities. To address these challenges, this paper proposes a network for simultaneous point cloud and image semantic segmentation based on sensor fusion. The proposed network adopts a modal-specific architecture to fully leverage the characteristics of sensor data and achieves geometrically accurate matching through the image, point, and voxel feature fusion module. Additionally, we introduce the dataset that provides semantic labels for synchronized images and point clouds. Experimental results show that the proposed fusion approach outperforms uni-modal based methods and demonstrates robust performance even in challenging real-world scenarios. The dataset is publicly available at https://github.com/ailab-konkuk/Multi-Modal-Dataset .}, archive = {J_EAAI}, author = {Jieun Cho and Jinsu Ha and Hamin Song and Sungmoon Jang and Kichun Jo}, doi = {10.1016/j.engappai.2025.110024}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110024}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modal dataset and fusion network for simultaneous semantic segmentation of on-road dynamic objects}, volume = {143}, year = {2025}, }
• (2025). Integrated spectrogram construction method on multi-channel signals for loose particle localization. EAAI, 143, 110023. (www)

  @article{ , abstract = {Loose particle is an important factor affecting the reliable operation of sealed electronic equipment. Loose particle localization research provides important reference for loose particle cleaning work, but there are still some problems. In this paper, for the first, the authors introduced the “image + deep learning model” scheme into the loose particle localization research, considered conducting multi-channel signal or image recognition research from the source link, thus proposed an integrated spectrogram construction method on multi-channel signals, and conducted two parts of research on signal groups and pulse groups. In the first part, the signal group that provides feedback on overall state information is converted into four signal spectrograms, horizontal merging, vertical merging, cross merging and linear superposition are performed, and four signal-integrated spectrogram sets are created, multiple deep learning models are trained, and the optimal transformation method and appropriate deep learning model are determined. In the second part, the two-stage variable energy threshold pulse extraction algorithm and noise-filling fused threshold-judgment pulse matching algorithm are newly proposed to extract multiple pulse groups from the signal group. The pulse group that provides feedback on relevant state information is converted into four pulse spectrograms, and the same four transformation methods are performed to create four pulse-integrated spectrogram sets and train deep learning models, then the optimal transformation method and appropriate deep learning model are determined. Comprehensive analysis shows that, the signal-group-to-spectrogram construction method that using vertical merging is the optimal integrated spectrogram construction method, Resnet-50 is a well-performing deep learning model, achieving the highest average classification accuracy of 98.31% and the highest localization accuracy of 100%, which has significant advantages over existing loose particle localization research. A large number of experiments have demonstrated the necessity and correctness of conducting this study, as well as the feasibility, practicality, and superiority of the proposed method. This study is an important supplement to existing loose particle detection research, which can effectively improve the reliability of sealed electronic equipment in aerospace systems. In theory, this study can provide important references for multi-channel signal processing and recognition research in the field of acoustic emission detection and fault diagnosis.}, archive = {J_EAAI}, author = {Zhigang Sun and Guofu Zhai and Min Zhang and Guotao Wang and Qi Liang and Hao Chen}, doi = {10.1016/j.engappai.2025.110023}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110023}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrated spectrogram construction method on multi-channel signals for loose particle localization}, volume = {143}, year = {2025}, }
• (2025). A thermodynamics-informed deep learning approach for lightweight modeling of gas turbine performance. EAAI, 143, 110022. (www)

  @article{ , abstract = {Lightweight performance modeling approaches are particularly crucial in the condition monitoring system of a heavy-duty gas turbine (HDGT) to track its efficiency changes over time accurately and promptly. This ensures sufficient productivity through a well-planned predictive maintenance strategy. This paper introduces a physics-informed deep learning methodology for lightweight modeling of HDGT performance, aimed at a real-time degradation monitoring for predictive maintenance. Initially, a mechanism-based thermodynamic model is established to serve as a performance benchmark. Subsequently, corrections are applied by adjusting base-load operating conditions to reference conditions, thereby mitigating the influence of ambient conditions and power demand. A substitute model called deep operator networks (DeepONet) is then constructed by integrating actual and simulation data to rapidly obtain crucial gas turbine performance parameters, such as compressor and turbine efficiency, as well as corrected power output and heat rate of the system. A standardized procedure is developed to automate the efficient performance modeling for gas turbines. To showcase the benefits of the proposed methodology and procedure, a comparative study with three different classical models is conducted using data from two various real-world HDGT machines. The DeepONet deep learning model is utilized to rapidly generate multivariate performance results for a gas turbine at 10ms for a single-step prediction, significantly faster than the physics-based model, which takes 6s. The prediction error is less than 0.1% on average when compared to the latter. Numerical results demonstrate that the proposed methodology offers a promising tool for real-time performance prediction of a gas turbine for predictive maintenance purposes.}, archive = {J_EAAI}, author = {Xiaomo Jiang and Yiyang Liu and Manman Wei and Xueyu Cheng and Zhicheng Wang}, doi = {10.1016/j.engappai.2025.110022}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110022}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A thermodynamics-informed deep learning approach for lightweight modeling of gas turbine performance}, volume = {143}, year = {2025}, }
• (2025). Optimizing quay crane scheduling using deep reinforcement learning with hybrid metaheuristic algorithm. EAAI, 143, 110021. (www)

  @article{ , abstract = {As global trade via maritime transport increases annually, competition among seaports necessitates dynamic scheduling in operational management to optimize port performance. Quay crane scheduling problem (QCSP) is a typical optimization problem for container terminal operations on the quayside. This paper proposes a modern approach based on model-free deep reinforcement learning (DRL) named proximal policy optimization (PPO), a gradient-based framework with a shared actor-critic network structure. Besides, a novel hybrid metaheuristic combined with a greedy randomized adaptive search procedure (GRASP) and ant colony optimization (ACO) is employed as an alternative approach. The performance of the artificial intelligence (AI) -powered solutions is verified through numerical simulation for single- and multi-agent scenarios with different starting positions. The results highlight the intelligent approach's remarkable performance in solving complex optimization problems and its flexibility compared to traditional metaheuristics, as it does not require clear insights into the system's dynamics.}, archive = {J_EAAI}, author = {Le Ngoc Bao Long and Sam-Sang You and Truong Ngoc Cuong and Hwan-Seong Kim}, doi = {10.1016/j.engappai.2025.110021}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110021}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing quay crane scheduling using deep reinforcement learning with hybrid metaheuristic algorithm}, volume = {143}, year = {2025}, }
• (2025). Unsupervised surface defect detection using dictionary-based sparse representation. EAAI, 143, 110020. (www)

  @article{ , abstract = {This study proposes a novel sparse encoding detection method that utilizes pre-trained deep neural networks to extract fused multi-scale features, addressing critical challenges in automated surface defect detection in manufacturing. By building sparse dictionaries with minimal redundancy, our technique overcomes issues related to poor feature representation and uneven features, requiring only defect-free samples for the model training. The pre-trained neural network extracts high-dimensional features uniformly, aligning and fusing the intermediate features across different network scales to enhance the image semantic information. We construct a representative sparse subset from the feature pool derived from defect-free samples using minimal redundancy as a constraint. The final step involves querying the sparse dictionary to calculate the image-level anomaly score and region. Extensive experiments on the MVTec Anomaly Detection (MVTec AD) dataset and the corrugated pipe dataset validate the efficacy of our approach. Specifically, we achieve a pixel-level area under the receiver operating characteristic (AUROC) score of 98.3% on the corrugated pipe dataset, demonstrating our method's superior performance compared to existing state-of-the-art defect detection and segmentation techniques. Our research contributes significantly to advancements in Artificial Intelligence for manufacturing applications by enhancing the reliability and accuracy of surface defect detection.}, archive = {J_EAAI}, author = {Meng Fanwu and Gong Tao and Wu Di and Xiang Xiangyi}, doi = {10.1016/j.engappai.2025.110020}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110020}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised surface defect detection using dictionary-based sparse representation}, volume = {143}, year = {2025}, }
• (2025). Synthetic image generation for effective deep learning model training for ceramic industry applications. EAAI, 143, 110019. (www)

  @article{ , abstract = {In the rapidly evolving field of machine learning engineering, access to large, high-quality, and well-balanced labeled datasets is indispensable for accurate product classification. This necessity holds particular significance in sectors such as the ceramics industry, in which effective production line activities are paramount and deep learning classification mechanisms are particularly relevant for streamlining processes; but real-world image samples are scarce and difficult to obtain, hindering dataset building and consequently model training and deployment. This paper presents a novel approach for dataset building in the context of the ceramic industry, which involves employing synthetic images for building or complementing datasets for image classification problems. The proposed methodology was implemented in CeramicFlow , an innovative computer graphics rendering pipeline designed to create synthetic images by employing computer-aided design models of ceramic objects and incorporating domain randomization techniques. As a result, a fully synthetic image dataset named Synthetic CeramicNet was created and validated in real-world ceramic classification problems. The results demonstrate that synthetic images provide an adequate basis for datasets and can significantly reduce reliance on real-world data when developing deep learning approaches for image classification problems in the ceramic industry. Furthermore, the proposed approach can potentially be applied to other industrial fields.}, archive = {J_EAAI}, author = {Fábio Gaspar and Daniel Carreira and Nuno Rodrigues and Rolando Miragaia and José Ribeiro and Paulo Costa and António Pereira}, doi = {10.1016/j.engappai.2025.110019}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110019}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Synthetic image generation for effective deep learning model training for ceramic industry applications}, volume = {143}, year = {2025}, }
• (2025). Collaborative semantic contrastive for all-in-one image restoration. EAAI, 143, 110017. (www)

  @article{ , abstract = {All-in-one image restoration, aiming to develop a unified model to deal with multiple image restoration tasks simultaneously, has great impact on the realistic application of Artificial Intelligence (AI). Owing to the diversity of multiple tasks, all-in-one image restoration cannot detour the feature entanglement problem. Thus, we propose Collaborative Semantic Contrastive for All-in-one Image Restoration (CSCAir) following the philosophy of normalizing multiple degradations into a common semantic category. Specifically, we propose a novel online knowledge distillation method of Collaborative Contrastive Learning (CCL). CCL employs a cohort of student models with identical architecture, namely Positive Network (PN), Negative Network (NN) and All-in-one Restoration Network (ARN), which are collaboratively trained on-the-fly. PN and NN are committed to establish a semantic space intersected by the clear and unclear manifolds. ARN is pulled closer to the clear semantic domain while being pushed out of the unclear semantic scope with contrastive learning. During the learning process, a synergy loss is established to keep communication among all student models. CCL can construct a good semantic embedding whilst enforcing all types of degraded images into the clear manifold, thus obtaining excellent restoration results. Furthermore, we justify that common metrics in contrastive learning are inadequate in measuring the semantic similarity. Therefore, we propose utilizing Wasserstein distance upon dense local descriptors to guarantee semantic similarity without fidelity in CCL. This paper is significant for addressing the feature entanglement problem in the context of all-in-one image restoration. Comprehensive experiments have been conducted across multiple benchmarks under both single-task and all-in-one settings to realize the implemented Artificial Intelligence (AI). Our method respectively achieves 35.87 decibel (dB) and 28.25 decibel (dB) in Peak Signal-to-Noise Ratio (PSNR), 0.9694 and 0.9643 in Structural Similarity Index Measure (SSIM) on deraining and dehazing task, outperforming the state-of-the-art approaches.}, archive = {J_EAAI}, author = {Bin Hu and Sai Yang and Fan Liu and Weiping Ding}, doi = {10.1016/j.engappai.2025.110017}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110017}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Collaborative semantic contrastive for all-in-one image restoration}, volume = {143}, year = {2025}, }
• (2025). Parameters identification of magnetorheological damper based on particle swarm optimization algorithm. EAAI, 143, 110016. (www)

  @article{ , abstract = {The parameter setting of the optimization algorithm is of significant importance in establishing a mechanical model with high accuracy. This study employs a combination of experimental and numerical methods to comprehensively examine the impact of optimization algorithm parameters on the accuracy of fitting results. The objective is to provide technical support for the precise prediction of the damping force in the control of the suspension system, as well as the optimization of vehicle driving performance. This paper employs the most prevalent particle swarm optimization algorithm and meticulously examines the impact of alterations in parameters, including the number of particles, the number of iterations and the learning factors, on the identification outcomes. The experimental data pertaining to the magnetorheological damper is obtained through investigation, and the parameters of the magnetorheological damper are identified through the utilisation of a numerical research methodology, specifically the particle swarm optimization algorithm. Finally, the veracity of the identified results is validated through a comparison of the identified damping force with the experimental damping force, thereby illustrating the significance of optimizing the algorithm parameter settings in enhancing the precision of the mechanical model.}, archive = {J_EAAI}, author = {Qianqian Guo and Xiaolong Yang and Kangjun Li and Decai Li}, doi = {10.1016/j.engappai.2025.110016}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110016}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Parameters identification of magnetorheological damper based on particle swarm optimization algorithm}, volume = {143}, year = {2025}, }
• (2025). An innovative approach to parasite classification in biomedical imaging using neural networks. EAAI, 143, 110014. (www)

  @article{ , abstract = {Background and objective In light of the increasing demand for precise and effective parasite detection in biomedical imaging, current models frequently depend on widely used convolutional neural networks (CNNs), with minimal innovation in feature selection and engineering for artificial intelligence. This study addresses these gaps by introducing a novel CNN, termed BLGSNet (Batch Normalization, Layer Normalization, GELU – Gaussian Error Linear Unit – and Swish functions-based network), specifically designed for parasite detection. BLGSNet increases feature engineering deploying a unique intersection-based feature selection method integrated into the deep learning model to improve classification accuracy. Materials and methods Using a publicly accessible image dataset with eight classes, including two blood cell types and six parasite types, we developed the BLGSNet architecture, which incorporates transformer-inspired design, convolution-based residual blocks, batch normalization, layer normalization, GELU, and Swish activation functions. The accompanying deep feature engineering model employs transfer learning and includes four key phases: (i) feature extraction using BLGSNet, (ii) feature selection using intersection methodology with neighborhood component analysis, chi-square, minimum redundancy maximum relevance, and ReliefF feature selectors, (iii) classification with k-nearest neighbors (kNN) and support vector machine (SVM), and (iv) automated selection of the optimal output. Results The proposed BLGSNet was trained on the designated dataset, achieving a test accuracy of 99.25%, while the deep feature engineering model attained an improved test accuracy of 99.59%, demonstrating the model's robustness. Conclusion The high classification accuracy achieved by BLGSNet and the feature engineering model highlights their potential in addressing the complex task of parasite detection.}, archive = {J_EAAI}, author = {Ozlem Aytac and Feray Ferda Senol and Ilknur Tuncer and Sengul Dogan and Turker Tuncer}, doi = {10.1016/j.engappai.2025.110014}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110014}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An innovative approach to parasite classification in biomedical imaging using neural networks}, volume = {143}, year = {2025}, }
• (2025). Attention-interactive horizontal–vertical graph-aware network for medical spine segmentation. EAAI, 143, 110013. (www)

  @article{ , abstract = {The health of the spine is vital to the human body. However, the single-scale convolutional network cannot fully capture the local detail features of the spine region. Meanwhile, the attribute decision-making ability of the boundary pixels between the spine region and the surrounding tissue structure could be better. To address these problems, we developed an attention-interactive horizontal–vertical graph-aware network to explore the semantic representation of the details of the spine region in magnetic resonance images. The network captures the coarse-grained multi-scale local spatial details of the spine region by dividing the subspace structure. Attention groups are used to refine the coarse-grained spatial features and establish rich channel dependencies and interactions between different subspace structures. Learning graph representation from horizontal and vertical directions and using the dynamic aggregation function between graph-aware nodes in the neighborhood to extract fine-grained semantics highlights the difference between the spinal region and the surrounding tissue structure. It improves boundary pixels’ attribute decision-making ability. The designed triple local attention embedding module establishes a complementary relationship. It balances coarse-grained features, fine-grained semantics, and prior knowledge to compensate for the shortcomings of a single feature representing the spine region. In addition, the designed weighted loss function performs separate supervised learning on each module and adaptively imposes penalties and constraints to encourage the network to achieve optimal performance. Finally, experimental results on the MRSpineSeg and CHAOS datasets show that the proposed framework has good segmentation performance and robustness, namely, mean intersection over union is 83.78% and 94.8%, respectively.}, archive = {J_EAAI}, author = {Yuyu Tian and Yalong Lv and XiaoYi Cai}, doi = {10.1016/j.engappai.2025.110013}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110013}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Attention-interactive horizontal–vertical graph-aware network for medical spine segmentation}, volume = {143}, year = {2025}, }
• (2025). A grey prediction model based on von bertalanffy equation and its application in energy prediction. EAAI, 143, 110012. (www)

  @article{ , abstract = {Accurately forecasting energy production and consumption is a critical challenge due to the inherent complexities and nonlinear behaviours of energy systems. Traditional grey forecasting models, while widely utilized, often struggle to adapt to these complexities. This study addresses these limitations by proposing a novel grey forecasting model, the Grey Von Bertalanffy Model ( G V B M ) model, which integrates Von Bertalanffy’s growth equation with Artificial Intelligence (AI)-based metaheuristic optimization techniques. The existing gap in the flexibility and adaptability of grey models is identified, and our approach introduces two additional parameters to better capture the dynamic behaviour of energy systems, especially for short-term forecasts. The primary aim of this research is to enhance the precision and reliability of energy forecasting by combining grey system theory with AI-driven optimization. The proposed G V B M leverages AI metaheuristic algorithms to fine-tune key parameters, thereby overcoming the structural limitations of traditional models. This method significantly improves the model’s ability to forecast energy production and consumption time series data with higher accuracy and adaptability. The G V B M is validated through application to six real-world energy scenarios involving electricity consumption and gas production. Across eight statistical metrics, the model consistently outperforms established grey forecasting models, demonstrating its superior forecasting accuracy, robustness, and adaptability. These results highlight the effectiveness of integrating AI with grey forecasting, offering a powerful tool for energy resource management and policy development.}, archive = {J_EAAI}, author = {Sajedeh Hedayatollahi Pour and Omid Solaymani Fard and Bo Zeng}, doi = {10.1016/j.engappai.2025.110012}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110012}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A grey prediction model based on von bertalanffy equation and its application in energy prediction}, volume = {143}, year = {2025}, }
• (2025). Integrating convolutional guidance and transformer fusion with markov random fields smoothing for monocular depth estimation. EAAI, 143, 110011. (www)

  @article{ , abstract = {Monocular depth estimation is a challenging and prominent problem in current computer vision research and is widely used in intelligent transportation like environment perception, navigation and localization. Accurately delineating object boundaries and ensuring smooth transitions in estimated depth images from a single image remain significant challenges. These issues place higher demands on the network’s global and local feature extraction capabilities. In response, we proposed a depth estimation framework, designed to address detection accuracy and the global smooth transition of predicted depth maps. Our method introduces a novel feature decoding structure named Co nvolutional G uided F usion (CoGF), which utilizes local features extracted by a convolutional neural network as a guide and fuses them with long-range dependent features extracted by a Transformer. This approach enables the model to retain both local details and global contextual information during the decoding process. To ensure global smoothness in the depth estimation results, we incorporate a smoothing strategy based on Markov Random Fields (MRF), enhancing pixel-to-pixel continuity and ensuring robust spatial consistency in the generated depth maps. Our proposed method is evaluated on current mainstream benchmarks. Experimental results demonstrate that our depth estimation method outperforms previous approaches. The code is available at https://github.com/pxrw/CGTF-Depth.git .}, archive = {J_EAAI}, author = {Xiaorui Peng and Yu Meng and Boqiang Shi and Chao Zheng and Meijun Wang}, doi = {10.1016/j.engappai.2025.110011}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110011}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating convolutional guidance and transformer fusion with markov random fields smoothing for monocular depth estimation}, volume = {143}, year = {2025}, }
• (2025). Enhancing glaucoma detection through multi-modal integration of retinal images and clinical biomarkers. EAAI, 143, 110010. (www)

  @article{ , abstract = {Glaucoma, a major cause of irreversible blindness globally, often progresses without early symptoms, making prompt and precise detection vital. This paper introduces a multi-modal glaucoma detection system that combines advanced deep learning architectures to analyze retinal images and clinical biomarkers. We developed three hybrid models: the first blends Vision Transformers (ViT) with Convolutional Neural Networks (CNN), specifically Residual Networks (ResNet), for comprehensive feature extraction; the second uses Object-Window-Location Vision Transformer (OWL-ViT) with Residual Networks for enhanced global contextual insights; and the third employs a Hierarchical Vision Transformer using Shifted Windows (Swin Transformer) with Residual Networks, which demonstrated the best performance. The strengths of these models, broad contextual capture by ViT, localized detail extraction by CNNs, and refined granularity by Swin Transformer, thereby improving both feature representation and computational efficiency, make them well-suited for clinical use. The best-optimized system, featuring the Swin Transformer hybrid model, achieved an F1-score of 0.993 for glaucoma and 0.995 for non-glaucoma, with an overall accuracy of 99.4% on a dataset of 2874 new cases, correctly classifying 2857 of them, thus confirming its efficacy in enhancing early-stage glaucoma detection and significantly advancing over existing methods.}, archive = {J_EAAI}, author = {Rishikesh Sivakumar and Anita Penkova}, doi = {10.1016/j.engappai.2025.110010}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110010}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing glaucoma detection through multi-modal integration of retinal images and clinical biomarkers}, volume = {143}, year = {2025}, }
• (2025). A distributed algorithm for fuzzy support vector machine on multi-source data-driven credit scoring. EAAI, 143, 110009. (www)

  @article{ , abstract = {Big data credit scoring is increasingly pivotal in credit systems. Yet, it faces several challenges that need addressing to enhance reliability. These challenges include dealing with uncertainty, ensuring privacy while integrating multi-source data across institutions, and overcoming limitations in memory space and computing power when handling massive data. To tackle these issues, we introduce a distributed algorithm leveraging a fuzzy support vector machine (SVM) to distinguish between fraudulent and trustworthy debtors. Our approach begins with employing a smoothing function to approximate the hinge loss in the original optimization problem of fuzzy SVM, rendering it differentiable. Then, we perform a multi-round distributed algorithm tailored to this new differentiable optimization problem. This algorithm requires only an initial fuzzy SVM estimator and iteratively refines the estimator through transmitting specific statistics and performing straightforward calculations. Furthermore, we provide theoretical proofs of the consistency of the initial estimator and the convergence of the algorithm throughout the iterative process, areas that have been rarely explored. Through simulation studies and empirical analysis, we validate the efficacy of our proposed method. Our analysis implies that our proposed distributed algorithm is helpful in the privacy preservation of isolated multi-source data to provide a fair and impartial credit scoring platform.}, archive = {J_EAAI}, author = {Wenjuan Li and Wei Shan and Maodian Liu}, doi = {10.1016/j.engappai.2025.110009}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110009}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A distributed algorithm for fuzzy support vector machine on multi-source data-driven credit scoring}, volume = {143}, year = {2025}, }
• (2025). Investigation of impact of aggregate shape on pervious concrete using machine learning classification methods. EAAI, 143, 110008. (www)

  @article{ , abstract = {Optimizing compaction energy reduces uncertainty in mass production of pervious concrete, but depends on aggregate-shape. This study analyses impact of aggregate size and shape on porosity and compressive strength. Aggregate-to-cement-ratio (AC) (3–5), compaction (Comp) (0–60 blows), aggregate-size (5–25 mm) and aggregate-shape (0, 200 and 1000 revolutions in Los Angeles Abrasion Value) were used to cast 486 samples. Porosity was computed from constituent-ratios (T.Poro) and apparent-weight (M.Poro) and Compressive-strength (Com.S) using Universal Testing Machine. M.Poro and T.Poro had significantly different impact from size and shape of aggregates. Feature selection models, Minimum-Redundancy-Maimum-Relevance (MRMR) and Kruskal-Wallis had significantly different ranking of features, owing to the different concepts. MRMR indicated dominance of M.Poro while a tenth of which was observed for Com.S and less than a twentieth for T.Poro. MRMR indicated significant impact of aggregate size (second highest score, half of M.Poro), indicating the impact aggregate shape had on performance was not uniform across different size of aggregates. Three performance parameters (M.Poro, T.Poro and Com.S) classified observations for aggregate shape (degree of milling) with an accuracy of 92.2% and 93% on Support-Vector-Machine (SVM) and K-Nearest-Neighbours (KNN) algorithms. Inclusion of aggregate size as feature improved accuracy to more than 96% while further inclusion of AC Comp improved the accuracy further to 99.6%. Classification of aggregate size was less accurate (70–80% with performance parameters alone and 92–93% with three design parameters included), indicating less impact on pervious concrete performance compared to shape of aggregates.}, archive = {J_EAAI}, author = {Sathushka Heshan Bandara Wijekoon and Navakulan Ahilash and Varatharaja Pravinjan and Karunanithy Virupashan and Navaratnarajah Sathiparan and Pratheeba Jeyananthan and Daniel Niruban Subramaniam}, doi = {10.1016/j.engappai.2025.110008}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110008}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Investigation of impact of aggregate shape on pervious concrete using machine learning classification methods}, volume = {143}, year = {2025}, }
• (2025). Self-supervised learning with ensemble representations. EAAI, 143, 110007. (www)

  @article{ , abstract = {Many computer vision applications, such as medical image processing, have struggled with a lack of labeled data. Recently, contrastive self-supervised learning has made remarkable progress in unsupervised representation learning and become a promising alternative in situations with limited labeled data. The simple Siamese network (SimSiam) is a well-known example in this area, known for its simplicity despite its powerful performance. However, it is known to be sensitive to changes in training configurations, such as hyperparameters and augmentation settings, due to its structural characteristics. To address this issue, we focus on the structural similarity between contrastive learning and the teacher–student framework in knowledge distillation. Inspired by the ensemble-based knowledge distillation approach, we propose a new self-supervised learning method, the ensemble representation model for simple Siamese networks (EnSiam), by introducing an ensemble representation in pseudo labels. This can reduce the variance in the contrastive learning procedure, providing better performance. Experiments demonstrate that EnSiam performs better than existing methods for popular datasets, showing that EnSiam is capable of learning high-quality representations. Furthermore, we extend our approach to various other self-supervised learning methods and empirically confirm that ensemble representations can consistently improve performance in general self-supervised learning.}, archive = {J_EAAI}, author = {Kyoungmin Han and Minsik Lee}, doi = {10.1016/j.engappai.2025.110007}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110007}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-supervised learning with ensemble representations}, volume = {143}, year = {2025}, }
• (2025). Inspection of cracking in stamping parts surfaces using anomaly detection. EAAI, 143, 110006. (www)

  @article{ , abstract = {Stamping parts are critical components of automobiles, and cracking represents the most serious quality issue in these parts. To effectively address the challenges of delay and low efficiency inherent in manual visual inspections, this article proposes an automated cracking detection framework. Owing to the difficulty in collecting cracking data in actual production, this research proposes a network named local and global self-supervision (LGSS) achieves cracking detection using normal data for model training and is utilized within the designed framework. The proposed LGSS network leverages collected normal samples of stamping parts to self-supervise the pre-trained model for fine-tuning, thereby enabling the model to extract features more effectively. The multivariate Gaussian distribution is employed to calculate the feature distribution of each pixel for anomaly detection (AD), addressing the issue of excessive exposure on stamping part surfaces being misidentified as defects. AD is conducted through both global and local branches, balancing hardware resource utilization while enhancing feature extraction and abnormal score calculation. The LGSS network achieved the score of area under the receiver operating characteristic (AUROC) curve of 100.0% for detection and 99.3% for localization in the actual cracking dataset. It can process stamping part images of sizes up to 1792 × 448 within 3 s using limited hardware resources. Experimental results demonstrate that the proposed framework can detect surface cracking on stamping parts both timely and accurately. Furthermore, the versatility of the proposed algorithm for defect detection in other industrial products was evaluated using the MVTec AD and BeanTech AD (BTAD) datasets.}, archive = {J_EAAI}, author = {Xingjun Dong and Changsheng Zhang and Dawei Wang and Qi Guo and Xinrui Deng and Chenyu Li}, doi = {10.1016/j.engappai.2025.110006}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110006}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Inspection of cracking in stamping parts surfaces using anomaly detection}, volume = {143}, year = {2025}, }
• (2025). A machine learning based framework for the probability failure envelope analysis of foundations in spatially variable clay. EAAI, 143, 110005. (www)

  @article{ , abstract = {Probabilistic failure envelopes are commonly used to represent the bearing capacity of skirted foundations in spatially variable soils. Traditional methods rely on probe loading and machine learning (ML) models, but these can lead to error propagation due to sparse data. This paper proposes a novel ML based framework for uncertainty quantification, replacing probe tests with a simplified modified swipe (SMS) method. The SMS method generates complete failure envelopes more quickly and accurately while providing more features to prevent error propagation, reducing errors by 43%. Among five ML methods—multilayer perceptron (MLP), convolutional neural network (CNN), k-nearest neighbour (KNN), decision tree (DT), and support vector machine (SVM)—MLP performed best, with prediction errors under 5% for coefficient of variation (COV) of 0.1 and below 10% for COV = 0.5. Additionally, the convex polynomial optimization method replaces the conventional fitting function and yields closed-form functions, which accurately represent various envelopes in random soil. The framework was validated across five different skirt lengths and soil conditions, providing safety factors for a 1% failure probability based on 4000 samples. This procedure holds promise for aiding in the reliability design of other offshore foundations.}, archive = {J_EAAI}, author = {Shuntao Fan and Yurong Zhang and Sa Li and Molin Han}, doi = {10.1016/j.engappai.2025.110005}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110005}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A machine learning based framework for the probability failure envelope analysis of foundations in spatially variable clay}, volume = {143}, year = {2025}, }
• (2025). Multimodal emotion recognition by fusing complementary patterns from central to peripheral neurophysiological signals across feature domains. EAAI, 143, 110004. (www)

  @article{ , abstract = {The implementation and application of artificial intelligence are propelling various advanced affective computing frameworks. Automatic recognition of emotions using multimodal physiological signals enhances the efficiency of systems such as health-care applications, pilot cognitive state monitoring, and passive brain-computer interfaces. However, challenges remain in capturing topological-frequency patterns in diverse electroencephalogram (EEG) electrode layouts, uncovering coupling dynamics across adjacent peripheral modalities, and integrating complementary affective patterns from the feature to modality levels. To address these issues, we propose a central-to-peripheral complementary integration network that employs hybrid encoders to extract and integrate affective patterns from EEG and peripheral signals. For EEG, the model unifies features from different channels into a single map to extract local-to-global representations, while for peripheral signals, adjacent cross-modal information is embedded into global affective patterns. These abstractions are systematically aggregated for emotion recognition by aligning affective relevance across domains and modalities within the central and peripheral nervous systems. The proposed model was evaluated on four publicly available multimodal databases using a leave-one-subject-out cross-validation approach. On the Database for Emotion Analysis using Physiological signal (DEAP), the binary recognition accuracy for valence and arousal scales was 75.00% and 77.33%, respectively. On the Human-Computer Interaction (HCI) database, the corresponding binary accuracies were 78.78% and 75.38%. For the SJTU Emotion EEG Datasets IV and V (SEED-IV and SEED-V), the four-class and five-class accuracies were 71.94% and 84.83%, respectively. These results validate the robustness and remarkable generalization capability of the proposed method.}, archive = {J_EAAI}, author = {Zhuang Ma and Ao Li and Jiehao Tang and Jianhua Zhang and Zhong Yin}, doi = {10.1016/j.engappai.2025.110004}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110004}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodal emotion recognition by fusing complementary patterns from central to peripheral neurophysiological signals across feature domains}, volume = {143}, year = {2025}, }
• (2025). Anomaly detection in all-sky images: An approach using robust ensemble modeling of cloud cover fraction and prediction bounds. EAAI, 143, 110003. (www)

  @article{ , abstract = {All-sky images (ASI) are widely used for sky monitoring, particularly in solar energy generation applications. Alignment issues and interferences demand a detection process of problematic images. With high sampling frequencies (1-2 images per minute), automating this process is crucial for managing large datasets and enabling integration into automatic systems, which is the objective of this work. For this purpose, a robust ensemble model, using the ApproxHull and Radial Basis Function (RBF) neural networks combined with Multi Objective Genetic Algorithms (MOGA) tools, was developed to compute the cloud cover fraction of each image. By computing the deviation between this result and the one obtained by the equipment, and by assessing if it lies within prediction bounds obtained in the design phase, an automatic method for detecting anomalies in All-sky images was obtained. ASI data collected during the Green Ocean (GoAmazon) Experiment 2014/5 was employed. The proposed approach obtained a Probability of Interval Coverage (PICP) similar to the user-specified level of confidence for several sets within what was classified as a “good” dataset, while being able to detect anomalies found within a “bad” dataset.}, archive = {J_EAAI}, author = {Vinicius Roggério da Rocha and Gilberto Fisch and Rodrigo Santos Costa and António Ruano}, doi = {10.1016/j.engappai.2025.110003}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110003}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Anomaly detection in all-sky images: An approach using robust ensemble modeling of cloud cover fraction and prediction bounds}, volume = {143}, year = {2025}, }
• (2025). An enhanced deep reinforcement learning approach for efficient, effective, and equitable disaster relief distribution. EAAI, 143, 110002. (www)

  @article{ , abstract = {Efficient disaster response, especially within the critical initial 72 h, is crucial for saving lives. However, allocating relief goods effectively to affected areas remains a complex challenge due to uncertainty, limited resources, and dynamic needs. This study addresses this challenge by proposing a multi-period integer nonlinear programming model for efficient, effective, and equitable distribution of relief goods during disaster response phase. To optimize relief allocation within entire 72-h, a novel decision-making approach is proposed that leverages the proximal policy optimization (PPO) algorithm. It uses deep residual neural networks for state-value and optimal action prediction with 5 value and 4 policy residual layers. Additionally, an algorithm-agnostic termination criterion based on episodic reward stall ensures effective convergence detection without requiring prior knowledge of optimal solution. The provided model and solution methods are validated through 30 hypothetical problem instances and a realistic earthquake response case study. The results demonstrate the superiority of proposed approach compared to traditional methods like dynamic programming, state-action-reward-state-action (SARSA), and Q-learning, in terms of both solution quality and sample efficiency. Notably, the deep residual networks and proposed termination criterion enable the PPO algorithm to achieve an average optimality gap of less than 10% for the majority of instances with consistent hyperparameters, while exhibiting significant sample efficiency gains, particularly for large-scale problems. This research empowers disaster managers with an efficient and timely relief delivery plan, ultimately contributing to saving lives in the face of disaster. Moreover, proposed termination criterion may improve the performance of reinforcement learning in other application areas.}, archive = {J_EAAI}, author = {Moiz Ahmad and Muhammad Tayyab and Muhammad Salman Habib}, doi = {10.1016/j.engappai.2025.110002}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110002}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An enhanced deep reinforcement learning approach for efficient, effective, and equitable disaster relief distribution}, volume = {143}, year = {2025}, }
• (2025). Identifying wagon numbers using transformers. EAAI, 143, 110000. (www)

  @article{ , abstract = {Identifying wagons in the railway industry from image data is essential in streamlining the process of wagon inspection and fleet monitoring. However, this is a laborious and time-consuming task, hence automating this process is pivotal in accelerating the identification process and removing human bias. This challenge is addressed in this work, and a multi-stage approach for identifying wagons by detecting and assembling their unique International Union of Railways (UIC) numbers from image data is introduced. Employing a Transformer architecture, the methodology surpasses other methods in accuracy obtained over real-world images and offers fast operation. The model’s effectiveness is demonstrated through extensive experiments performed on a unique dataset of more than 3,000 images captured in various weather conditions. The top-performing machine learning model exhibits exceptional speed and real-time performance, processing an average of 100 UIC-containing frames in less than 0.1 s on an NVIDIA A10G Tensor Core Graphics Processing Unit, and offering the accuracy of assembling UIC numbers exceeding 95% in scenarios where 1 to 4 digits are missing. The findings presented in this article not only open avenues to apply sequence-to-sequence models to various industrial object detection tasks, but can also significantly improve railway operations.}, archive = {J_EAAI}, author = {Bartłomiej Gładys and Bartłomiej Wojakowicz and Michał Bober and Jakub Nalepa}, doi = {10.1016/j.engappai.2024.110000}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {110000}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identifying wagon numbers using transformers}, volume = {143}, year = {2025}, }
• (2025). Mamba-enhanced spectral-attentive wavelet network for underwater image restoration. EAAI, 143, 109999. (www)

  @article{ , abstract = {Underwater image restoration is a critical research area in computer vision, owing to its inherent complexities and diverse applications in challenging environments. Existing methods generally neglect the fusion of frequency domain and spatial domain features, and the recovery effect is limited due to the loss of local details caused by wavelet transform. Although some studies have tried to improve the recovery effect by introducing frequency domain information or enhancing spatial features, it is still difficult to reduce the colour bias and improve the image contrast while preserving the details. To address these limitations, this paper introduces a Mamba-enhanced Spectral-Attentive Wavelet Network for effective underwater image recovery. The proposed method comprises three pivotal modules: a spatial detail enhancement encoder, a state–space model for local information compensation, and a spectral cross-attention residual module. The spatial detail enhancement encoder utilises differential convolution and Atrous Spatial Pyramid Pooling to refine average coefficients post-wavelet transform, ensuring precise detail restoration. The state–space model for local information compensation leverages Two-Dimensional Selective Scanning to dynamically integrate multi-scale data, enhancing global context understanding. The spectral cross-attention residual module focuses on optimising high-frequency component recovery through Fourier feature extraction and adaptive convolution, thereby improving both spectral and spatial fidelity. Extensive experimental evaluations demonstrate that the proposed method excels in underwater image recovery tasks, outperforming existing state-of-the-art techniques and showcasing significant potential for practical applications. The code for our method is available at https://github.com/makabala/MESA-WNet .}, archive = {J_EAAI}, author = {Baocai Chang and Genji Yuan and Jinjiang Li}, doi = {10.1016/j.engappai.2024.109999}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109999}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mamba-enhanced spectral-attentive wavelet network for underwater image restoration}, volume = {143}, year = {2025}, }
• (2025). Forest-ORE: Mining an optimal rule ensemble to interpret random forest models. EAAI, 143, 109997. (www)

  @article{ , abstract = {Random Forest (RF) is well-known as an efficient ensemble learning method with strong predictive performance. However, it is often regarded as a “black box” due to its reliance on hundreds of deep decision trees. This lack of interpretability can be a real drawback for the acceptance of RF models in several real-world applications, especially those affecting individuals’ lives. In this work, we present Forest-ORE, a method that makes RF interpretable via an optimized rule ensemble (ORE) for local and global interpretation. Unlike other rule-based approaches aimed at interpreting the RF model, this method simultaneously considers several parameters that influence the choice of an interpretable rule ensemble. Existing methods often prioritize predictive performance over interpretability coverage and do not account for existing overlaps or interactions between rules. Forest-ORE uses a mixed-integer optimization program to build an ORE that considers the trade-off between predictive performance, interpretability coverage, and model size (ensemble size, rule length, and overlap). In addition to producing an ORE competitive with RF in predictive performance, this method enriches the ORE through other rules that afford complementary information. This framework is illustrated through an example, and its robustness is evaluated across 36 benchmark datasets. A comparative analysis with well-known methods shows that Forest-ORE achieves an excellent trade-off between predictive performance, interpretability coverage, and model size.}, archive = {J_EAAI}, author = {Maissae Haddouchi and Abdelaziz Berrado}, doi = {10.1016/j.engappai.2024.109997}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109997}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Forest-ORE: Mining an optimal rule ensemble to interpret random forest models}, volume = {143}, year = {2025}, }
• (2025). Interpretable manifold learning for T-wave alternans assessment with electrocardiographic imaging. EAAI, 143, 109996. (www)

  @article{ , abstract = {T-wave alternans (TWA) is a biomarker for sudden cardiac death prediction, characterized by subtle variations in the amplitude or morphology of consecutive T-waves in electrocardiographic studies. Electrocardiographic imaging (ECGI) offers increased spatial resolution, enabling TWA distribution analysis across the epicardium. However, existing TWA estimation methods disregard ECGI spatial information by analyzing each signal independently. To address this gap, we present a novel, subject-specific, interpretable manifold learning-based TWA estimation method tailored to ECGI. First, Uniform Manifold Approximation and Projection (UMAP) reduces input data dimensions. Second, the Louvain algorithm detects communities and identifies the TWA-dominant community. Finally, the location of this community and the rest of the communities is compared, and a Bootstrap-based TWA classifier is applied. A customized Shapley additive explanations method was developed to identify the signal segments most affecting the algorithm decisions to enhance explainability. Reducing the input data to 18 dimensions improved the separation of the TWA-dominant community, with an average normalized distance of 0.28. The Bootstrap analysis showed that the TWA-dominant community had a distance metric up to 0.2 above the confidence interval upper limit. The TWA-dominant community input signals showed different TWA patterns, namely, hump-shaped and amplitude-shifted TWA, and the interpretability algorithm revealed that UMAP focuses on them when projecting points into the latent space. Our method achieved maximum accuracy in subjects with known outcomes and made consistent patient decisions based on input signals. This study introduces the first ECGI-specific TWA detection method. Its subject-specific nature enables the extraction of individual-specific characteristics, offering personalized diagnostic insights.}, archive = {J_EAAI}, author = {E. Sánchez-Carballo and F.M. Melgarejo-Meseguer and R. Vijayakumar and J.J. Sánchez-Muñoz and A. García-Alberola and Y. Rudy and J.L. Rojo-Álvarez}, doi = {10.1016/j.engappai.2024.109996}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109996}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interpretable manifold learning for T-wave alternans assessment with electrocardiographic imaging}, volume = {143}, year = {2025}, }
• (2025). An effective skeleton-based approach for multilingual sign language recognition. EAAI, 143, 109995. (www)

  @article{ , abstract = {Effective communication with the deaf community necessitates the recognition of sign language; however, identifying skeleton-based sign language poses numerous challenges. These challenges include capturing temporal dynamics, sequential patterns, and other gesture nuances such as signer variability, gesture class imbalance, and semantic ambiguity. Therefore, developing robust methods for creating skeletal representations that accurately capture temporal dynamics, address semantic ambiguity, and remain connected to various communities is crucial for enhancing usability and inclusiveness. This paper presents a Skeleton-Based Sign Language Recognition (SB-SLR) framework for sign language recognition, addressing challenges by leveraging features derived from pivotal frames of sign videos. Our technique comprises three steps: Identification of Indicator Frame (IIF), Identification of Pivotal Frames (IPF), and Generation of Skeleton Sequence (GSS). We utilize a Convolutional Neural Network model for classification, which takes these sequences of skeleton images as input to recognize sign words. The SB-SLR algorithm demonstrates superior performance in labeling specific sign words, potentially enhancing communication accessibility for the hearing impaired.}, archive = {J_EAAI}, author = {S. Renjith and M.S. Sumi Suresh and Manazhy Rashmi}, doi = {10.1016/j.engappai.2024.109995}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109995}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An effective skeleton-based approach for multilingual sign language recognition}, volume = {143}, year = {2025}, }
• (2025). Target recognition and localization of environmental perception system based on binocular cameras for unmanned cleaning vehicle. EAAI, 143, 109994. (www)

  @article{ , abstract = {Cleaning vehicles, vital in municipal projects, automated via unmanned driving, boost urban cleaning efficiency and sustainability, cutting costs and minimizing environmental impact. Environmental perception is key to their automation, differing from passenger cars. Cleaning vehicles often encounter issues such as low cleaning efficiency, high energy consumption, and suboptimal cleaning results when operating in complex and variable scenarios. The automated cleaning vehicle uses an environmental perception system to identify waste and provides vital information for controlling its cleaning devices, boosting both cleaning and energy efficiency. However, existing object detection algorithms struggle with challenges such as background interference and the small size of road debris, leading to limited detection capabilities. This paper proposes an environmental perception solution specifically tailored for the operational context of cleaning vehicles. The solution uses a binocular camera system, integrating stereo vision technology with object detection algorithms to identify and locate road debris. Attention mechanisms are incorporated to enhance model detection accuracy. An additional layer for small object detection is implemented to improve the ability to detect smaller targets, increasing the mean average precision by 5.1% and the recall rate by 6.61%. These improvements could lead to more accurate and efficient automated cleaning systems in urban environments, with far-reaching implications for the future of smart cities.To validate the feasibility of the proposed solution, tests are conducted using a road debris dataset and real-vehicle application experiments. The experimental results demonstrate that the system effectively performs debris identification and localization tasks on cleaning vehicles.}, archive = {J_EAAI}, author = {Qihuai Chen and Jianhe Wen and Tianliang Lin and Haoling Ren}, doi = {10.1016/j.engappai.2024.109994}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109994}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Target recognition and localization of environmental perception system based on binocular cameras for unmanned cleaning vehicle}, volume = {143}, year = {2025}, }
• (2025). Study on image acquisition and camera positioning of depth recognition model in the tobacco curing stage. EAAI, 143, 109992. (www)

  @article{ , abstract = {To address the challenges in the digital transformation of the tobacco industry regarding the recognition of data acquisition standards and the construction of an Internet of Things (IoT) system for tobacco leaf curing stages, these deep neural networks are applied in this field to construct an identification model. Data was collected under various conditions, including different camera types (with or without distortion, focal length), installation positions, and lighting conditions, to obtain curing stage data. After preprocessing each type of image data, ten-stage classification recognition datasets were established based on the “three stages and six steps” curing process. Six recognition models were developed using ResNeXt-50, ShuffleNetV2(1.0), MobileNetV3-S, VanillaNet-10, EfficientNetV2-S, and EfficientNetV2-M as backbone networks. Evaluation and analysis were conducted using multiple performance metrics such as accuracy, F1 score, as well as various graphical representations including curve plots, radar charts, bar charts, and confusion matrices. The results indicate: 1. ShuffleNetV2(1.0) (MT: 98.71%) and EffcientNetV2(MT: 99.84%) series networks exhibit superior recognition performance. 2. Recognition performance varies between high-temperature and low-temperature areas. 3. Combining multiple perspectives can improve recognition accuracy. 4. Cameras with larger focal lengths, cool white lighting, and distortion are more conducive to recognition. 5. The application prospects of image-assisted tobacco leaf curing are promising. Code available at: https://github.com/vontran2021/CuringStage .}, archive = {J_EAAI}, author = {Chuan Feng and Shiping Zhu and Maojie Tang and Hu Zhao and Quan Yuan and Bojun Wang}, doi = {10.1016/j.engappai.2024.109992}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109992}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Study on image acquisition and camera positioning of depth recognition model in the tobacco curing stage}, volume = {143}, year = {2025}, }
• (2025). A force neural network framework for structural optimization. EAAI, 143, 109991. (www)

  @article{ , abstract = {In this paper, an efficient Force Neural Network (FNN) is developed to reformulate the size optimization of truss structures as an operator learning problem. A Deep Neural Network (DNN) is designed to directly map the connectivity information of truss members to the corresponding design variables. Therein, the entire unlabeled training data contains only the connectivity information of members, without any structural responses, weights, or cross-sectional areas. By integrating Force Method (FM), our framework embeds the optimal design problem represented by the objective and constraint functions in the loss function to guide the training process. And it guarantees that the generated solution is consistent with the underlying physical principles. In addition to enhance efficiency in finding the optimum structural weight, Bayesian Optimization (BO) is applied for automatic hyper-parameters tuning instead of the trial and error method. As soon as the training phase ends, the optimal weight of truss structures is found without using any other numerical methods. Several numerical examples are investigated to demonstrate the effectiveness and applicability of the FNN for the optimization of truss structures. The obtained results indicate that it not only be simple to perform but also overcomes the local optimal problem and reduces the computational cost in high-dimensional problems.}, archive = {J_EAAI}, author = {Dai D. Mai and Si T. Do and Seunghye Lee and Hau T. Mai}, doi = {10.1016/j.engappai.2024.109991}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109991}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A force neural network framework for structural optimization}, volume = {143}, year = {2025}, }
• (2025). A recurrent lyapunov neural network for fast tracking of unknown single-input single-output nonlinear systems. EAAI, 143, 109990. (www)

  @article{ , abstract = {Lyapunov stability-based neural network (NN) controllers have been widely applied to nonlinear systems, yet challenges such as high computational cost and reliance on signal estimators persist. This paper introduces a novel estimator-free neuro-adaptive control approach using a Lyapunov Stable Machine Learning Recurrent Neural Network (LSMRN). The key contributions are twofold: reducing computational complexity by replacing the traditional neural network estimator (NNE) and controller (NNC) with a single recurrent neural network, and significantly improving tracking performance. The LSMRN relies solely on input/output data, eliminating the need for explicit system models and control signal estimators in the closed loop. Lyapunov stability analysis ensures closed-loop stability and error convergence. Simulations on the Lorenz Chaotic and 3D Chaotic Satellite systems demonstrate the controller’s superior tracking accuracy with reduced control effort, highlighting its robustness and efficiency compared to conventional methods.}, archive = {J_EAAI}, author = {Muhammad Sadiq and Muhammad Shafiq and Naveed R. Butt}, doi = {10.1016/j.engappai.2024.109990}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109990}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A recurrent lyapunov neural network for fast tracking of unknown single-input single-output nonlinear systems}, volume = {143}, year = {2025}, }
• (2025). Model-free output feedback optimal tracking control for two-dimensional batch processes. EAAI, 143, 109989. (www)

  @article{ , abstract = {To address the challenge of linear quadratic tracking for batch processes without taking into account the specifics regarding the system dynamics and the state observer, a two-dimensional off-policy model-free output feedback control method is proposed. Previous solutions have mostly failed to fully explore the historical input and output information of the system along time and batch dimensions, thereby constraining the flexibility and efficiency of the controller. However, the method proposed aims to reconstruct the state space by delving into these historical input and output data of the system. The correlation between the two-dimensional value function and the two-dimensional Q-function with output feedback characteristics is found, thereby revealing the corresponding two-dimensional Bellman equation. This method, through the use of reinforcement learning algorithms, can effectively learn the optimal control strategy without the need for prior knowledge of system dynamics. To this end, the proposed method not only converges faster but also has smaller errors, making it more suitable for complex and ever-changing industrial production processes in reality. This innovative research provides new ideas and methods for the design and optimization of control systems, which is expected to bring higher efficiency and more stable performance to industrial production. At last, simulation outcomes for the injection velocity process validate the suggested method's effectiveness.}, archive = {J_EAAI}, author = {Huiyuan Shi and Jiayue Ma and Qiang Liu and Jinna Li and Xueying Jiang and Ping Li}, doi = {10.1016/j.engappai.2024.109989}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109989}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Model-free output feedback optimal tracking control for two-dimensional batch processes}, volume = {143}, year = {2025}, }
• (2025). A cloud-based leakage current classified system for high voltage insulators with improved particle swarm optimization and hybrid deep learning technique. EAAI, 143, 109987. (www)

  @article{ , abstract = {Determining the severity level of the deteriorative insulator is the most critical challenge in power system maintenance. The flashover phenomenon due to pollution and aging causes severe damage to the power system and reduces power quality. Therefore, a cloud leakage current (LC) classified system is essential for preventing flashover phenomena and effectively providing indicators in maintenance services. This research develops an innovative cloud system to identify the LC levels of high voltage 15 kV insulators based on convolution neural network bidirectional gated recurrent unit (CNN-BGRU) deep learning algorithms. The LC data were thoroughly examined and experimented with different weather conditions in polluted coastal areas of Taiwan for more than 1 year. The improved particle swarm optimization (IPSO) is proposed to identify the most optimum architecture of the hybrid CNN-BGRU model through hyperparameter fine-tuning operation. The LC-classified performance of developed frameworks is analyzed and compared with other state-of-the-art deep learning techniques, including long short-term memory (LSTM), convolutional neural network (CNN), gated recurrent unit (GRU), BGRU, recurrent neural network (RNN), and CNN-GRU models. The classified ability of proposed IPSO-CNN-BGRU frameworks is demonstrated through different comparative benchmarks, which provides the advantages of the cloud system for classifying LC level of 15 kV insulators in maintenance services.}, archive = {J_EAAI}, author = {Thanh-Phuong Nguyen and Ming-Yuan Cho}, doi = {10.1016/j.engappai.2024.109987}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109987}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cloud-based leakage current classified system for high voltage insulators with improved particle swarm optimization and hybrid deep learning technique}, volume = {143}, year = {2025}, }
• (2025). A novel convolutional neural network with global perception for bearing fault diagnosis. EAAI, 143, 109986. (www)

  @article{ , abstract = {Bearings are key support components in rotating machinery, and their stability is crucial to the reliability of the entire mechanical system. To address the limitations of existing Transformer architectures in edge-side optimization and convolutional neural networks in global feature extraction, especially the resulting poor real-time performance and low accuracy in bearing fault diagnosis based on acoustic signals, this paper proposes a novel global-aware convolutional neural network based on residual masking and position-aware strategies (ParC-ReSMNet). Firstly, the network is based on the residual network (ResNet-18) and designed with a residual mask block combined with an improved spatial pyramid mask attention mechanism, which effectively removes redundant spatial information and focuses on critical fault features, thereby enhancing the robustness and generalization of the network. Secondly, a position-aware circular module is introduced to replace specific residual blocks in the original network, achieving an effective fusion of positional and global information, thereby augmenting the modeling capability of the convolutional neural network. Experiments are conducted on a self-made belt conveyor idler dataset and the Detection and Classification of Acoustic Scenes and Events (DCASE) 2023 Task2 bearing dataset, with results showing that ParC-ReSMNet achieves 95.49% and 96.67% accuracy, respectively. Compared to seven state-of-the-art models, it has the highest precision and recall, along with good real-time performance, which demonstrates great application value for fault monitoring of belt conveyors used in coal mines, power plants, ports, and other rotating machinery. The code library is available at: https://github.com/xgli411/Parc-ResMNet .}, archive = {J_EAAI}, author = {Xianguo Li and Ying Chen and Yi Liu}, doi = {10.1016/j.engappai.2024.109986}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109986}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel convolutional neural network with global perception for bearing fault diagnosis}, volume = {143}, year = {2025}, }
• (2025). Autoregressive integrated moving average with semantic information: An efficient technique for intelligent prediction of dengue cases. EAAI, 143, 109985. (www)

  @article{ , abstract = {Herein, we proposed a novel approach, AutoRegressive Integrated Moving Average with Semantic Information (ARIMAS), an extension of the ARIMA model that incorporates a knowledge base about dengue outbreaks. This extension considerably improves the analysis capabilities of traditional time-series forecasting models. ARIMAS differentiates itself by integrating semantic information from external knowledge bases, enabling a comprehensive understanding of the relationships between data points using an ontological model. This approach extracts related data from a dataset to enhance the autoregressive technique beyond ARIMA's reliance on historical time-series data alone. We evaluate ARIMAS using a provincial dengue dataset from Thailand, comparing its performance with those of state-of-the-art methods: ARIMA, simple exponential smoothing (SES), and Holt's exponential smoothing (HES). The results demonstrate that ARIMAS achieves an average mean square error of 4.88E+05, outperforming ARIMA (5.04E+05), SES (7.60E+05), and HES (7.67E+05). ARIMAS represents a promising advancement in time-series forecasting for dengue outbreaks by addressing the key limitations of ARIMA models. By combining statistical analysis with domain-specific knowledge, ARIMAS can provide more accurate, interpretable, and contextually relevant predictions, ultimately supporting effective public health interventions. This advancement in prediction accuracy highlights the prediction effectiveness of ARIMAS, particularly in unusual situations that may exhibit short, nonlinear trends. This enhanced method can eliminate nonlinearity through outlier reduction and is thus a robust, superior tool for time-series analysis that overcomes the constraints of traditional models.}, archive = {J_EAAI}, author = {Wanarat Juraphanthong and Kraisak Kesorn}, doi = {10.1016/j.engappai.2024.109985}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109985}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Autoregressive integrated moving average with semantic information: An efficient technique for intelligent prediction of dengue cases}, volume = {143}, year = {2025}, }
• (2025). Underwater moving target detection and tracking based on enhanced you only look once and deep simple online and realtime tracking strategy. EAAI, 143, 109982. (www)

  @article{ , abstract = {Addressing the challenges posed by factors like light attenuation, turbid water quality, large target scale changes, and dense distribution in underwater moving target images, we propose an enhanced strategy combining YOLOv5 (You Only Look Once version 5) and Deep SORT (Simple Online and Realtime Tracking) for target detection and tracking. Initially, various image enhancement techniques are employed to improve detection performance, while a Deep Convolutional Generative Adversarial Networks (DCGAN) approach augments the small underwater image dataset. Subsequently, we enhance the YOLOv5 method by integrating a convolutional attention module into the detection network to enhance target salience in large-scale scenes. Additionally, a tiny target detection head is introduced to enhance the detector’s capability in adapting to scale changes, particularly for small objects. Finally, the Deep SORT algorithm is integrated for tracking detected targets. Experimental comparisons with state-of-the-art methods validate the efficacy of the proposed approach. Our method achieves a detection accuracy of 94.3% on a self-made underwater fish target dataset, with a detection speed of 71FPS and robust tracking performance. Moreover, its applicability extends to various underwater targets, demonstrating superior performance.}, archive = {J_EAAI}, author = {Bing Sun and Wei Zhang and Cheng Xing and Yingyao Li}, doi = {10.1016/j.engappai.2024.109982}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109982}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Underwater moving target detection and tracking based on enhanced you only look once and deep simple online and realtime tracking strategy}, volume = {143}, year = {2025}, }
• (2025). Predicting postural risk level with computer vision and machine learning on multiple sources of images. EAAI, 143, 109981. (www)

  @article{ , abstract = {Postural risk level assesses the degree of potential harms to a human body due to bad postures often encountered in workplaces. If accurate 3-dimensional coordinates of body joints can be found from images, then postural risk level can be calculated analytically. Unfortunately, even with today's very successful deep learning technologies, human pose estimation algorithms still cannot give accurately enough coordinates for the task. Machine learning (ML) may be needed to improve risk estimation from outputs of pose estimators. In this study, we apply ML to train a risk level predictor based on joint angles calculated from outputs of pose estimators. We validate our work with publicly available image datasets. Two state-of-the-art pose estimators with pre-trained models were used as is. Multiple sources of images were combined in different ways to tackle covariate shift and concept drift issues in ML. It is found that a weakly supervised algorithm increased balanced accuracy and F1 score substantially from the baseline approach on two public image datasets. F1 score is the harmonic mean of precision and recall. A two-stage multi-source domain adaptation algorithm also improved these measures, though the improvement was not as impressive as the weak supervision approach. The baseline approach calculated risk levels analytically with outputs of pose estimators. The balanced accuracy is 0.676/0.735/0.745 for baseline/two-stage/weak-supervision with one pose estimator, and 0.749/0.745/0.761 with another one. The F1 score is 0.692/0.726/0.736 for baseline/two-stage/weak-supervision with one pose estimator, and 0.745/0.732/0.748 with another estimator. Applications to private images with the proposed procedure are explained.}, archive = {J_EAAI}, author = {Shing Hwang Doong}, doi = {10.1016/j.engappai.2024.109981}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109981}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predicting postural risk level with computer vision and machine learning on multiple sources of images}, volume = {143}, year = {2025}, }
• (2025). Cattle weight estimation model through readily photos. EAAI, 143, 109976. (www)

  @article{ , abstract = {Cattle weight is a key indicator for assessing health, meat quality, productivity, and overall well-being. Manual measurement methods are time-consuming and induce stress reactions. In contrast, non-contact measurement methods based on computer vision can effectively and rapidly estimate cattle weight. These non-contact methods typically involve image processing and three-dimensional reconstruction, capturing image information with cameras and combining deep learning algorithms to estimate cattle weight. Despite their advantages, these methods are mainly designed for large-scale professional cattle farms and are characterized by high costs and complex operations, which do not meet the needs of individual cattle farmers. To address this issue, we propose the Lightweight Network-based Cattle Weight Estimation (LaWE) model, deployed on mobile phones and can measure cattle weight by capturing readily available photos. The LaWE model uses an optimized lightweight multi-scale fusion network to accurately extract keypoints from images and calculate relevant body measurements, which are then input into a specially designed deep regression network to estimate cattle weight. We conducted extensive experiments on both publicly available datasets and our self-collected dataset of 108 Horqin Yellow cattle. The results demonstrate that the LaWE model achieves an accuracy rate of over 97% in cattle weight estimation, meeting the requirements for low cost and real-time performance. Our research makes a significant contribution to the field of non-contact cattle weight estimation, providing practical technical support for smallholder farmers in the industry.}, archive = {J_EAAI}, author = {Lili Bai and Chaopeng Guo and Jie Song}, doi = {10.1016/j.engappai.2024.109976}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109976}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cattle weight estimation model through readily photos}, volume = {143}, year = {2025}, }
• (2025). Bi-objective predictive modelling and optimization in micro-electrical discharge machining process using an artificially intelligent system. EAAI, 143, 109975. (www)

  @article{ , abstract = {Electrical Discharge Machining (EDM) is a highly complex, non-linear system whose process modelling using normal techniques such as regression is not much favorable. Hence, soft computing technique such as artificial neural network (ANN), is applied for modelling. Numerous issues are encountered in conventional machining of Nimonic 80A and hence EDM is one of the favorable techniques for machining. Based on obtained value of mean square error , rigorous checks were done to obtain the best performance and it was found that ANN model trained by employing gradient descent with momentum and adaptive learning rate (GDX) algorithm outperformed others due to its exhaustive training and fine tuning. Reasonable accuracy of the ANN model was indicated by its prediction results. Considering the conflicting nature of the objectives, an evolutionary algorithm, non-dominated sorting genetic algorithm-II (NSGA-II) is utilized for performing multi-objective optimization. The distance of the non-dominated pareto optimal solutions from the ideal point, is proposed as a solution for selection of optimum point. Confirmatory experiments were carried out and the error obtained was under acceptable limits. Scanning Electron Microscopy (SEM) images indicated that at the optimal setting, considerable finer hole quality, lower value of overcut, less dense re-cast layer and less dense heat affected zone (HAZ), were seen. This efficient intelligent method may serve as a decision support system through its improved prediction accuracy with less dependency on the experimental data and simultaneous optimization of the process conditions.}, archive = {J_EAAI}, author = {Piyush Pant and Pushpendra S. Bharti}, doi = {10.1016/j.engappai.2024.109975}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109975}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bi-objective predictive modelling and optimization in micro-electrical discharge machining process using an artificially intelligent system}, volume = {143}, year = {2025}, }
• (2025). Ship re-identification in foggy weather: A two-branch network with dynamic feature enhancement and dual attention. EAAI, 143, 109974. (www)

  @article{ , abstract = {Most existing ship re-identification methods are only applicable under sunny days. However, the presence of fog makes ship re-identification difficult. Therefore, this study proposes a two-branch network with dynamic feature enhancement and dual attention, which includes the defogging and re-identification branch, employing an end-to-end training approach to simultaneously learn defogging and re-identification tasks. To address issues related to blurring images and changes in ship appearance caused by fog, this study proposes a dynamic feature enhancement module, utilizing the deformable group convolution to adaptively adjust the shapes and positions of kernels. The adjustment enables better matching with ship actual outlines and size changes, thereby accurately locating salient regions in ship images. To further improve the network’s ability in capturing ship details in foggy weather and recover clear and fog-free images, a dual attention module fusing multiple pooling operations and using convolution kernels of different sizes is proposed, which strengthens the network’s attention to capture ship characteristics in different channels and spaces. To enhance the discrimination of learned ship vital attributes, this study introduces the adaptive sparse pair loss, dynamically adjusting the network’s attention to positive and negative sample pairs with different attributes. In addition, a new dataset is constructed specifically for foggy ship re-identification. The atmospheric scattering model is employed for the synthesis of foggy ship images, effectively addressing the challenge associated with training and evaluating foggy ship re-identification networks. Experimental results show that the proposed method outperforms existing ship re-identification methods in foggy weather.}, archive = {J_EAAI}, author = {Wei Sun and Fei Guan and Xiaorui Zhang and Xinyi Shen and Kaibo Wang}, doi = {10.1016/j.engappai.2024.109974}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109974}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ship re-identification in foggy weather: A two-branch network with dynamic feature enhancement and dual attention}, volume = {143}, year = {2025}, }
• (2025). Transductive gradient injection for improved hyperspectral image denoising. EAAI, 143, 109973. (www)

  @article{ , abstract = {In real-world hyperspectral imaging, noise disproportionately affects specific spectral bands. However, existing denoising techniques struggle to discern the varied contributions of different signal-to-noise ratios across spectral bands, leading to suboptimal performance. To fill this gap, we propose a transductive gradient learning framework that utilizes high signal-to-noise ratio bands to guide the inference of gradient patterns in low signal-to-noise ratio bands, substantially enhancing denoising effectiveness. Unlike existing approaches that only recover global low-rank structures, our framework introduces a transductive gradient injection regularization term to capture both global low-rank and local sparse gradient patterns. This term combines a low-rank matrix for global patterns and a sparse matrix for local patterns, leveraging pre-learned feature matrices from high signal-to-noise ratio band gradients to accurately inject spatial gradient textures, avoid excessive singular value constraints, and achieve efficient noise separation. Additionally, we have developed an efficient alternating direction method of multipliers algorithm for optimization. Extensive synthetic and real experiments on hyperspectral image datasets, along with applications in remote sensing, highlight significant performance gains. Across all datasets and noise conditions, our method achieves an average improvement of nearly 33% in overall peak signal-to-noise ratio and a 23% increase in spectral angle mapper compared to state-of-the-art hyperspectral image denoising methods.}, archive = {J_EAAI}, author = {Yuanyang Bu and Yongqiang Zhao and Jize Xue and Seong G. Kong and Jiaxin Yao and Jonathan Cheung-Wai Chan and Pan Liu and Xun Zhang}, doi = {10.1016/j.engappai.2024.109973}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109973}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transductive gradient injection for improved hyperspectral image denoising}, volume = {143}, year = {2025}, }
• (2025). Application of deep learning-based multimodal fusion technology in cancer diagnosis: A survey. EAAI, 143, 109972. (www)

  @article{ , abstract = {Relying solely on a single medical data for cancer diagnosis may increase the risk of misdiagnosis and missed diagnosis. Multi-modal data provides comprehensive information on disease characteristics and can effectively promote the development of precision oncology. This paper first introduces the genomic, pathological, radiological and clinical information in cancer multimodal data. Secondly, the common subfields of cancer multimodal data fusion are reviewed, with emphasis on data fusion techniques. The evolution of architectures under different fusion classes is compared, highlighting their comparative advantages and limitations. Importantly, we systematically reviewed the last five years of deep learning-based multimodal cancer data fusion, focusing on the application of multimodal techniques to cancer survival prediction and subtype typing. Finally, we present the challenges and possible solutions for multimodal applications in cancer. The purpose of this paper is to promote the fusion and application of multimodal tumor data and highlight potential research directions in the future.}, archive = {J_EAAI}, author = {Yan Li and Liangrui Pan and Yijun Peng and Xiaoyu Li and Xiang Wang and Limeng Qu and Qiya Song and Qingchun Liang and Shaoliang Peng}, doi = {10.1016/j.engappai.2024.109972}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109972}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Application of deep learning-based multimodal fusion technology in cancer diagnosis: A survey}, volume = {143}, year = {2025}, }
• (2025). Artificial intelligence driven laser parameter search: Inverse design of photonic surfaces using greedy surrogate-based optimization. EAAI, 143, 109971. (www)

  @article{ , abstract = {Photonic surfaces designed with specific optical characteristics are becoming increasingly crucial for novel energy harvesting and storage systems. The design of these surfaces can be achieved by texturing materials using lasers. The optimal adjustment of laser fabrication parameters to achieve target surface optical properties is an open challenge. Thus, we develop a surrogate-based optimization approach. Our framework employs the Random Forest algorithm to model the forward relationship between the laser fabrication parameters and the resulting optical characteristics. During the optimization process, we use a greedy, prediction-based exploration strategy that iteratively selects batches of laser parameters to be used in experimentation by minimizing the predicted discrepancy between the surrogate model’s outputs and the user-defined target optical characteristics. This strategy allows for efficient identification of optimal fabrication parameters without the need to model the error landscape directly. We demonstrate the efficiency and effectiveness of our approach on two synthetic benchmarks and two specific experimental applications of photonic surface inverse design targets. By calculating the average performance of our algorithm compared to other state of the art optimization methods, we show that our algorithm performs, on average, twice as well across all benchmarks. Additionally, a warm starting inverse design technique for changed target optical characteristics enhances the performance of the introduced approach.}, archive = {J_EAAI}, author = {Luka Grbcic and Minok Park and Juliane Müller and Vassilia Zorba and Wibe Albert de Jong}, doi = {10.1016/j.engappai.2024.109971}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109971}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence driven laser parameter search: Inverse design of photonic surfaces using greedy surrogate-based optimization}, volume = {143}, year = {2025}, }
• (2025). An improved traffic coordination control integrating traffic flow prediction and optimization. EAAI, 143, 109969. (www)

  @article{ , abstract = {The rapid growth of the number of vehicles and the inadaptability of signal control have become major factors restricting traffic efficiency and people’s travel experience. Optimizing traffic signal timing can alleviate congestion and reduce delays, but accurately predicting traffic flow for the next signal cycle remains a complex challenge. To address this, this paper proposes an analytical signal control optimization algorithm that integrates prediction and coordination at urban regional intersections to improve traffic efficiency. First, an intelligent learning scheme is designed, embedding real datasets into the Long Short-Term Memory (LSTM) network to predict traffic condition information in subsequent signal queues. Then, an objective optimization model is established based on kinematic wave theory and flow-density diagram. This model seeks the globally optimal signal timing solution by dynamically adjusting signal timing, cycle length, and phase splits. For traffic congestion scenarios, the phase configuration of multi-phase intersections is improved to enhance green light time utilization and traffic capacity. Taking the traffic operation scenario of Songwei South Road in Shanghai as an example, the simulation study verifies the performance of the proposed strategy.}, archive = {J_EAAI}, author = {Wei Su and Chaoxu Mu and Lei Xue and Xiaobao Yang and Song Zhu}, doi = {10.1016/j.engappai.2024.109969}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109969}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An improved traffic coordination control integrating traffic flow prediction and optimization}, volume = {143}, year = {2025}, }
• (2025). Meta learning-based deep reinforcement learning algorithm for task offloading in dynamic vehicular network. EAAI, 143, 109966. (www)

  @article{ , abstract = {The dramatic growth of the Internet of Vehicles (IoV) has led to an explosion in the number of latency-sensitive and computation-intensive tasks. Task offloading allows the migration of complex tasks from vehicle terminals to the Mobile Edge Computing (MEC) server or remote cloud server, thereby reducing their computational burden and energy consumption while improving the efficiency of task processing. However, since the high mobility of vehicles will cause dynamic changes in the network environment, how to perform efficient task offloading in IoV is a significant challenge. In this paper, we first propose a three-layer task offloading network architecture consisting of vehicles, multiple MEC servers and cloud server. Then, we formulate the task offloading decision problem with the aim of minimizing latency, energy consumption, and enhancing service quality. Finally, we model it as a markov decision process and propose a MEta learning-based Distributed Reinforcement learning Offloading (ME-DRO) algorithm to find the optimal offloading strategy. Simulation results show that compared with other baseline algorithms, the ME-DRO algorithm reduces the offloading latency and energy consumption by about 5.97% and 13.57%, respectively, and improves the service quality by about 20.17%. In addition, the algorithm has strong generalization ability when making real-time offloading decisions, and can quickly adapt to new tasks and environmental changes.}, archive = {J_EAAI}, author = {Liang Liu and Tengxiang Jing and Wenwei Li and Jie Duan and Wuping Mao and Huan Liu and Guanyu Liu}, doi = {10.1016/j.engappai.2024.109966}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109966}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Meta learning-based deep reinforcement learning algorithm for task offloading in dynamic vehicular network}, volume = {143}, year = {2025}, }
• (2025). Extended kalman filter-based maximum likelihood estimation for dynamic soft tissue characterisation. EAAI, 143, 109965. (www)

  @article{ , abstract = {Background and objective Dynamic characterisation of soft tissue properties plays a vital role in medical diagnosis and treatment, while it is a challenging issue due to sophisticated mechanical behaviours of soft tissue. This paper introduces a novel extended Kalman filter-based maximum likelihood (EKF-ML) method for dynamic soft tissue characterisation. Research gap Most of the existing studies assume soft tissue properties constant, ignoring the discrepancy between the model and experimental measurement. They also assume mechanical tool's kinematics exactly known without noise, leading to the difficulty in practical applications. Method The proposed method integrates extended Kalman filter (EKF) with maximum likelihood (ML) learning principle for dynamic identification of soft tissue properties based on online estimation of mechanical tool's kinematics. An EKF is developed based on the nonlinear Hunt–Crossley (H–C) model to estimate mechanical tool's motion, while the ML learning principle is established to evaluate the probability densities of soft tissue properties to address the discrepancy between the H–C model and experimental measurement. Results Simulation and experimental results show the proposed method can effectively identify soft tissue properties with an absolute mean deviation of 0.01N, leading to improved accuracy for soft tissue characterisation. Conclusion The proposed method addresses the discrepancy between the H–C model and experimental measurement while accounting for the unknown and noisy kinematics of mechanical tool. In addition to improved accuracy for soft tissue characterisation, it also overcomes the existing methods' limitation that requires prior knowledge of mechanical tool's motion, thus more suitable for practical applications.}, archive = {J_EAAI}, author = {Xinhe Zhu and Jiankun Li and Yongmin Zhong and Chengfan Gu and Kup-Sze Choi}, doi = {10.1016/j.engappai.2024.109965}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109965}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Extended kalman filter-based maximum likelihood estimation for dynamic soft tissue characterisation}, volume = {143}, year = {2025}, }
• (2025). Broad learning system-based model-free adaptive robust control for hypersonic morphing aircraft with appointed-time prescribed performance. EAAI, 143, 109962. (www)

  @article{ , abstract = {This paper proposes a model-free attitude tracking controller for hypersonic morphing aircraft with appointed-time prescribed performance. An error transformation methodology is developed, ensuring adherence to performance constraints irrespective of initial tracking conditions. Subsequently, an adaptive control scheme for the transformed system is devised, comprising two broad learning system-based adaptive neural network compensators and an adaptive robust integral of the signum of error (RISE) controller. The neural network compensators, featuring a node update strategy, are developed to compensate for uncertain system dynamics and unknown lumped disturbances. Subsequently, the adaptive RISE controller is integrated with the compensators to ensure precise tracking. The stability of the closed-loop system is analyzed by employing the Lyapunov’s direct method. Finally, the effectiveness and enhanced performance of the proposed control scheme are validated via numerical simulations.}, archive = {J_EAAI}, author = {Shihao Xu and Yingzi Guan and Changzhu Wei and Hechuan Xu}, doi = {10.1016/j.engappai.2024.109962}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109962}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Broad learning system-based model-free adaptive robust control for hypersonic morphing aircraft with appointed-time prescribed performance}, volume = {143}, year = {2025}, }
• (2025). An objective-guided multi-strategy evolutionary algorithm for multi-objective coalition formation. EAAI, 143, 109961. (www)

  @article{ , abstract = {The coalition formation (CF) problem is crucial for reasonably organizing agents with diverse and complementary capabilities to address complex scenarios in collaborative environments. While CF has received some research attention, the multi-objective coalition formation (MOCF) problem remains relatively unexplored and presents significant challenges. In the context of disaster relief and emergency response, this paper delves into the MOCF problem and constructs the mathematical model, which minimizes both the latest arrival time and the total cost of coalition members under mission-specific capability constraints. To tackle this, this paper proposes an innovative objective-guided multi-strategy evolutionary algorithm (OGMSEA) for effective capability aggregation regarding mission requirements and objective trade-offs, which leverages the problem characteristics of multiple objectives and lower-bound constraints. The initialization strategies leverage various objective weights to generate a uniformly distributed and extensive set of initial solutions. The repair strategies restore unsatisfied coalitions by evaluating the alignment of idle agents with the remaining capability requirements and optimization objectives. The restart strategies reconstruct repetitive solutions to maintain the population diversity. Comprehensive experiments demonstrate OGMSEA’s superior performance in terms of applicability and adaptability, better achieving inverted generational distance and hypervolume metrics across 135 various cases compared with advanced algorithms. In large-scale complex scenarios (e.g., more than 100 agents, 10 missions, and a demand-supply ratio on capabilities of 0.5), OGMSEA consistently achieves a high-quality Pareto front due to its well-designed strategies. Additionally, a forest fire scenario is constructed and addressed by forming firefighting coalitions, demonstrating the practical applicability of this study.}, archive = {J_EAAI}, author = {Miao Guo and Bin Xin and Jie Chen and Shuxin Ding}, doi = {10.1016/j.engappai.2024.109961}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109961}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An objective-guided multi-strategy evolutionary algorithm for multi-objective coalition formation}, volume = {143}, year = {2025}, }
• (2025). Local interpretation of deep learning models for aspect-based sentiment analysis. EAAI, 143, 109947. (www)

  @article{ , abstract = {Currently, deep learning models are commonly used for Aspect-Based Sentiment Analysis (ABSA). These deep learning models are often seen as black boxes, meaning that they are inherently difficult to interpret. To improve deep learning models, it is crucial to understand their inner workings. We aim to interpret black box models by implementing model-agnostic local interpretation methods. Inspired by Local Interpretable Model-agnostic Explanations (LIME) and Local Rule-based Explanations (LORE) and combined with a Similarity-based Sampling (SS) method, we propose SS-LIME and SS-LORE, and use Anchor to explain two state-of-the-art ABSA deep learning models. The deep learning models build upon the Left-Center-Right separated neural network with Rotatory attention (LCR-Rot) model, extended by iterating multiple times over the rotatory attention mechanism (LCR-Rot-hop) and hierarchical attention and context-dependent word embeddings (LCR-Rot-hop++). We evaluate the proposed models in terms of fidelity, hit rate, and user interpretability using the SemEval 2016 dataset consisting of restaurant reviews for ternary sentiment classification. Results show that the LCR-Rot-hop and LCR-Rot-hop++ models are best explained by SS-LIME and SS-LORE, respectively. Furthermore, we conclude that the LCR-Rot-hop++ model can be better interpreted than the LCR-Rot-hop model.}, archive = {J_EAAI}, author = {Stefan Lam and Yin Liu and Max Broers and Jasper van der Vos and Flavius Frasincar and David Boekestijn and Finn van der Knaap}, doi = {10.1016/j.engappai.2024.109947}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109947}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Local interpretation of deep learning models for aspect-based sentiment analysis}, volume = {143}, year = {2025}, }
• (2025). Intelligent modeling and analysis of hybrid organic rankine plants: Data-driven insights into thermodynamic efficiency and economic viability. EAAI, 143, 109946. (www)

  @article{ , abstract = {Hybrid organic Rankine cycle (HORC) is a hydrodynamic plant used from industrial processes for low-temperature heat sources, such as geothermal, solar, and waste heat. Intelligent models were developed to predict the first and the second thermodynamic efficiencies and the levelized energy cost to optimize the overall thermal and economic efficiency of hybrid organic Rankine cycle-powered plants. Deep learning, gradient-boosting framework, and Kernel models such as Long Short-Term Memory (LSTM), Light Gradient Boosting Machine (LGBM), and Kernel Ridge Regression (KRR) models were developed to predict the three outputs of HORC according to five subsets: subset-1: design variables, subset-2: temperature variables, subset-3: power variables, subset-4: heat exchanger variables, and subset-5: all previous variables. The LSTM model generally achieved superior performance across the multiple input variables used in predicting the three model outputs. The LSTM model attained the lowest mean absolute percentage error (MAPE) (4.8%–13%), the highest coefficient of determination (R 2 ) (up to 0.994), and the lowest root mean square error (RMSE) as low as 0.002. This demonstrated superior predictive accuracy across the various model input subsets. The LGBM model, however, showed moderate performance, with the MAPE reaching up to 25.6% and the R 2 ranging from 0.624 to 0.986. In contrast, the KRR model struggled to demonstrate exceptional performance, especially with the heat exchanger dataset, thus exhibiting a MAPE up to 54.4% and an R 2 value as low as 0.408. Therefore, we advocate that the LSTM model could be the most reliable model for predicting the system efficiency of HORC.}, archive = {J_EAAI}, author = {Hai Tao and Mohammed Suleman Aldlemy and Mohammed Ayad Saad and Swee Pin Yeap and Atheer Y. Oudah and Omer A. Alawi and Leonardo Goliatt and Shamsad Ahmad and Zaher Mundher Yaseen and Ravinesh C. Deo}, doi = {10.1016/j.engappai.2024.109946}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109946}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent modeling and analysis of hybrid organic rankine plants: Data-driven insights into thermodynamic efficiency and economic viability}, volume = {143}, year = {2025}, }
• (2025). Lightweight train image fault detection model based on location information enhancement. EAAI, 143, 109941. (www)

  @article{ , abstract = {In addition to the imperative for accurate detection, the computational and storage costs of defect detection models are often considered in computing environments with limited resources, such as mobile devices and edge nodes. Consequently, a lightweight fault detection network model (Light-LIENet), based on location information enhancement, is proposed. It is capable of achieving real-time and high-precision detection while minimizing computational overhead and parameter complexity. Light-LIENet uses a convolutional neural network as its foundational architecture for extracting pertinent image features. Furthermore, it integrates a location feature network (LFNet) to accomplish multiscale feature fusion and enhance object location information. LFNet includes a location sensitive block, a feature enhancement block, and a lightweight convolution block. These blocks collectively contribute to enhancing location information, augmenting feature extraction, and minimizing computational amount. Experimental results show that, when compared to state-of-the-art lightweight detectors on the freight train image fault dataset, Light-LIENet exhibits impressive performance. With a detection speed of 62 frames per second, a mere 1.611 million model parameters, and a computational requirement of just 3.984 G, it unequivocally demonstrates its effectiveness and practicality.}, archive = {J_EAAI}, author = {Longxin Zhang and Yang Hu and Runti Tan and Wenliang Zeng and Jianguo Chen}, doi = {10.1016/j.engappai.2024.109941}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109941}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight train image fault detection model based on location information enhancement}, volume = {143}, year = {2025}, }
• (2025). Monitoring the deterioration state of braking friction performance of high-speed train under imbalanced data. EAAI, 143, 109939. (www)

  @article{ , abstract = {To solve the problem of poor friction block classification when the number of samples of different degradation states is imbalanced in the braking process of high-speed trains, a friction block deterioration state monitoring model based on generative adversarial network (GAN) and lightweight network is proposed in this study. First, the vibration signal of the friction block is transformed into a two-dimensional heat map to enhance the distinguishability between different types of signals. Second, the proposed method is used to expand the samples under different friction blocks. The modified MobileNetv3 (MMN3) lightweight network framework is then used to construct the monitoring model of the friction block degradation state, reducing the number of parameters while maintaining high precision recognition performance. Finally, the proposed method is applied to the imbalanced brake pad friction block data obtained from the self-developed high-speed train test bench. The proposed model can maintain high recognition accuracy in different degrees of friction block wear state imbalance proportion datasets. It can effectively monitor the deterioration state of high-speed train friction blocks under imbalanced data.}, archive = {J_EAAI}, author = {Min Zhang and Wenxin Xu and Jiliang Mo and Zaiyu Xiang and Zhongrong Zhou}, doi = {10.1016/j.engappai.2024.109939}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109939}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Monitoring the deterioration state of braking friction performance of high-speed train under imbalanced data}, volume = {143}, year = {2025}, }
• (2025). Sustainable benchmarking of e-scooter micromobility systems: A hybrid q-rung orthopair fuzzy score function and distance measure-based ranking approach. EAAI, 143, 109934. (www)

  @article{ , abstract = {Micromobility is an innovative urban transport solution that can effectively tackle greenhouse gases and reduce the use of private vehicles, especially for short-distance travel options. As a type of shared micromobility service, the electric scooter (e-scooter) is designed to provide convenient and quick rides for short distances. The constantly growing adoption of these emerging transportation technologies brings many factors to be considered. In this paper, we evaluate the available e-scooters and identify the most appropriate choice with respect to several factors. To this aim, we propose a decision support tool with the combination of modified symmetry point of criterion (SPC), the pivot pairwise relative criteria importance assessment (PIPRECIA), and the weighted aggregated sum product assessment (WASPAS) approaches under the setting of q-rung orthopair fuzzy (q-ROF) sets. First, the developed method calculates the significance values of decision experts through a new q-ROF-score function. Next, the criteria weights are determined using an incorporated objective-subjective weighting model. It comprises a q-ROF-modified SPC tool for objective weighting and a q-ROF-PIPRECIA tool for subjective weighting. Subsequently, we present the stepwise mathematical algorithm of a q-ROF-WASPAS approach using q-ROF aggregation operators and criteria weighting tools. To verify the compatibility and effectiveness of the introduced model, it is applied to evaluate and prioritize the e-scooters under the q-ROF environment. Sensitivity analysis is made to examine the impact of varying values of weighting and utility parameters. The efficiency of the introduced approach is confirmed by different comparative investigations.}, archive = {J_EAAI}, author = {Arunodaya Raj Mishra and Pratibha Rani and Ahmad M. Alshamrani and Adel Fahad Alrasheedi and Vladimir Simic}, doi = {10.1016/j.engappai.2024.109934}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109934}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Sustainable benchmarking of e-scooter micromobility systems: A hybrid q-rung orthopair fuzzy score function and distance measure-based ranking approach}, volume = {143}, year = {2025}, }
• (2025). A comprehensive review of simulation optimization methods in agricultural supply chains and transition towards an agent-based intelligent digital framework for agriculture 4.0. EAAI, 143, 109930. (www)

  @article{ , abstract = {Simulation-based optimization is a potent methodology that synergises simulation modelling with optimization techniques to enhance decision-making and system performance within complex contexts. This comprehensive review underscores the pivotal role of supply chain management in agriculture and elucidates the pivotal role of simulation-based optimization in optimizing processes, mitigating waste, and curbing supply chain vulnerabilities. The review not only identifies prevalent trends and patterns in simulation-based optimization applications within agricultural supply chains but also unearths existing gaps. Beyond analyzing the evolution of traditional-to-modern trends in supply chain problem-solving, this article delves into contemporary simulation-based optimization approaches. Noteworthy among these are agent-based simulation techniques for the simulation facet and artificial intelligence methodologies, including machine learning, for optimization endeavours. In a culminating stride, the article introduces a conceptual framework for a digital and intelligent agricultural supply chain, seamlessly integrating agent-based simulation and reinforcement learning strategies. Through optimization grounded in simulation, this digital twin engenders supply chain designs that hold substantial practicality and efficacy, particularly in the context of Agriculture 4.0. This review amalgamates insights from simulation-based optimization and agricultural supply chain management, culminating in a forward-looking framework poised to fortify the capabilities of Agriculture 4.0.}, archive = {J_EAAI}, author = {Elham Shadkam and Elnaz Irannezhad}, doi = {10.1016/j.engappai.2024.109930}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109930}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A comprehensive review of simulation optimization methods in agricultural supply chains and transition towards an agent-based intelligent digital framework for agriculture 4.0}, volume = {143}, year = {2025}, }
• (2025). Physics descriptors enhanced bayesian learning method for permeability of random media under sparse data. EAAI, 143, 109929. (www)

  @article{ , abstract = {Permeability is a significant property in microstructure-based material design. Currently, the main research gap in such material design can be divided into three aspects: Firstly, experimental methods face stringent conditions and poor reproducibility in accurately describing internal media structures. Secondly, machine learning methods require impractically large datasets to learn the strong nonlinear relationship between permeability and microstructure. Thirdly, there is a challenge in quantifying the inherent uncertainty of randomly distributed phases and the uncertainty propagation in the model. In this work, we propose a novel physics descriptors enhanced Bayesian learning method, aiming to predict the permeability of random media under sparse data. The key aspects of the method are the physics descriptors with clear physical indications, which characterize porous microstructure from multiple perspectives, and an encoder–decoder strategy that reduces microstructural complexity while utilizing these physics descriptors as basis functions to reconstruct permeability. The numerical experiments indicate that porosity, chord length density, and lineal-path dominate in determining permeability. Moreover, predictions with a 5%–11% error can be achieved under 32 training samples. Finally, the method provides a complete posterior distribution, where the posterior variances quantify uncertainty and benefit robust decisions in simulation-assisted materials design.}, archive = {J_EAAI}, author = {Hang Qi and Xiaofei Guan and Qing Chen and Zhengwu Jiang and Fang Liu and Jieqiong Zhang and Hehua Zhu}, doi = {10.1016/j.engappai.2024.109929}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109929}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics descriptors enhanced bayesian learning method for permeability of random media under sparse data}, volume = {143}, year = {2025}, }
• (2025). MAViT: A lightweight hybrid model with mutual attention mechanism for driver behavior recognition. EAAI, 143, 109921. (www)

  @article{ , abstract = {Driver behavior plays a critical role in transportation safety. Studies show that over half of traffic accidents are caused by distracted driving, making it essential to accurately recognize various driver behaviors. These behaviors can be reflected by their overall postures and local details, both of which need to be considered. A new lightweight hybrid model, Mutual Attention Vision Transformer (MAViT), is proposed in this paper with a mutual attention mechanism to recognize these behaviors. In this model, a multi-scale inverse residual (MSIR) block is designed to extract key features of different scales from the driving image. In addition, the local and global mutual attention module (LGMA) is proposed, providing stronger representation capabilities in driving scenarios by establishing the mutual influence relationship between local features and global features. The proposed model utilizes only 0.2 million parameters and achieves accuracy rates of 99.64% and 99.40% on the State Farm Distracted Driver Detection (SFD3) and the American University in Cairo Distracted Driver v2 (AUCv2) public datasets, respectively. This surpasses many other state-of-the-art methods.}, archive = {J_EAAI}, author = {Haibin Sun and Yujie Ma}, doi = {10.1016/j.engappai.2024.109921}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109921}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MAViT: A lightweight hybrid model with mutual attention mechanism for driver behavior recognition}, volume = {143}, year = {2025}, }
• (2025). Imitation learning of a model predictive controller for real-time humanoid robot walking. EAAI, 143, 109919. (www)

  @article{ , abstract = {Bipedal walking is an especially challenging task for humanoid robots, as it requires a robust controller to make a humanoid robot walk stably and react to disturbances. State-of-the-art algorithms have made use of model predictive controllers, but they require an exceedingly high computational cost and are often impractical to embed in a real robot’s hardware. This paper contributes by showing how imitation learning may be employed to copy the behavior of a model predictive controller to a neural network, being able to predict the trajectory of the Center of Mass (CoM), as well as the position, rotation, and duration of each step. Our method was tested on a simplified simulation, a realistic full-body simulation, and on a real humanoid robot. The results showcase an algorithm that is close in terms of performance to the original controller, while requiring only a small fraction of its computational cost — with a speedup of 650 times, enabling it to be used in real time on real robots.}, archive = {J_EAAI}, author = {Vítor G.B. de A. Porto and Dicksiano C. Melo and Marcos R.O.A. Maximo and Rubens J.M. Afonso}, doi = {10.1016/j.engappai.2024.109919}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109919}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Imitation learning of a model predictive controller for real-time humanoid robot walking}, volume = {143}, year = {2025}, }
• (2025). Multi-objective multi-path COVID-19 medical waste collection problem with type-2 fuzzy logic based risk using partial opposition-based weighted genetic algorithm. EAAI, 143, 109916. (www)

  @article{ , abstract = {The transportation and handling of the most infectious coronavirus disease 2019 (COVID-19) waste are vital from the point of view of risk and safety. Here, a multi-objective, multi-path medical waste collection routing problem with occupational and transportation risks is considered. A specially equipped medical collection van with appropriately dressed collectors starts from a dumping campus and returns to it for disposal after collecting the COVID-19 hazardous wastes from rural medical centers/hospitals. Transportation, fixed charge, and loading costs are considered in this system. The occupational risk in each medical center depends on the number of workers involved, the collected amount of hazardous waste, and the storage period. The transportation risk depends on the probability of an accident which is determined by Type-1 and Type-2 fuzzy logic, the environmental condition of the accident, the length of the route, the weight of carried waste, and the population around the accident site which are the key applications of artificial intelligence. Here, the objective is to minimize the total cost, occupational, and transportation risks. This multi-objective problem is converted to a single-objective one through the weighted sum method. To solve it, we develop and implement a genetic algorithm with partial opposition-based learning-dependent initialization and mutation, probabilistic selection, and weightage-based comparison crossover. The proposed algorithm is tested against some standard benchmark instances, its effectiveness is shown through statistical tests and performance metrics. From the numerical experiments, it is observed that total cost is inversely related to both risks, but their behavior is non-linear. Some managerial insights are presented.}, archive = {J_EAAI}, author = {Somnath Maji and Samir Maity and Debasis Giri and Izabela Nielsen and Manoranjan Maiti}, doi = {10.1016/j.engappai.2024.109916}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109916}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective multi-path COVID-19 medical waste collection problem with type-2 fuzzy logic based risk using partial opposition-based weighted genetic algorithm}, volume = {143}, year = {2025}, }
• (2025). Exponential neutrosophic programming contrivance in multiple criteria for type-2 uncertain multi-objective transportation problem. EAAI, 143, 109903. (www)

  @article{ , abstract = {Exploration of type-2 uncertainty in decision-making contexts is ongoing, and this article takes a pioneering approach by introducing the integration of type-2 normal uncertain parameters in the formulation of a multi-objective transportation problem. The transformation of the type-2 normal uncertain model into a deterministic one is achieved through methods based on expected value chance constraint programming. The problem is further investigated within a neutrosophic framework, where exponential membership functions with optimistic, pessimistic, and mixed variations are considered. The Pareto optimal solutions are obtained using truth, indeterminacy, and falsity membership functions in confluence with exponential neutrosophic programming techniques, including optimistic, pessimistic, and mixed senses. These techniques offer a flexible approach, empowering the decision-maker to select confidence levels according to their priorities. They can then analyze the Euclidean distance of the obtained optimal values of the objective functions, enhancing their understanding of the results. The benefits of employing our suggested strategies in a practical sense are clearly demonstrated via numerical experimentation, and finally, a comparison of the proposed approaches is also shown based on Euclidean distance. Furthermore, MATLAB software (version R2023b, 64-bit (maci64)) is utilized to develop the code to solve the suggested issue.}, archive = {J_EAAI}, author = {Anesh Kumar and Pitam Singh}, doi = {10.1016/j.engappai.2024.109903}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109903}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exponential neutrosophic programming contrivance in multiple criteria for type-2 uncertain multi-objective transportation problem}, volume = {143}, year = {2025}, }
• (2025). Rethinking the message passing for graph-level classification tasks in a category-based view. EAAI, 143, 109897. (www)

  @article{ , abstract = {Message-Passing Neural Networks (MPNNs) have emerged as a popular framework for graph representation in recent years. However, the graph readout function in MPNNs often leads to significant information loss, resulting in performance degradation and computational waste for graph-level classification tasks. Despite the common explanation of “local information loss,” the underlying essence of this phenomenon and the information that the MPNN framework can capture in graph-level tasks have not been thoroughly analyzed. In this paper, we present a novel analysis of the MPNN framework in graph-level classification tasks from a node category-based perspective. Our analysis reveals that the graph-level embeddings learned by MPNNs essentially correspond to category-based contribution measurements. Building upon this insight, we propose a groundbreaking Category-Based Non-Message-Passing (CANON) paradigm for graph-level representation learning. By leveraging a novel numerical encoding mechanism, CANON achieves superior performance even without incorporating structural information, surpassing state-of-the-art MPNN methods. CANON also offers substantial computational advantages, including a model size that is hundreds of times smaller and reduced time complexity, enabling faster inference and reduced costs for real-world applications, particularly in domains such as chemistry, biology, and computer vision. To further enhance the method’s effectiveness, we introduce domain-specific structural incorporation. Our extensive experiments across multiple datasets demonstrate CANON’s efficacy and its potential to serve as a highly efficient alternative to MPNNs, opening up new possibilities for graph representation learning and its downstream applications.}, archive = {J_EAAI}, author = {Lei Han and Xu Jiaxing and Ni Jinjie and Ke Yiping}, doi = {10.1016/j.engappai.2024.109897}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109897}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rethinking the message passing for graph-level classification tasks in a category-based view}, volume = {143}, year = {2025}, }
• (2025). Efficient safe learning for controller tuning with experimental validation. EAAI, 143, 109894. (www)

  @article{ , abstract = {Optimization-based controller tuning is challenging because it requires formulating optimization problems explicitly as functions of controller parameters. Safe learning algorithms overcome the challenge by creating surrogate models from measured data. To ensure safety, such data-driven algorithms often rely on exhaustive grid search, which is computationally inefficient. In this paper, we propose a novel approach to safe learning by formulating a series of optimization problems instead of a grid search. We also develop a method for initializing the optimization problems to guarantee feasibility while using numerical solvers. The performance of the new method is first validated in a simulated precision motion system, demonstrating improved computational efficiency, and illustrating the role of exploiting numerical solvers to reach the desired precision. Experimental validation on an industrial-grade precision motion system confirms that the proposed algorithm achieves 30% better tracking at sub-micrometer precision as a state-of-the-art safe learning algorithm, improves the default auto-tuning solution, and reduces the computational cost seven times compared to learning algorithms based on exhaustive search.}, archive = {J_EAAI}, author = {Marta Zagorowska and Christopher König and Hanlin Yu and Efe C. Balta and Alisa Rupenyan and John Lygeros}, doi = {10.1016/j.engappai.2024.109894}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109894}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient safe learning for controller tuning with experimental validation}, volume = {143}, year = {2025}, }
• (2025). Optimization of built-in self-test test chain configuration in 2.5D integrated circuits using constrained multi-objective evolutionary algorithm. EAAI, 143, 109876. (www)

  @article{ , abstract = {2.5D Integrated Circuit (2.5D IC) is an advanced packaging technology. This technology facilitates the dense integration of multiple dies by adding passive components to the silicon interposer. However, the high-density stacking of dies has increased the complexity of IC. Its highly interconnected nature presents a series of challenges for IC testing. As for 2.5D IC testing utilizing the Built-In Self-Test(BIST) architecture, a dual-objective optimal model is established for test chain configuration. The model aims to minimize both hardware testing costs and test time, considering constraints related to maximum power consumption and maximum testing time. Subsequently, A generic encoding and decoding method is devised to enable the direct application of existing Multi-Objective Evolutionary Algorithms (MOEAs) and Constrained Multi-Objective Evolutionary Algorithms (CMOEAs) to solve the test chain configuration problem. Simultaneously, a dual-stage dual-population CMOEA is devised, integrating diverse local search strategies and adaptively adjusting them based on a successful incentive mechanism. Additionally, a constraint handling framework is introduced, utilizing the relationship between the Constrained Pareto Front (CPF) and Unconstrained Pareto Front (UPF) to assist in searching for solutions that satisfy constraints. This framework also adaptively allocates computational resources by assessing the collaborative effects of auxiliary tasks. Multiple experiments are designed and conducted to demonstrate the correctness and effectiveness of the proposed model and algorithm. Comparative analysis with 8 MOEAs or CMOEAs reveals that the proposed algorithm outperforms in solving the test chain configuration problem.}, archive = {J_EAAI}, author = {Zhe Yang and Libao Deng and Chunlei Li and Lili Zhang}, doi = {10.1016/j.engappai.2024.109876}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109876}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimization of built-in self-test test chain configuration in 2.5D integrated circuits using constrained multi-objective evolutionary algorithm}, volume = {143}, year = {2025}, }
• (2025). Proposal of a sequential machine learning modelling approach for optimal cementitious composites. EAAI, 143, 109837. (www)

  @article{ , abstract = {This paper presents an innovative approach to designing and manufacturing optimal composites. The proposed approach is based on a new concept of sequential machine learning (ML) that is proposed to enhance accuracy and reliability whilst overcoming the common bottleneck of existing ML approaches related to datasets of limited size. The sequential ML modelling is shown to create a synergistic analytical pipeline for precise reverse engineering. An application is made to estimate the modulus of rupture (MoR) of steel fibre-reinforced concrete (SFRC). Compared to the existing ML-assisted MoR modelling, the sequential approach enables significantly higher inclusion of literature data (>1000 mixes) in the training process to reach a higher accuracy (error of 0.8 MPa) with relevant stability. A multi-objective optimisation (MOO) framework with a population-based metaheuristic enabled the design of optimal concrete mixes, verified for strength against advanced numerical simulations. The optimised mixes showed up to two times superior performance for production cost and environmental footprint to those in the collected dataset.}, archive = {J_EAAI}, author = {Elyas Asadi Shamsabadi and Saeed Mohammadzadeh Chianeh and Daniel Dias-da-Costa}, doi = {10.1016/j.engappai.2024.109837}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109837}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Proposal of a sequential machine learning modelling approach for optimal cementitious composites}, volume = {143}, year = {2025}, }
• (2025). Human fall detection using pose estimation: From traditional machine learning to vision transformers. EAAI, 143, 109809. (www)

  @article{ , abstract = {Human activity recognition research for healthcare has drawn global attention in recent era. Recent advancements have led to various approaches capable of detecting diverse movements like walking, running, jumping, and falling. Fall detection is crucial due to its potential fatality, especially for older individuals. Sensors are widely employed to perceive environmental changes, and they can be integrated into wearable devices like phones, necklaces, or wristbands. However, these devices may be uncomfortable or unsuitable for continuous use. Video imagery, in principle, surpasses wearable sensors for fall detection. The proposed method uses video frames to identify falls, reducing the need for environmental sensors. We present an empirical analysis of vision-based human fall detection, employing multiple techniques to estimate human poses including a transformer-based pose estimation technique. These techniques yield foundational features used for training diverse networks, including machine learning classifiers to vision transformers. Our methodology achieves cutting-edge outcomes across the UR-Fall, UP-Fall, and Le2i fall detection datasets.}, archive = {J_EAAI}, author = {Ali Raza and Muhammad Haroon Yousaf and Waqar Ahmad and Sergio A. Velastin and Serestina Viriri}, doi = {10.1016/j.engappai.2024.109809}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109809}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Human fall detection using pose estimation: From traditional machine learning to vision transformers}, volume = {143}, year = {2025}, }
• (2025). Artificial intelligence in healthcare and medicine technology development review. EAAI, 143, 109801. (www)

  @article{ , abstract = {This paper deals with the need for artificial intelligence (AI) such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Natural Language Processing (NLP), Computer Vision (CV), application technologies and fields, and key developments of specialized companies that are completely changing the experience of doctors, patients, and healthy ordinary people in healthcare and medical fields. At the end of this article, AI-based healthcare and medical technology that can innovatively improve medical professionals' examination errors due to differences in data acquisition and analysis or proficiency of existing medical equipment. First, historical development is mentioned chronologically, and detailed contributions to doctors, patients, and the general public in terms of medical image precision diagnosis, patient diagnosis and treatment, monitoring day and operation management, and new drug discovery and development are highlighted. AI technolgoies and their applications in radiology, dermatology, hematology, ophthalmology, oncology, cardiovascular, gastroenterology, pathology, musculoskeletal and electro-health record-based treatment process efficiency and new drug development are discussed. It also analyzes AI-based solutions and detailed application technologies from six U.S. companies specializing in healthcare and medical care, Elitic, Butterfly Network, Merative, Viz.ai, Path AI, and BioXel Therapeutics, and four Korean companies JLK, VUNO, Lunit and Deepnoid. In conclusion, AI has significant potential to contribute to addressing ongoing issues of medical professional shortage and improving the quality of medical service. However, there are challenges in terms of data privacy, security, safety and accuracy, integration with existing systems, acceptance and trust of doctors, prejudice and ethics, and autonomy and they should be discussed in depth in the future.}, archive = {J_EAAI}, author = {Daeun Chung}, doi = {10.1016/j.engappai.2024.109801}, journal = {Engineering Applications of Artificial Intelligence}, month = {3}, pages = {109801}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence in healthcare and medicine technology development review}, volume = {143}, year = {2025}, }
• (2025). Causality-inspired surface defect detection by transferring knowledge from natural images. EAAI, 142, 109984. (www)

  @article{ , abstract = {Deep neural networks have become a mainstream method for dyed fabric surface defect detection, which requires a substantial number of training images to exploit its performance. However, building large-scale datasets of dyed fabric surface defects is difficult due to the scarcity of defects, which leads to deep networks that are often under-trained and overfitted in the training set. In this paper, we attempt to mitigate model overfitting caused by insufficient defect samples by transferring the knowledge learned from natural images. Specifically, we explore the correlation between natural image salient object detection (SOD) and defect image detection, utilizing multi-domain learning to jointly train natural and defect images. To mitigate the biased behavior in model learning caused by the differences between natural and defect images, we design a causality-based removal framework to find the true causal relationships between defects and contexts. To facilitate integrity for multi-domain learning, we propose a diverse synergy module (DSM) based on foreground, background, and global feature fusion. The DSM is used to enhance the ability of the network to comprehensively perceive defect regions. In addition, we propose a multi-convolutional aggregation module (MCAM) to enhance feature diversity by aggregating features with different receptive fields. Extensive experiments on our proposed fabric dyeing defects dataset (FDD) and three of the most widely used surface defects datasets show that our proposed method achieves state-of-the-art results and has good generalization ability. Our method shows remarkable detection accuracy even with extremely limited training samples. Our dataset and code will be available at https://github.com/DEF-21/UMDNet .}, archive = {J_EAAI}, author = {Fangfang An and Shaolei Cao and Shuai Ma and Dawu Shu and Bo Han and Wanxin Li and Ruigang Liu}, doi = {10.1016/j.engappai.2024.109984}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109984}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Causality-inspired surface defect detection by transferring knowledge from natural images}, volume = {142}, year = {2025}, }
• (2025). Removing visual occlusion of construction scaffolds via a two-step method combining semantic segmentation and image inpainting. EAAI, 142, 109983. (www)

  @article{ , abstract = {With increasing computer vision (CV) applications in automated construction management, the visual occlusion issue caused by crisscrossing, wide-coverage, and immovable scaffolds has become one of the most challenging. This study proposes a novel deep learning-based two-step method combining pixel-level semantic segmentation and contextual image inpainting to remove scaffolds visually and restore the occluded visual information. A low-cost data synthesis method using only unlabeled data has also been developed to alleviate the shortage of labeled data for deep neural network (DNN) training. Experiments on the synthesized test data show that the proposed method achieves performances of 92% mean intersection over union (MIoU) for scaffold segmentation and over 82% structural similarity (SSIM) for scene restoration after removing scaffolds. This research set a precedent for addressing the visual occlusion issue of scaffolds, and the proposed method is verified in real-world cases that it helps existing CV models perform better in scaffolding scenarios.}, archive = {J_EAAI}, author = {Yuexiong Ding and Muyang Liu and Ming Zhang and Xiaowei Luo}, doi = {10.1016/j.engappai.2024.109983}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109983}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Removing visual occlusion of construction scaffolds via a two-step method combining semantic segmentation and image inpainting}, volume = {142}, year = {2025}, }
• (2025). Few-shot remaining useful life prediction based on bayesian meta-learning with predictive uncertainty calibration. EAAI, 142, 109980. (www)

  @article{ , abstract = {Deep learning-based methods have been developed and widely used for predicting remaining useful life (RUL), a critical task in predictive maintenance aimed at minimizing machine downtime and optimizing maintenance schedules. These methods typically rely on substantial degradation data. However, in some real-world scenarios, degradation data are extremely limited. Meta-learning, a prominent few-shot learning method, leverages degradation data from auxiliary tasks to facilitate predictions in the target tasks. While meta-learning has improved prediction accuracy, there is inevitably uncertainty in predictions. To provide decision-makers with more reliable information, accurately quantifying predictive uncertainty is crucial. However, in meta-learning, both quantifying and calibrating predictive uncertainty are challenging due to data scarcity. This paper proposes a few-shot learning method for RUL prediction, named Bayesian model-agnostic meta-learning with predictive uncertainty calibration (BMLPUC), incorporating an uncertainty calibration term into the objective function for the model training. Additionally, the meta-training is optimized using two adaptive hyper-parameters according to the model performance. The effectiveness of the proposed method is validated using two bearing datasets. Superior prediction accuracies and more accurately calibrated predictive uncertainty compared to the baseline and three other state-of-the-art methods are achieved by the proposed method.}, archive = {J_EAAI}, author = {Liang Chang and Yan-Hui Lin}, doi = {10.1016/j.engappai.2024.109980}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109980}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Few-shot remaining useful life prediction based on bayesian meta-learning with predictive uncertainty calibration}, volume = {142}, year = {2025}, }
• (2025). Unsupervised learning for motion correction and assessment in brain magnetic resonance imaging using severity-based regularized cycle consistency. EAAI, 142, 109978. (www)

  @article{ , abstract = {Magnetic resonance imaging is an important non-invasive diagnostic tool, yet is vulnerable to motion artifacts due to the long acquisition time. With the recent development of deep learning, numerous methods for improving motion artifacts have been proposed. Due to the difficulty in constructing paired datasets for motion artifact correction, various unsupervised learning-based models have emerged recently. Cycle-Consistent Generative Adversarial Network (CycleGAN), which is widely used as a base model for unsupervised learning, has ill-posed problem and high parameter complexity that poses limitations towards an unstable training process. To overcome these limitations, this paper proposes an unsupervised learning method for motion artifact correction and assessment by replacing one generator of CycleGAN with a motion artifact simulator and regularizing the cycle-consistency loss with a motion severity prior. In adopting this approach, the overall model's training complexity can be significantly reduced due to the decrease in the quantity of learnable parameters. Moreover, the calculation of the cycle-consistency loss becomes more stable, thereby addressing the inherent ill-posed problem of CycleGAN. As a result, the proposed method demonstrates superior performance quantitatively and qualitatively compared to existing motion artifact correction methods and succeeds in reconstructing clearer images qualitatively compared to supervised learning methods.}, archive = {J_EAAI}, author = {Seuk Kim and Mohammed A. Al-masni and Seul Lee and Sunyoung Jung and Kyu-Jin Jung and Chuanjiang Cui and Sung-Min Gho and Young Hun Choi and Dong-Hyun Kim}, doi = {10.1016/j.engappai.2024.109978}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109978}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised learning for motion correction and assessment in brain magnetic resonance imaging using severity-based regularized cycle consistency}, volume = {142}, year = {2025}, }
• (2025). Dual-path information enhanced pyramid unet for COVID-19 lung infection segmentation. EAAI, 142, 109977. (www)

  @article{ , abstract = {The coronavirus disease 2019 (COVID-19) pandemic has brought computer-aided diagnosis into the spotlight. COVID-19 computed tomography (CT) images often have redundant background, and the proportion of infected area and healthy areas is unbalanced, which may confuse the model and make it difficult to identify the infected areas, resulting in the inability of the model to make accurate determinations. The low contrast between infected and normal areas makes the boundaries difficult to distinguish. The infected areas are small, irregular and unevenly distributed. Thus, models often suffers from incomplete and inadequate segmentation. To solve these challenges, a dual-path information enhanced pyramid Unet network (DIEP-Unet) was proposed for COVID-19 infection segmentation. First, an approach of coarse multi-scale feature map (CMFM) was proposed, which included a Fourier image boundary enhancement (FIBE) module to enhance the boundary features of infected areas, and a multi-scale information extraction and fusion (MSIE) module to locate the position of the infected areas. Second, the hybrid attention global context awareness (HAGCA) module collected hybrid attention information and multi-scale features from different branches, which can better aggregate information between the encoder and the decoder. Experimental results revealed that the proposed method achieves an accuracy of 0.9982 and a specificity of 0.9989. Numerical examples and ablation studies demonstrate that the proposed DIEP-Unet achieves high accuracy and outperforms existing segmentation methods. Therefore, the proposed method has the potential applications in the detection, localization, and labeling of other diseased areas.}, archive = {J_EAAI}, author = {Yan Zhang and Qi Mao and Yi Tian and Wenfeng Wang and Lijia Ren and Haibo Li}, doi = {10.1016/j.engappai.2024.109977}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109977}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-path information enhanced pyramid unet for COVID-19 lung infection segmentation}, volume = {142}, year = {2025}, }
• (2025). Identification and validation of key predictive factors for heart attack diagnosis using machine learning and fuzzy clustering. EAAI, 142, 109968. (www)

  @article{ , abstract = {Heart disease remains a leading cause of mortality worldwide, making early diagnosis and prevention critical for improving patient outcomes. However, accurate diagnosis often depends on physician expertise, which can lead to medical errors. This study develops a machine learning-based predictive model to assist physicians in diagnosing and predicting heart attacks with higher precision. Patient data from an Iranian hospital were collected and preprocessed, including handling missing values, normalization, and outlier removal. Analysis using RapidMiner software identified seven significant predictors of heart attack from thirteen initial factors, achieving a high accuracy of 97.86%, confirmed by hospital physicians. Among various machine learning algorithms, the decision tree demonstrated the highest accuracy (97%) and provided interpretable decision pathways for clinical use. Decision rules extracted from the decision tree highlighted chest pain, trait-anxiety, age, smoking, and resting electrocardiographic results as the most influential factors. To validate these findings, fuzzy clustering was employed, revealing an alignment rate of 76% for the two most critical factors—chest pain and trait-anxiety. These findings emphasize the prioritization of these factors during initial evaluations, followed by age, smoking, and resting electrocardiographic results to enhance diagnostic accuracy. The integration of machine learning and fuzzy clustering offers a robust approach to clinical decision-making, verified by expert feedback from hospital physicians, and contributes to reducing diagnostic errors.}, archive = {J_EAAI}, author = {Mohammad Dehghani Saryazdi and Ali Mostafaeipour}, doi = {10.1016/j.engappai.2024.109968}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109968}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Identification and validation of key predictive factors for heart attack diagnosis using machine learning and fuzzy clustering}, volume = {142}, year = {2025}, }
• (2025). Efficient network architecture for target detection in challenging low-light environments. EAAI, 142, 109967. (www)

  @article{ , abstract = {Target detection in low-light and dark conditions presents significant challenges compared to normal scenarios due to blurriness, low contrast, and information loss from shadows and darkness. Color distortion and noise further hinder color-based detection algorithms. This study introduces an improved detection model that incorporates image enhancement techniques specifically designed for low-light conditions. The focus is on preprocessed image restoration to achieve realistic and natural visual quality. Key enhancements include a Multi-level Fine-Grained Feature Prediction Network architecture, enabling layered feature fusion and refined feature extraction. This leads to improved model accuracy, reduced parameters, and a dynamic detection output head that enhances target region precision, ultimately improving overall detection accuracy. Validation on a public dataset reveals a 5.6% increase in mean average precision, a 6.2% reduction in computational costs in terms of Floating-point Operations, and a 20% decrease in model parameter size. Our method outperforms mainstream low-light target detection algorithms in terms of performance and versatility.}, archive = {J_EAAI}, author = {Qiang Liu and Wei Huang and Tao Hu and Xiaoqiu Duan and Jie Yu and Jiahuan Huang and Jianghao Wei}, doi = {10.1016/j.engappai.2024.109967}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109967}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient network architecture for target detection in challenging low-light environments}, volume = {142}, year = {2025}, }
• (2025). Short-term prediction of dissolved oxygen and water temperature using deep learning with dual proportional-integral-derivative error corrector in pond culture. EAAI, 142, 109964. (www)

  @article{ , abstract = {Dissolved oxygen (DO) and water temperature (WT) are the most important water quality indicators that directly affect the metabolism and development of aquatic products in pond cultures. Therefore, the accurate prediction of these two water quality factors is crucial for improving aquaculture efficiency; however, the complexity and variability of pond aquaculture environments make accurate and efficient prediction of water quality challenging in this field. To address this problem, this study proposes a deep learning method combined with a proportional-integral-derivative (PID) error corrector for the accurate and rapid prediction of DO and WT. This method consists of two parts: a benchmark deep-network prediction model and a PID error corrector. The benchmark model is used for the multistep forward prediction of DO and WT, providing feedback input for the PID error corrector, which obtains the prediction error by comparing the feedback input with the actual values of the two predicted variables and then based on the prediction error, calculates the error correction amount for the current predicted values of the benchmark model. The proposed method was validated using three chaotic time-series datasets and an actual pond aquaculture water environment dataset. The results demonstrate the effectiveness of the proposed PID error corrector which significantly improved prediction accuracy of the benchmark model despite its simple structure (with only three adjustable parameters). Thus, the PID error corrector has great engineering application prospects in the field of pond culture water quality monitoring.}, archive = {J_EAAI}, author = {Xinhui Zhou and Yinfeng Hao and Yang Liu and Lanxue Dang and Baojun Qiao and Xianyu Zuo}, doi = {10.1016/j.engappai.2024.109964}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109964}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Short-term prediction of dissolved oxygen and water temperature using deep learning with dual proportional-integral-derivative error corrector in pond culture}, volume = {142}, year = {2025}, }
• (2025). Dynamic weighted adversarial domain adaptation network with sparse representation denoising module for rotating machinery fault diagnosis. EAAI, 142, 109963. (www)

  @article{ , abstract = {Domain adaptation effectively addresses the issue of varying vibration data conditions in fault diagnosis. However, current domain adaptation methods rarely consider the impact of imbalanced distributions and background noise, limiting their applicability. To alleviate these problems, a dynamic weighted adversarial domain adaptation network with the sparse representation denoising module (DWADAN-SRDM) is designed for rotating machinery fault diagnosis. First, a sparse representation denoising module is designed to effectively suppress noise, enabling more accurate extraction of vibration signal features. Then, an adaptive exponential loss is introduced address class imbalance. Finally, a weighted adversarial domain adaptation strategy, integrated with a dynamic sample weighting mechanism that can assign greater weight to source domain samples with higher relevance to the target domain, is used to extract domain-invariant features while reducing negative transfer. Extensive experiments on three unbalanced rotating machinery datasets demonstrate that the fault identification accuracy of DWADAN-SRDM improves by over 8.5% compared to existing methods.}, archive = {J_EAAI}, author = {Maogui Niu and Hongkai Jiang and Haidong Shao}, doi = {10.1016/j.engappai.2024.109963}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109963}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic weighted adversarial domain adaptation network with sparse representation denoising module for rotating machinery fault diagnosis}, volume = {142}, year = {2025}, }
• (2025). Development of six-degree-of-freedom prosthetic hand with integrated voice and electromyography control using machine learning. EAAI, 142, 109949. (www)

  @article{ , abstract = {Objective Upper limb amputations impact millions worldwide, limiting daily activities and reducing independence. Current prosthetic hand control systems face limitations: body-powered prostheses require high physical effort, EMG-based systems struggle with complex movement control, and voice-only interfaces lack reliability in noisy environments. This study aims to integrate voice recognition with Electromyography (EMG) to enhance control accuracy and user experience in a six-degree-of-freedom prosthetic hand. Methods In this study, we developed a machine learning framework to integrate voice and EMG signals for seamless prosthetic control. Data were collected from eight healthy male participants performing five distinct hand gestures and from one female participant producing five voice commands. The EMG dataset consisted of 6080 samples, while the voice command dataset included 200 samples. The EMG data were preprocessed to reduce noise, and 16 time-domain features were extracted. Voice data were processed to extract Mel-Frequency Cepstral Coefficients (MFCCs). We evaluated 29 classifiers to identify the optimal model for EMG data classification, while Dynamic Time Warping (DTW) was applied for classifying voice commands. Results The EMG data achieved a classification accuracy of 76.00% using a Medium Neural Network (MNN) classifier, while the voice commands attained a 94.80% accuracy with DTW. The integrated system allows for intuitive control of five distinct hand gestures using both EMG and voice inputs. Conclusion The combined EMG and voice-based control framework provides a more flexible and effective solution for prosthetic hand control, addressing the limitations of single-modality systems and supporting user's functional independence.}, archive = {J_EAAI}, author = {Fariha Shabbir and Muzzaffar Iqbal and Ali Asghar and Saba Khan and Saad Jawaid Khan}, doi = {10.1016/j.engappai.2024.109949}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109949}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Development of six-degree-of-freedom prosthetic hand with integrated voice and electromyography control using machine learning}, volume = {142}, year = {2025}, }
• (2025). ENVQA: Improving visual question answering model by enriching the visual feature. EAAI, 142, 109948. (www)

  @article{ , abstract = {Visual Question Answering (VQA) is pivotal in various industries, including medicine. Current approaches typically rely on identifying patterns between image regions and questions, using attention-learning techniques to highlight essential information and suppress noise. However, existing VQA systems often overlook crucial foreground and background-related features in images, limiting their ability to tackle complex questions effectively. Most VQA models employ either spatial or channel attention mechanisms. Spatial attention localizes the region of interest (ROI) but may overlook global semantic relationships between salient objects. Conversely, channel attention enhances feature representation but disregards spatial dynamics within images. To address these limitations, we propose ”ENVQA” (Enriching V in VQA), a novel VQA model that integrates enriched visual features by leveraging both spatial and object-level features, alongside spatial and channel attention networks. Our model aims to enhance understanding by capturing both local and global contexts within images. Experimental evaluations on benchmark datasets such as VQA 2.0, TDIUC, and GQA demonstrate that ENVQA outperforms state-of-the-art (SOTA) models utilizing attention mechanisms.}, archive = {J_EAAI}, author = {Souvik Chowdhury and Badal Soni}, doi = {10.1016/j.engappai.2024.109948}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109948}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {ENVQA: Improving visual question answering model by enriching the visual feature}, volume = {142}, year = {2025}, }
• (2025). Physics-based self-adaptive algorithm for estimating the long-term performance of concrete shrinkage. EAAI, 142, 109945. (www)

  @article{ , abstract = {Drying shrinkage of concrete, a complex phenomenon susceptible to humidity diffusion out of concrete, represents one of the critical reasons for the structural deterioration that always affects the long-term service performance of concrete. Although various methods have been developed to forecast concrete shrinkage, more reliable and accurate prediction methods are required. This study proposes a self-adaptive physics-informed neural network (SA-PINN) for modeling the mechanism of humidity diffusion and predicting the long-term performance of concrete shrinkage. The model was enriched with a Gaussian probabilistic approach to accurately adjust the self-adaptive loss function. Specifically, the model exploits the feature settings in the neural network and leverages the loss of physical constraints as a way to fine-tune the gradient descent algorithm, which can produce more accurate results than traditional methods. The performance of the SA-PINN model is compared to that of a physics-informed network and a purely data-driven method. The results revealed that the proposed model could reasonably simulate the humidity diffusion behavior and significantly improve the prediction precision of shrinkage in concrete. Model testing provided the highest correlation coefficient (0.954), lowest mean square error (0.266), and lowest mean absolute percentage error (0.035), indicating the accuracy of this model in predicting the drying shrinkage of concrete.}, archive = {J_EAAI}, author = {Wafaa Mohamed Shaban and Shui-Long Shen and Ayat Gamal Ashour and Annan Zhou and Khalid Elbaz}, doi = {10.1016/j.engappai.2024.109945}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109945}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-based self-adaptive algorithm for estimating the long-term performance of concrete shrinkage}, volume = {142}, year = {2025}, }
• (2025). A novel integration of multiple learning methods for detecting misleading information from different datasets during the pandemic. EAAI, 142, 109944. (www)

  @article{ , abstract = {Coronavirus Disease 2019 (COVID-19) was an intensely and commonly discussed topic on social media platforms during the pandemic due to uncertainty about the virus, especially as new variants of the virus emerged around the world. Unfortunately, during the pandemic, people shared many posts about COVID-19 on their social media accounts without paying attention or checking whether they were true or not. In this way, intentionally or unintentionally, they highly manipulated public opinion through their posts. The majority of these posts contained misleading information that negatively affected readers' cognitive and mental health, leading to a new neologism associated with the pandemic: “infodemic.” Therefore, the present study focuses on the classification of Fake News disseminated during the pandemic to mislead people. To this end, five different datasets were first trained independently using natural language processing and machine learning methods, and the results obtained were compared. Later, these datasets were combined according to the different scenarios to improve the model performance. According to the results, the highest accuracy value of 98.1% was obtained with the model Efficiently Learning an Encoder that Classifies Token Replacements Accurately (ELECTRA) when the datasets were trained independently. Similarly, the highest training accuracy of 94.12% was obtained with the ELECTRA method and the highest test accuracy of 91.71% was obtained with the Random Forest method. In summary, the model ELECTRA, which is less preferred than other pre-trained models, achieved the highest performance scores in all study-specific scenarios.}, archive = {J_EAAI}, author = {Muhammed Coşkun Irmak and Tolga Aydın and Mete Yağanoğlu}, doi = {10.1016/j.engappai.2024.109944}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109944}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel integration of multiple learning methods for detecting misleading information from different datasets during the pandemic}, volume = {142}, year = {2025}, }
• (2025). Regional coverage balance and efficient worker recruitment for self-organized mobile crowdsourcing. EAAI, 142, 109943. (www)

  @article{ , abstract = {With the widespread adoption of smart devices, self-organized mobile crowdsourcing has become a popular method for decentralized data collection, where mobile users autonomously participate in task completion by leveraging their mobility and proximity to tasks. A key challenge in this context is achieving regional coverage balance, ensuring that tasks are equitably distributed across geographic areas to prevent both underserved and overserved regions. However, focusing solely on regional coverage without considering user satisfaction can negatively impact the quality of service. On the other hand, optimal worker recruitment in self-organized mobile crowdsourcing can maximize the expected overall quality of service. To address this, we develop a framework that balances regional coverage while enhancing user satisfaction by predicting user trajectories and identifying optimal service providers. Additionally, we tackle the worker recruitment problem by formulating a model that maximizes the expected quality of service. Our approach incorporates two collaborative deep learning networks: we first employ Proximal Policy Optimization (PPO) for matching candidates and training batches during the training phase, and then use Long Short-Term Memory (LSTM) to extract learning patterns of candidates, aiding PPO in making more effective recruitment decisions. We evaluate the performance of the proposed approach through extensive experiments using real-world data and by comparing it with existing strategies from previous research. Simulation results demonstrate that our method significantly improves both coverage balance and service quality in large-scale, decentralized mobile crowdsourcing environments.}, archive = {J_EAAI}, author = {Ruiqing Liu and Yonghong Wang and Xiaofeng Wang}, doi = {10.1016/j.engappai.2024.109943}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109943}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Regional coverage balance and efficient worker recruitment for self-organized mobile crowdsourcing}, volume = {142}, year = {2025}, }
• (2025). Intra- and inter-instance location correlation network for human–object interaction detection. EAAI, 142, 109942. (www)

  @article{ , abstract = {Objective: Human–object interaction detection is to detect human–object pairs and identify their interactions, which is of great significance to improve the perception and decision-making ability of embodied artificial intelligence robots such as service robots. Under the robot view in complex indoor scenes, some situations with imbalanced sizes and crowded instances make human–object detection and matching particularly difficult. Novelty: Focus on the new task of robotic human–object interaction detection in complex indoor scenes, we propose an intra- and inter-instance location correlation network (ILCN) for human–object interaction detection. First, to improve the detecting and matching of complex combinations, we put forward parallel human–object spatial association method (PSA). PSA links humans and objects in a parallel manner by splicing their instance features and then exploits spatial prior knowledge to enhance the focus of features on instances layer by layer. Then, we design an interaction network block through instance spatial layout relationship guidance (SLG). SLG extracts spatial layout relations from the inner human–object location and fuses the layout relation with interaction queries through self-attention, while in cross-attention the fused features will be split by a skip-connection mechanism to guarantee interaction classification. Findings: The results of the comparison experiments with state-of-the-art methods on three public benchmark datasets show that ILCN achieves the best comprehensive performance on the dataset with complex combinations.}, archive = {J_EAAI}, author = {Minglang Lu and Guanci Yang and Yang Wang and Kexin Luo}, doi = {10.1016/j.engappai.2024.109942}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109942}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intra- and inter-instance location correlation network for human–object interaction detection}, volume = {142}, year = {2025}, }
• (2025). Stampede detector based on deep learning models using dense optical flow. EAAI, 142, 109940. (www)

  @article{ , abstract = {The world’s population has grown in recent decades, increasing social events and leading to more crowd situations with potential issues, such as bottlenecks, stampedes, or falls. In this context, this paper presents an approach for stampede detection from image sequences in low- and medium-crowd. It is based on a feature vector extracted from the dense optical flow, using the Gunner-Farneback method, and a deep learning-based classification model capable of determining, frame by frame, whether a stampede is happening. It has been evaluated on four different datasets: two widely used in the state-of-the-art— University of Minnesota (UMN) and Performance Evaluation of Tracking and Surveillance (PETS-2009)— and two new labeled datasets, Geintra-Behaviour-Analysis (GBA-Stampedes) and Geintra-Santander Multiple Actions Dataset in Cruises (GSMADC), which include realistic indoor and outdoor scenarios, as well as diverse crowd types and sizes (up to 6 people in GSMADC and a minimum of 15 in GBA). Both datasets have been made publicly available to increase the limited number of sequences for validating stampede detection in videos, with more than 43000 frames. The proposed method was evaluated across various training scenarios to test its adaptability to new environments. In the most challenging scenario, using a limited training set, our system achieved average metrics of around 99% on UMN and PETS-2009, 95% on GBA, and 91% on GSMADC. In comparison, other models achieved only 90% on UMN and PETS-2009, 80% on GBA, and below 80% on GSMADC, demonstrating the accuracy and robustness of the stampede detector across scenarios.}, archive = {J_EAAI}, author = {Antonio Carlos Cob-Parro and Cristina Losada-Gutiérrez and Marta Marrón-Romera}, doi = {10.1016/j.engappai.2024.109940}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109940}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Stampede detector based on deep learning models using dense optical flow}, volume = {142}, year = {2025}, }
• (2025). Multi-objective optimization of buckling load and natural frequency in functionally graded porous nanobeams using non-dominated sorting genetic algorithm-II. EAAI, 142, 109938. (www)

  @article{ , abstract = {This study investigates the fundamental natural frequency and critical buckling load of Functionally Graded Porous nanobeams supported by an elastic medium, addressing the need for optimized designs in advanced nanostructures. Utilizing a Genetic Algorithm and Non-Dominated Sorting Genetic Algorithm-II, the research aims to identify the Pareto front for these two objectives while incorporating surface effects. The nanobeam is modeled using Nonlocal Strain Gradient Theory and Gurtin-Murdoch surface elasticity theory, with governing equations solved via the Generalized Differential Quadrature Method based on Reddy's Third-order Shear Deformation Theory. Key input parameters, including temperature gradient, residual surface stress, porosity, and elastic foundation properties, are varied to train two Artificial Neural Networks for output prediction. Results indicate that for the fundamental frequency, significant factors include the material length scale and the Pasternak shear foundation parameter, while the critical buckling load is mainly influenced by the temperature gradient and the same material parameters. These findings provide critical insights for designers, allowing them to make informed decisions based on optimal values for eight input parameters.}, archive = {J_EAAI}, author = {Hao Liu and Ali Basem and Dheyaa J. Jasim and Mohammad Hashemian and S. Ali Eftekhari and Halah Jawad Al-fanhrawi and Barno Abdullaeva and Soheil Salahshour}, doi = {10.1016/j.engappai.2024.109938}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109938}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-objective optimization of buckling load and natural frequency in functionally graded porous nanobeams using non-dominated sorting genetic algorithm-II}, volume = {142}, year = {2025}, }
• (2025). Adaptive filtering algorithm based on beta fluctuation and fractional order differential evolution in impulsive noise. EAAI, 142, 109937. (www)

  @article{ , abstract = {As a widely used optimization algorithm in adaptive filtering, the gradient method is sensitive to large outliers, leading to poor robustness of adaptive filtering algorithm against impulsive noise. Due to its stochastic mutation and crossover operations without external inputs, the differential evolution (DE) method can reduce the probability of unstable convergence of adaptive filtering algorithm in impulsive noise. However, the smooth surface of cost function is easily destroyed by large outliers, which may make DE fall into local optimum, thereby resulting in lower convergence accuracy. To improve the convergence performance of adaptive filtering algorithms in impulsive noise, we propose a Beta fluctuation and fractional order differential evolution (BF-FDE) adaptive filtering algorithm. For the proposed BF-FDE, the Beta fluctuation is used to break the local equilibrium by controlling the distribution of generated individuals through adjusting the shape parameters, and the fractional order difference mutation strategy with memory characteristics is applied to improve the convergence accuracy and enhance robustness. Subsequently, the mechanism of the BF-FDE against impulsive noise is theoretically analyzed. Finally, the superiority of BF-FDE adaptive filtering algorithm in impulsive noise has been demonstrated by applications including system identification, channel equalization and noise cancellation.}, archive = {J_EAAI}, author = {Yongjiang Luo and Jiyang Li and Jiali Yang and Susu Yan and Yuhang Bai and Xiang Zhang}, doi = {10.1016/j.engappai.2024.109937}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109937}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive filtering algorithm based on beta fluctuation and fractional order differential evolution in impulsive noise}, volume = {142}, year = {2025}, }
• (2025). DSRS: DELIGHT sequential recommender system. EAAI, 142, 109936. (www)

  @article{ , abstract = {Sequential recommendation is becoming more critical in a variety of e-commerce platforms. The aim of sequential recommender systems is to model the dynamic preferences of users based on their previous actions and predict what they will do next. The collected user activity logs on real-world platforms could be quite long. This wealth of information provides options to follow users’ actual interests. Prior efforts primarily aimed at providing recommendations following recent behaviors. Meanwhile, the entire sequential data may not be used efficiently since early actions may influence users’ decisions at present. Furthermore, scanning the whole behavior sequence while doing inference for every user is unbearable due to the need for prompt reaction time in real-world applications. To this end, we propose the DELIGHT Sequential Recommender System (DSRS), which takes the above properties into account to recommend the next item the user might be interested in. DSRS divides the entire user behavior sequence into long- and short-term segments and models them through independent networks before integrating their learned representations. In particular, the first network learns user long-term, whereas the second one learns short-term preferences and then combines them for an efficient joint recommendation. Experimental findings across four datasets show that our model outperforms other state-of-the-art sequential models in apprehending long-term dependence.}, archive = {J_EAAI}, author = {Syed Tauhid Ullah Shah and Fazlullah Khan and Shirin Yamani and Ryan Alturki and Foziah Gazzawe and Muhammad Imran Razzak}, doi = {10.1016/j.engappai.2024.109936}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109936}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {DSRS: DELIGHT sequential recommender system}, volume = {142}, year = {2025}, }
• (2025). Preference ranking organization method for enrichment evaluation-based feature selection for multiple source ordered information systems. EAAI, 142, 109935. (www)

  @article{ , abstract = {Feature selection in multi-source ordered decision-making information systems is an underexplored domain that poses unique challenges. This paper introduces a novel approach to fill this research gap by integrating the preference function from the Preference Ranking Organization Method for Enrichment Evaluation method into a feature selection framework. Our method diverges from traditional single-source focused techniques, providing a tailored solution for the intricacies of multi-source environments, especially those characterized by ordered information. We propose an algorithm that synergizes rough set theory with the preference function to quantitatively assess and rank the relevance of features for decision-making. This is achieved through the construction of preference index matrices, which encapsulate the attribute bias order relationships within the multi-source sequential decision-making information system. To validate the efficacy of our method, we conducted comprehensive experiments on 12 diverse and complex University of California, Irvine public datasets. The results indicate that our feature selection method surpasses existing approaches in terms of accuracy and efficiency, while also providing enhanced insights into the decision-making processes within multi-source ordered information systems.}, archive = {J_EAAI}, author = {Weihua Xu and Zishuo Yang}, doi = {10.1016/j.engappai.2024.109935}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109935}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Preference ranking organization method for enrichment evaluation-based feature selection for multiple source ordered information systems}, volume = {142}, year = {2025}, }
• (2025). Spatio-temporal collaborative multiple-stream transformer network for liver lesion classification on multiple-sequence magnetic resonance imaging. EAAI, 142, 109933. (www)

  @article{ , abstract = {Accurate identification of focal liver lesions is essential for determining the appropriate therapeutic approach in clinical practice. Magnetic resonance imaging (MRI) is a valuable technology for precise classification, revealing diverse physical and biological details of lesions. However, due to the wide variety and morphological variability of the lesions, unsystematic mixing analysis of multiple-sequences MRI may cause aliasing of lesion information, obscuring the inter-tissue relationships between various imaging sequences and impeding a comprehensive diagnosis. In this paper, we proposed a Spatio-Temporal Collaborative Multiple-Stream Transformer Network that simultaneously considers spatial contrast and temporal variations to obtain detailed information on anatomical structures and tissue dynamics, effectively organizing and utilizing multiple-sequence MRI for analysis. Specifically, multiple-sequence MRI is first grouped into multiple streams based on MRI diagnostic characteristics. To reduce the interference of redundancy across multiple streams, we design a bottleneck bridge structure for spatial information aggregation. Additionally, we adopt a bidirectional Long Short-Term Memory to simulate radiologists observing the vascular morphology and hemodynamics in lesion sites from contrast-enhanced sequences. Experiments conducted on public MRI dataset, which includes seven categories of focal liver lesions from 498 patients, demonstrate that our framework achieves state-of-the-art performance, with an accuracy of 85.6%, a precision of 87.4%, a recall of 84.2%, an F1-score of 85.3%, and an Area Under the Curve of 97.1%. The experimental results indicate that the network performs well in predicting focal liver lesions, advancing the application of precision medicine.}, archive = {J_EAAI}, author = {Shuangping Huang and Zinan Hong and Bianzhe Wu and Jinglin Liang and Qinghua Huang}, doi = {10.1016/j.engappai.2024.109933}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109933}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatio-temporal collaborative multiple-stream transformer network for liver lesion classification on multiple-sequence magnetic resonance imaging}, volume = {142}, year = {2025}, }
• (2025). Cross-modal augmentation for few-shot multimodal fake news detection. EAAI, 142, 109931. (www)

  @article{ , abstract = {The nascent topic of fake news requires automatic detection methods to quickly learn from limited annotated samples. Therefore, the capacity to rapidly acquire proficiency in a new task with limited guidance, also known as few-shot learning, is critical for detecting fake news in its early stages. Existing approaches either involve fine-tuning pre-trained language models which come with a large number of parameters, or training a complex neural network from scratch with large-scale annotated datasets. This paper presents a multimodal fake news detection model which augments multimodal features using unimodal features. For this purpose, we introduce Cross-Modal Augmentation (CMA), a simple approach for enhancing few-shot multimodal fake news detection by transforming n-shot classification into a more robust (n × z)-shot problem, where z represents the number of supplementary features. The proposed CMA achieves state-of-the-art (SOTA) results over three benchmark datasets, utilizing a surprisingly simple linear probing method to classify multimodal fake news with only a few training samples. Furthermore, our method is significantly more lightweight than prior approaches, particularly in terms of the number of trainable parameters and epoch times. The code is available here: https://github.com/zgjiangtoby/FND_fewshot}, archive = {J_EAAI}, author = {Ye Jiang and Taihang Wang and Xiaoman Xu and Yimin Wang and Xingyi Song and Diana Maynard}, doi = {10.1016/j.engappai.2024.109931}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109931}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-modal augmentation for few-shot multimodal fake news detection}, volume = {142}, year = {2025}, }
• (2025). Integrating the group-based fuzzy best-worst method and fuzzy technique for order of preference by similarity to ideal solution for selecting contractors for heritage building renovation projects. EAAI, 142, 109928. (www)

  @article{ , abstract = {Heritage buildings are increasingly vulnerable to environmental degradation and human activities, necessitating the development of effective, context-specific restoration methods. This study investigates the factors influencing the efficiency of design and implementation in heritage conservation projects through a statistical survey involving 85 participants and 2 experts, which identified 12 key criteria for contractor selection during the design and construction phases of a heritage building restoration case study. Based on these findings, a novel multi-criteria decision-making model was developed, integrating the Fuzzy Best-Worst Method, which accommodates multiple decision-makers simultaneously, with the Fuzzy Technique for Order of Preference by Similarity to Ideal Solution. This integration facilitates the ranking of contractors and the selection of the most suitable alternative. The proposed model was applied to a case study of a historical heritage building in the Mekong Delta region of Vietnam, where it computes fuzzy weights for the 12 criteria using a system of nonlinear equations based on Triangular Fuzzy Numbers. Criterion 1, "Assurance level of material and architectural requirements," received the highest fuzzy weight (0.156, 0.169, 0.178), while Criterion 5, "Duration and schedule," received the lowest fuzzy weight (0.029, 0.030, 0.033). Implemented in Python, the model was effectively automated for problems with a large number of constraints (207 constraints) and achieved a high consistency ratio of 0.039, ensuring robustness in decision-making. Results from the case study validate the model's ability to enhance accuracy and flexibility in contractor selection, with multiple decision-makers evaluating alternatives based on Triangular Fuzzy Numbers simultaneously for heritage restoration projects.}, archive = {J_EAAI}, author = {L.D. Long}, doi = {10.1016/j.engappai.2024.109928}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109928}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating the group-based fuzzy best-worst method and fuzzy technique for order of preference by similarity to ideal solution for selecting contractors for heritage building renovation projects}, volume = {142}, year = {2025}, }
• (2025). A prior segmentation knowledge enhanced deep learning system for the classification of tumors in ultrasound image. EAAI, 142, 109926. (www)

  @article{ , abstract = {Breast and thyroid cancers are prevalent among women worldwide. Ultrasound (US) examination is widely used for the early detection of breast and thyroid cancers. However, due to the blurred tumor boundaries and irregular shapes, the computer-aided diagnosis (CAD) of tumors based on US is challenging. Numerous studies have introduced deep learning-based multi-task learning approaches to address this issue, but these methods may result in feature redundancy and misinformation. Tumor segmentation is a prerequisite step for US CAD, and a higher Dice coefficient is associated with more accurate classification outcomes. Therefore, this paper introduces a novel deep-learning system that fully utilizes segmentation knowledge to boost classification performance. The system starts with a hybrid convolutional neural network (CNN)-Transformer for tumor localization and coarse segmentation, then uses a lightweight CNN-based U-Net to refine segmentation results. Subsequently, segmentation knowledge is harnessed to augment the network input and enhance multimodal feature extraction, resulting in improved classification performance. Our proposed method yielded a Dice coefficient of 83.62% and 77.20% for breast and thyroid tumor segmentation and area under curve (AUC) values of 0.9536 and 0.9475 for their respective classifications. Compared to non-segmentation knowledge-based classification models, our method obtained an increase in AUC of 0.1054 and 0.0566 on the breast and thyroid datasets, respectively. It outperformed the performance of the State-Of-The-Art (SOTA) methods across various datasets. In summary, our proposed system shows promise for application in US tumor analysis and holds potential to be extended to additional diseases and modalities.}, archive = {J_EAAI}, author = {Tao Jiang and Jun Guo and Wenyu Xing and Ming Yu and Yifang Li and Bo Zhang and Yi Dong and Dean Ta}, doi = {10.1016/j.engappai.2024.109926}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109926}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A prior segmentation knowledge enhanced deep learning system for the classification of tumors in ultrasound image}, volume = {142}, year = {2025}, }
• (2025). A novel data-driven framework for driving range prognostics in electric vehicles. EAAI, 142, 109925. (www)

  @article{ , abstract = {Electric vehicle (EV) driving range prediction is crucial for enhancing EV adoption and mitigating range anxiety among drivers. Despite advancements in battery technology, accurately estimating the remaining driving range under varying conditions remains a significant challenge. To address this, we propose a novel approach capable of prognosticating the Maximum Driving Range (MDR) an EV can achieve. Our method intelligently segments routes and integrates machine learning with physics-based models to predict vehicle speed, energy consumption, and power usage, offering a more precise and dynamic driving range estimation. Specifically, the approach utilizes stochastic dropout-based Long Short-Term Memory (LSTM) networks to predict vehicle speed, which serves as input to a Light Gradient Boosting Machine (LightGBM) model for estimating energy and power consumption. In addition, the Maximum Driving Range (MDR) concept is introduced as a new metric for identifying “hazard zones” where the likelihood of battery disconnection is high. Our approach was validated through real-world driving tests across three case studies in San José, Costa Rica. The model achieved a mean absolute error of 6.87 km/h for speed, 0.067 kWh for energy consumption, and 8.57 kW for power consumption, successfully predicting hazard zones 3.90 to 8.70 km before battery disconnection. These findings demonstrate the potential of this hybrid model in enhancing EV range predictions, offering a practical tool for improved route planning and driver confidence. Future work could involve adapting the model to diverse driving scenarios, incorporating user-specific risk tolerance and environmental factors, and improving computational efficiency to support real-time applications.}, archive = {J_EAAI}, author = {Jorge E. García Bustos and Cesar Baeza and Benjamín Brito Schiele and Violeta Rivera and Bruno Masserano and Marcos E. Orchard and Claudio Burgos-Mellado and Aramis Perez}, doi = {10.1016/j.engappai.2024.109925}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109925}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel data-driven framework for driving range prognostics in electric vehicles}, volume = {142}, year = {2025}, }
• (2025). Optimal machine learning- and deep learning-driven algorithms for predicting the future value of investments: A systematic review and meta-analysis. EAAI, 142, 109924. (www)

  @article{ , abstract = {The COVID-19 pandemic and the increasing competitive landscape have led asset management companies to consider investing in applying Artificial Intelligence (AI)-driven technologies to minimise the risk and maximise the profitability of the investment funds they manage. Thus, a systematic review and a meta-analysis of the relevant literature were conducted to provide evidence-based informed recommendations on which AI-driven technologies could be leveraged for such purpose. Data on both Machine Learning (ML)- and Deep Learning (DL)-driven technologies applied to aid the management of investment funds in China and, specifically, in and around Shenzhen, were pooled from eleven eligible and recent studies (since September 15, 2017) and analysed accordingly. The key business-relevant and human-interpretable metrics representing their performance were identified in the root mean squared error (RMSE), in the same unit of currency of the investment funds, and the correlation strength between the predicted and actual values. One ML- and one DL-based algorithms were recommended to be used in the short and long terms respectively. In particular, the ML-based Gradient Boosting Decision Tree (GBDT) algorithm was found the most accurate in the relevant literature, e.g., 28.16% more accurate than the Support Vector Regressor (SVR), also having a highly competitive ability to capture trends in the actual values of investment funds (83.7% of correlation strength), whilst the Long-Short Term Memory (LSTM)-GBDT model was identified as the most accurate DL-based algorithm, 15.05% more accurate than the GBDT and with 13.2% higher ability to capture trends in the actual investment funds’ values than the GBDT.}, archive = {J_EAAI}, author = {Luca Parisi and Marianne Lyne Manaog}, doi = {10.1016/j.engappai.2024.109924}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109924}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimal machine learning- and deep learning-driven algorithms for predicting the future value of investments: A systematic review and meta-analysis}, volume = {142}, year = {2025}, }
• (2025). Estimation of the orientation of potatoes and detection bud eye position using potato orientation detection you only look once with fast and accurate features for the movement strategy of intelligent cutting robots. EAAI, 142, 109923. (www)

  @article{ , abstract = {The accurate detection of potato orientation and bud eye positions is critical for guiding the end-effector of intelligent cutting robots. This study introduced Potato Orientation Detection You Only Look Once (POD-YOLO), a novel lightweight model based on YOLOv8n, designed for fast and precise detection of potato orientation and bud eye locations. Key innovations include replacing the Cross Stage Partial Dark Network (CSPDarkNet) with the Cross Stage Partial and Dual Partial Network (CSPDPNet) to reduce parameter count and improve detection accuracy. Additionally, the "no more strided convolutions or pooling" approach replaced downsampling modules in the backbone and neck, enhancing detection of small targets and low-resolution images. The regression loss function was further optimized by substituting Kalman Filtering Intersection over Union (KFIoU) for improved rotated bounding box performance. Experimental results showed that POD-YOLO achieved a mean Average Precision (mAP) of 97.2%, with a precision of 95.2%, recall of 94.0%, and detection time of 9.01 ms. With only 1.75 million parameters, POD-YOLO was lightweight and efficient, meeting real-time requirements. This research offers a robust and effective solution for automated potato orientation and bud eye detection, laying the groundwork for advanced agricultural automation.}, archive = {J_EAAI}, author = {Jie Huang and Xiangyou Wang and Chengqian Jin and Fernando Auat Cheein and Xinyu Yang}, doi = {10.1016/j.engappai.2024.109923}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109923}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Estimation of the orientation of potatoes and detection bud eye position using potato orientation detection you only look once with fast and accurate features for the movement strategy of intelligent cutting robots}, volume = {142}, year = {2025}, }
• (2025). Physics-guided neural network for predicting international roughness index on flexible pavements considering accuracy, uncertainty and stability. EAAI, 142, 109922. (www)

  @article{ , abstract = {An outstanding amount of funds are allocated to maintain road network conditions. To ensure the serviceability of roads, the accurate prediction of its roughness or International Roughness Index plays a pivotal role in road management. Artificial neural network, typically used in roughness prediction, is a powerful machine learning algorithm that learns complex patterns in data with non-linear relationships. However, it remains a black box solution and relies heavily on the utilized data and its internal structures, causing model's overfitting and instability. To address such issues, a physics-guided neural network modelling framework is proposed for short- and long-term predictions of roughness aimed at improving model's accuracy, uncertainty and stability. This framework fuses the output of physics-based model simulations along with field observational data acquired from the Long-Term Pavement Performance public database as inputs to develop a neural network architecture. Additionally, the framework uses a physics-based loss function in the network's learning process to ensure the predictions are consistent with the known physics. The performances are evaluated and compared to traditional artificial neural network. The comparison results indicate that the proposed modelling framework can increase the accuracy by 4%, and 26.08%, reduce the uncertainty by 4% and at least 22.15%, and improve the stability by 24.09% and by 46.34%, for one year and multi-year predictions, respectively. This framework offers great potential for accurate, reliable and stable predictions for engineering asset conditions by leveraging the complementary strengths of numerical simulations and data-driven models.}, archive = {J_EAAI}, author = {Kun Chen and Mehran Eskandari Torbaghan and Nick Thom and Asaad Faramarzi}, doi = {10.1016/j.engappai.2024.109922}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109922}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-guided neural network for predicting international roughness index on flexible pavements considering accuracy, uncertainty and stability}, volume = {142}, year = {2025}, }
• (2025). A physics-based iterative learning framework for quantitative parametric imaging with application to photoacoustic imaging. EAAI, 142, 109920. (www)

  @article{ , abstract = {Quantitative parametric imaging represents a typical application of knowledge discovery from visual scenes. In numerous instances, purely data-driven neural networks may face challenges in accurately interpreting the physical characteristics of complex visual scenarios, especially when faced with limited or insufficiently diverse training data to encompass the full spectrum of physical phenomena. However, the integration of physical priors into network architectures continuously enhances their capabilities in this domain. This paper introduces a physics-informed iterative learning framework aimed at recovering multiple characteristic parameters of imaging objects from visual scenes. The framework incorporates forward imaging models into the learning process, facilitating mapping from raw measurement signals acquired by the detector to multiple-parameter maps. The application of this framework focuses on the joint quantification of the optical absorption coefficient and the absolute Grüneisen parameter at the pixel level in biological photoacoustic imaging. The iterative gradient network is trained and evaluated on simulation and phantom datasets with established reference values. The results demonstrate significant improvements in the recovery of parametric maps, both qualitatively through visual inspection and quantitatively using evaluation metrics. In a phantom study, the structural similarity and peak signal-to-noise ratio of the recovered parametric maps exhibit improvements of at least 19.5% and 33.5%, respectively, compared to state-of-the-art nonlearning two-step methods. To the best of our knowledge, this represents the first deep learning solution for quantifying absolute Grüneisen parameter from photoacoustic measurements. This work contributes a versatile iterative learning framework for extracting physical insights from detector data.}, archive = {J_EAAI}, author = {Sun Zheng and Zhu Aoying and Hou Yingsa and Sun Meichen}, doi = {10.1016/j.engappai.2024.109920}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109920}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A physics-based iterative learning framework for quantitative parametric imaging with application to photoacoustic imaging}, volume = {142}, year = {2025}, }
• (2025). Corrigendum to “An innovative attention infused- BiRecurrenTwin network assisted hybrid segmentation technique for accurate heart disease prediction” [Eng. appl. artif. intell. 138 (2024) 109389]. EAAI, 142, 109918. (www)

  @article{ , archive = {J_EAAI}, author = {R. Raja Aswathi and K. Pazhanikumar and B. Ramakrishnan}, doi = {10.1016/j.engappai.2024.109918}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109918}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Corrigendum to “An innovative attention infused- BiRecurrenTwin network assisted hybrid segmentation technique for accurate heart disease prediction” [Eng. appl. artif. intell. 138 (2024) 109389]}, volume = {142}, year = {2025}, }
• (2025). Contrastive prototype learning with semantic patchmix for few-shot image classification. EAAI, 142, 109917. (www)

  @article{ , abstract = {Few-shot image classification aims to learn unseen classes with only a few training samples for each class. However, most existing models still suffer from weak feature representation due to data scarcity. To this end, a novel contrastive learning framework is proposed for few-shot image classification that utilizes patch-wise and class-wise features. Concretely, a semantic patchmix scheme is designed to effectively capture patch-wise features with more discriminative representation. Specifically, a new information noise contrastive estimation loss with modulating factor is proposed to adjust the weights of samples, which is adaptive and trades off different samples. For class-wise features, contrastive prototype learning on two correlated views is leveraged to enhance the generalization of representations. Experiments demonstrate that our method achieves competitive performances on five popular datasets for few-shot image classification. In particular, our method brings a 1.74% improvement in accuracy over state-of-the-art methods on 5-way 1-shot.}, archive = {J_EAAI}, author = {Mengping Dong and Fei Li and Zhenbo Li and Xue Liu}, doi = {10.1016/j.engappai.2024.109917}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109917}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Contrastive prototype learning with semantic patchmix for few-shot image classification}, volume = {142}, year = {2025}, }
• (2025). A knowledge-driven scatter search algorithm for the distributed hybrid flow shop scheduling problem. EAAI, 142, 109915. (www)

  @article{ , abstract = {The disparity in means of production and costs across different regions has prompted manufacturing companies to develop intelligent distributed production scheduling methods. The distributed hybrid flow shop scheduling problem (DHFSP) is a fundamental issue in the process of distributed production scheduling. Exploring effective methods to address the DHFSP is crucial for enhancing the efficiency of real-world distributed production scheduling. A knowledge-driven scatter search algorithm (KSS) is proposed to improve the efficiency of solving the DHFSP by leveraging domain knowledge. An enhanced reinforcement learning approach is integrated into the scatter search algorithm, leveraging knowledge at different search stages to drive evolutionary power. Simultaneously, a depth-variable search strategy based on knowledge is designed to explore the optimal solution by leveraging neighborhood knowledge at different depths. The development of an evaluation method that integrates problem and algorithm characteristics is undertaken to enhance the efficacy of learning. The experimental results demonstrate that KSS outperforms other algorithms in solving the DHFSP.}, archive = {J_EAAI}, author = {Yang Zuo and Fuqing Zhao and Jianlin Zhang}, doi = {10.1016/j.engappai.2024.109915}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109915}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A knowledge-driven scatter search algorithm for the distributed hybrid flow shop scheduling problem}, volume = {142}, year = {2025}, }
• (2025). Weighted symmetric nonnegative matrix factorization and graph-boosting to improve the attributed graph clustering. EAAI, 142, 109914. (www)

  @article{ , abstract = {Graph clustering is a critical task in network analysis, aimed at grouping nodes based on their structural or attributed similarities. In particular, attributed graph clustering, which considers both structural links and node attributes, is essential for complex networks where additional node information is available. Nonnegative Matrix Factorization (NMF) has shown promise in graph clustering; however, it faces limitations when applied to attributed graph clustering, such as an inability to detect outliers, distortion of geometric data point structures, and disregard for attributed information. Moreover, many existing attributed graph clustering methods overlook distant node relationships due to network sparsity, which hinders further performance improvements. To address these challenges, this paper introduces Weighted Symmetric NMF with graph-Boosting for attributed Graph Clustering (WSBGC), an innovative extension of NMF. WSBGC leverages attribute similarity among nodes to create a weighted version of NMF, enabling outlier detection while preserving the geometric structure of data points. Additionally, WSBGC employs graph-boosting, leveraging attribute information to account for distant node relationships and improve clustering accuracy. A graph attention auto-encoder is then used to construct the final clustering model. The effectiveness of WSBGC is validated through extensive experiments on real-world datasets. Notably, our algorithm improves accuracy by 2.5% compared to the best available method, demonstrating its superior clustering performance in attributed graphs.}, archive = {J_EAAI}, author = {Shunlei Li and Lili Wan and Yin Zhang and Lixia Luo}, doi = {10.1016/j.engappai.2024.109914}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109914}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weighted symmetric nonnegative matrix factorization and graph-boosting to improve the attributed graph clustering}, volume = {142}, year = {2025}, }
• (2025). Hyper boundary conditions: Data-driven operator for boundary value problems. EAAI, 142, 109913. (www)

  @article{ , abstract = {Applying machine learning methods and Neural Operators to solve differential equations has proven effective in generating reliable surrogates. However, incorporating boundary conditions still poses a significant challenge to the training process. While it is possible to integrate the equations into a physics-informed training approach, we propose a novel explicit data-driven representation of the boundary conditions for learning operators. The mathematical data associated with the boundary conditions allows the solution of multiple instances of boundary value problems as a function of training data. The latent representation of the boundary conditions forms a family of neural models, each solving a particular boundary value problem. Our approach, Hyper Boundary Conditions, employs a hypernetwork approach to use the latent boundary condition representation to validate multiple constraint instances. Our method presented high accuracy and low computational cost compared to another classical Neural Operator approach, whose validation employed several prediction and generalization experiments.}, archive = {J_EAAI}, author = {Mateus Mussi Brugnolli and Leonardo Correia and Bruno A. Angélico and João F. Justo}, doi = {10.1016/j.engappai.2024.109913}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109913}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hyper boundary conditions: Data-driven operator for boundary value problems}, volume = {142}, year = {2025}, }
• (2025). Reshaping the future of sports with artificial intelligence: Challenges and opportunities in performance enhancement, fan engagement, and strategic decision-making. EAAI, 142, 109912. (www)

  @article{ , abstract = {The rapid advancement of artificial intelligence has significantly transformed the sports industry over the past decade. In sports performance, artificial intelligence-driven analytics has become essential for optimizing athlete training, injury prevention, and performance enhancement, with sophisticated algorithms analyzing player data to develop personalized training programs and identify areas for improvement. The impact of artificial intelligence on fan engagement has been profound, enabling the delivery of highly personalized content and recommendations that foster stronger connections between fans and their favorite teams or athletes. Leveraging artificial intelligence-powered algorithms, ticket pricing strategies are optimized, resulting in increased ticket sales and revenue. Artificial intelligence-based performance evaluation tools assist sports managers in making well-informed decisions regarding team composition and tactics. This study shows the use of an artificial neural network with artificial intelligence technologies, including data analysis, to improve sports performance, management, and decision-making. Data-driven insights aid in talent scouting and recruitment, allowing managers to identify promising athletes with precision. Issues about data privacy, algorithmic bias, and fair competition need to be addressed to ensure responsible and equitable utilization of artificial intelligence in augmenting the sports industry. The integration of artificial intelligence technologies has profoundly transformed both sports and sports management. From enhancing athlete performance and fan engagement to optimizing administrative tasks and strategic decision-making, the impact of artificial intelligence on the sports domain continues to expand. Results show significant enhancements in all areas, with artificial intelligence outperforming data analysis alone. Prediction errors of the artificial neural network model align well with experimental targets.}, archive = {J_EAAI}, author = {Ting Xu and S. Baghaei}, doi = {10.1016/j.engappai.2024.109912}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109912}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reshaping the future of sports with artificial intelligence: Challenges and opportunities in performance enhancement, fan engagement, and strategic decision-making}, volume = {142}, year = {2025}, }
• (2025). Pareto selective error feedback suppression for popularity–diversity balanced session-based recommendation. EAAI, 142, 109911. (www)

  @article{ , abstract = {Session-based recommendation approaches have garnered substantial attention for their ability to deliver tailored recommendations across diverse domains. However, balancing recommendation accuracy and diversity remains a significant challenge, with most research focusing primarily on accuracy at the expense of diversity. In this paper, we address this challenge by introducing a novel approach, the P areto selective e rror feedback s uppression a lgorithm (Pesa). This model-agnostic back-propagation algorithm simultaneously enhances recommendation accuracy and diversity by strategically modifying gradient distributions during training. Specifically, Pesa operates by randomly freezing the gradients of diverse items during back-propagation, thereby suppressing error feedback selectively and enabling the model to better capture critical features of these items while maintaining a balance between accuracy and diversity. Comprehensive evaluations on four benchmark datasets demonstrate the superiority of Pesa, obtaining relative improvement in accuracy, while achieving a 10% increase in diversity compared to state-of-the-art baselines. These results highlight the effectiveness of our approach in addressing the trade-off between accuracy and diversity in session-based recommendation systems. We provide our source code at: https://github.com/Jackymeister/Pesa .}, archive = {J_EAAI}, author = {Yanglei Gan and Qiao Liu and Da Luo and Rui Hou and Yuxiang Cai and Run Lin and Xiaojun Shi}, doi = {10.1016/j.engappai.2024.109911}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109911}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pareto selective error feedback suppression for popularity–diversity balanced session-based recommendation}, volume = {142}, year = {2025}, }
• (2025). Artificial intelligence and machine learning in mechanical engineering: Current trends and future prospects. EAAI, 142, 109910. (www)

  @article{ , abstract = {This review examines the transformative influence of artificial intelligence (AI) and machine learning (ML) on mechanical engineering, emphasizing application-specific advancements that have contributed to the field's progress. Key applications, including predictive maintenance, design optimization, structural health monitoring, quality control, and renewable energy optimization, illustrate how AI techniques, including reinforcement learning, deep learning, and neural networks, improve efficiency, reduce costs, and promote sustainable practices. This review also looks at the more general, overarching features of AI, such as its ability to adapt, to be interpreted, and to combine physics-based and data-driven models, which makes it easier to use these applications in a range of engineering settings. Despite significant advancements, challenges remain, including the model's robustness, computational demands, ethical considerations, and data quality. This paper endeavors to provide a comprehensive resource for researchers and practitioners in the mechanical engineering domain by synthesizing current advancements, identifying critical challenges, and predicting future trajectory.}, archive = {J_EAAI}, author = {Madhu Puttegowda and Sharath Ballupete Nagaraju}, doi = {10.1016/j.engappai.2024.109910}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109910}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence and machine learning in mechanical engineering: Current trends and future prospects}, volume = {142}, year = {2025}, }
• (2025). Reinforced fuzzy neural networks based on maximum entropy clustering and conjugate gradient method. EAAI, 142, 109909. (www)

  @article{ , abstract = {This paper presents a novel fuzzy clustering-based Takagi–Sugeno–Kang (TSK) neural network for classification tasks, aiming to enhance the accuracy of existing models through a more effective design approach. Unlike previous fuzzy neural networks (FNNs) models, our classifier leverages the maximum entropy clustering algorithm to establish the first layer structure, while updating the hidden layer weights to address limitations of traditional clustering methods. Additionally, to alleviate computational burden, we employ the conjugate gradient method to adaptively adjust connection parameters of the output layer, employing a softmax function to represent probability distribution. Furthermore, we introduce the cross-entropy error function as the objective function and include an L 2 regularization term to mitigate multicollinearity, thereby preventing overfitting and enhancing generalization performance. Experimental evaluations on two-dimensional and other machine learning datasets demonstrate the superiority of our model through comprehensive comparative analyses, highlighting the impact of various parameters on classifier performance. By implementing this new model, we explore the potential of combining fuzzy clustering with TSK neural networks for applications in the fields of pattern recognition and optimization of intelligent systems.}, archive = {J_EAAI}, author = {Qingmei Dong and Qinwei Fan and Zhiwei Xing}, doi = {10.1016/j.engappai.2024.109909}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109909}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reinforced fuzzy neural networks based on maximum entropy clustering and conjugate gradient method}, volume = {142}, year = {2025}, }
• (2025). Pseudo-label attention-based multiple instance learning for whole slide image classification. EAAI, 142, 109908. (www)

  @article{ , abstract = {Automating disease classification in whole slide images (WSIs) is crucial for improving clinical diagnostic efficiency. However, existing multiple instance learning (MIL) approaches for this task often struggle with challenges such as insufficient focus on positive regions and data imbalance between positive and negative regions. These issues can lead to suboptimal performance in practical applications. To address these problems, in this paper, we propose a novel embedding-based MIL technique called pseudo-label attention-based multiple instance learning (PAMIL). PAMIL aggregates each instance’s features regarding their contributions to improving downstream classification performance. The key insight of PAMIL involves training the model in a supervised manner by introducing pseudo-labels to emphasize positive regions. Additionally, we propose a fine-tuning strategy to effectively refine the dataset, eliminating the interference of false-positive data and alleviating data imbalance. The effectiveness of PAMIL was demonstrated through comparisons with six state-of-the-art MIL techniques across two large-scale, real-world datasets. Empirical results show that the proposed method outperforms other methods, achieving up to a 2.15% improvement in accuracy and a 1.61% increase in area under the curve (AUC) on the Cancer Genome Atlas Non-Small Cell Lung Cancer (TCGA-NSCLC) dataset, highlighting the superiority of our method in practical applications, such as helping clinicians diagnose quickly.}, archive = {J_EAAI}, author = {Jing He and Ping Wang and Jingwen Cai and Dan Tang and Shaowen Yao and Renyang Liu}, doi = {10.1016/j.engappai.2024.109908}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109908}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Pseudo-label attention-based multiple instance learning for whole slide image classification}, volume = {142}, year = {2025}, }
• (2025). A collaborative multi-objective meta-heuristic for deadline-constrained multi-workflows scheduling in cloud environment. EAAI, 142, 109907. (www)

  @article{ , abstract = {As cloud computing technology rapidly advances, various organizations are increasingly migrating their complex workflow applications to cloud platforms. Given the diverse demands of users, multi-objective scheduling for multiple workflows has become both necessary and challenging. To tackle the deadline-constrained multi-objective multi-workflows scheduling problem, this paper introduces a collaborative multi-objective meta-heuristic (CMOMH), aiming to optimize a set of schedules that adhere to deadline constraint of each workflow while minimizing total execution time and cost. Within CMOMH, a bi-population co-evolutionary mechanism is crafted to prevent information exchange between feasible and infeasible schedules. A constraint-aware objective function is employed to evaluate infeasible schedules, and a differential evolution-based framework is leveraged to optimize the feasible ones. To maintain the task-to-virtual machine mapping, a sampling-based crossover operator is utilized to explore non-dominated schedules using a multi-point crossover strategy and a probability model. Additionally, two perturbation operators are integrated to enhance population diversity. Simulation results reveal that CMOMH significantly outperforms related state-of-the-art algorithms at the 95% confidence level, demonstrating its effectiveness in addressing the multi-objective scheduling challenges in cloud environments.}, archive = {J_EAAI}, author = {Shuo Qin and Zhongshi Shao}, doi = {10.1016/j.engappai.2024.109907}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109907}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A collaborative multi-objective meta-heuristic for deadline-constrained multi-workflows scheduling in cloud environment}, volume = {142}, year = {2025}, }
• (2025). Frequency-guided dual-collapse transformer for low-light image enhancement. EAAI, 142, 109906. (www)

  @article{ , abstract = {Inadequate exposure of imaging devices in low-light environments results in a loss of image information, significantly deteriorating the image quality. However, current low-light image enhancement algorithms commonly suffer from issues such as color distortion and loss of fine details and textures. In this paper, we propose a frequency-guided dual-collapse Transformer (FDCFormer) network. First, in response to color distortion after enhancement, we propose a dual-collapse Transformer that effectively aggregates features from both spatial and channel dimensions, thus capturing global information. Subsequently, because relying solely on enhancement in the spatial domain often makes it difficult to preserve fine details and textures, we design multiple mixed residual fast Fourier transform blocks as additional frequency information guidance branches, focusing on local detail information at the image edges. Additionally, we employ an adaptive dual-domain information fusion module that combines spatial domain and frequency domain information to enrich the final output features. Extensive experiments on multiple publicly available datasets demonstrate that our FDCFormer outperforms state-of-the-art methods, exceeding Retinexformer by up to 0.93 dB on average across five paired datasets. We also employ our method as a preprocessing step in dark detection, our method improves mean average precision (mAP) by 1.9% over the baseline model on ExDark dataset, revealing the latent practical values of our method. The corresponding codes will be available at https://github.com/Fly175/FDCFormer .}, archive = {J_EAAI}, author = {Jianpu Lin and Fangwei Lai and Shanling Lin and Zhixian Lin and Tailiang Guo}, doi = {10.1016/j.engappai.2024.109906}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109906}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Frequency-guided dual-collapse transformer for low-light image enhancement}, volume = {142}, year = {2025}, }
• (2025). Reconstruction and fusion: Using pseudo physiological modality for pain recognition. EAAI, 142, 109905. (www)

  @article{ , abstract = {Pain often manifests alongside changes in physiological features. The detection of physiological modalities often requires specialized contact-based medical devices. Researchers have explored the possibility of using generated pseudo-physiological modality. However, the generated physiological signals may have low confidence due to various video lighting and change of face positions. This paper presents a novel model named Pseudo-physiological Modality Reconstruction and Fusion Network for pain assessment utilizing both video and pseudo-physiological modality signals. The confidence of the generated signals is estimated and considered. The unreliable pseudo-physiological modality features are replaced by randomized vectors and reconstructed. The modal reconstruction module utilizes Variational Autoencoder based intra-modal reconstruction units and cross-modal reconstruction units to learn modal features and the relation between different modalities. Experimental validation conducted on two widely used public datasets corroborates the effectiveness of the proposed method, demonstrating its potential for practical application in automatic pain assessment systems.}, archive = {J_EAAI}, author = {Mingze Sun and Haoxiang Wang and Bilian Li and Jianming Lv and Guancheng Yao and Yuhao Han}, doi = {10.1016/j.engappai.2024.109905}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109905}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reconstruction and fusion: Using pseudo physiological modality for pain recognition}, volume = {142}, year = {2025}, }
• (2025). Prediction of the load–moment interaction curve for rectangular concrete-filled steel tube columns using composite multi-fidelity neural network. EAAI, 142, 109904. (www)

  @article{ , abstract = {Rectangular concrete-filled steel tubes (RCFSTs) are widely utilized structural solutions due to their exceptional mechanical behavior under combined compression and bending, characterized by the load–moment ( N – M ) interaction curve. However, with the advancement of high-performance materials, existing codes are inapplicable, and the scarcity of pure bending experimental data constrains the development of general calculation formulae. In this study, a composite multi-fidelity neural network (CMFNN) is proposed to predict the parabolic N – M curve by incorporating a high-fidelity network and an empirical bending strength method. A total of 1005 axial compression tests and 401 eccentric compression tests collected from a comprehensive literature review are employed to train the proposed model. Compared with specifications and a back propagation neural network, the high-fidelity network, which predicts axial strength through the sequential acquisition of concrete and steel coefficients, demonstrates superior accuracy, with a mean absolute percentage error (MAPE) of 6.04%. To capture the curve's transition with slenderness, the eccentric compression data is subdivided into three subsets. This approach significantly improves prediction performance, especially in high-slenderness tests. Each subset exhibits a coefficient of determination (R 2 ) above 95% and a MAPE below 8.72%, outperforming traditional methods. The analysis of the feature importance conducted by the SHapley Additive exPlanations (SHAP) algorithm indicates that the primary factors affecting the curve shape shift from the strength and thickness of steel tube to concrete strength and specimen length as slenderness increases. Additionally, sensitivity analysis demonstrates that height feature predominantly impacts eccentric strength, providing effective guidance for practical engineering applications.}, archive = {J_EAAI}, author = {Yixuan Li and Junbin Lou and Qian Feng and Guannan Wang and Rongqiao Xu}, doi = {10.1016/j.engappai.2024.109904}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109904}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction of the load–moment interaction curve for rectangular concrete-filled steel tube columns using composite multi-fidelity neural network}, volume = {142}, year = {2025}, }
• (2025). An advanced multi-source data fusion method utilizing deep learning techniques for fire detection. EAAI, 142, 109902. (www)

  @article{ , abstract = {Fire poses a major risk to ecosystems, human life, and property. Due to the complexity of fire scenarios and the challenge of detecting small, dispersed targets, this paper proposes FireSmoke-YOLO, an innovative fire detection algorithm based on the You Only Look Once (YOLO) framework. Firstly, the Funnel Space Pyramid Pooling Fast (FSPPF) is proposed to replace the Spatial Pyramid Pooling Fast (SPPF) in the backbone module, enhancing the feature extraction of deep networks. Secondly, to improve the detection accuracy of small objects, a Small Object Detection Layer is integrated into the Neck module. Additionally, to better fuse deep and shallow features, a Dynamic Snake Convolution (DSC) is added after each concatenation layer in the Neck module. FireSmoke-YOLO was trained and evaluated with a custom fire dataset. Experimental results show FireSmoke-YOLO achieves 81.4% a mean average precision at an intersection over union threshold of 0.5 ( mAP50 ) and 59% mAP50-95 , surpassing the original model by 1.9% and 8.9%, respectively. The FSPPF layer reduces model size by 4.4 megabytes, ensuring a lightweight design. The model shows strong generalization in processing satellite remote sensing data, validating its effectiveness in fire monitoring. This study enhances fire detection accuracy while providing an efficient, lightweight model for firefighting and environmental protection. FireSmoke-YOLO presents a new approach to fire monitoring technology, with broad application potential. Future work will focus on further optimization and expanding its use in diverse fire monitoring scenarios to improve public safety and ecological protection.}, archive = {J_EAAI}, author = {Shikuan Wang and Mengquan Wu and Xinghua Wei and Xiaodong Song and Qingtong Wang and Yanchun Jiang and Jinkun Gao and Lingyi Meng and Zhipeng Chen and Qiyue Zhang and Yike Zhang and Shuang Liang and Shengtao Wei and Longxing Liu and Changbo Yi and Jinyi Lv}, doi = {10.1016/j.engappai.2024.109902}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109902}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An advanced multi-source data fusion method utilizing deep learning techniques for fire detection}, volume = {142}, year = {2025}, }
• (2025). Deep learning inspired game-based cognitive assessment for early dementia detection. EAAI, 142, 109901. (www)

  @article{ , abstract = {This paper introduces a gaming approach inspired by deep learning for the early detection of dementia. This research employs a convolutional neural network (CNN) model to analyze health metrics and facial images via a cognitive assessment gaming application. We have collected 1000 samples of health metric data from Apollo Diagnostic Center and hospitals, labeled “demented” or “nondemented,” to train a modified 1-dimensional convolutional neural network (MOD-1D-CNN) for game level 1. Additionally, a dataset of 1800 facial images, also labeled “demented” or “non-demented,” is collected in our work to train a modified 2-dimensional convolutional neural network (MOD-2D-CNN) for game level 2. The MOD-1D-CNN has achieved a loss of 0.2692 and an accuracy of 70.50% in identifying dementia traits via health metric data; in comparison, the MOD-2D-CNN has achieved a loss of 0.1755 and an accuracy of 95.72% in distinguishing dementia from facial images. A rule-based linear weightage method combines these models and provides a final decision. In addition, a better fusion neural network strategy is also explored in the results analysis with an ablation study. The proposed models are computationally efficient alternatives with significantly fewer parameters than other state-of-the-art models. The performance and parameter counts of these models are compared with those of existing deep learning models, emphasizing the role of AI in enhancing early dementia.}, archive = {J_EAAI}, author = {Paramita Kundu Maji and Soubhik Acharya and Priti Paul and Sanjay Chakraborty and Saikat Basu}, doi = {10.1016/j.engappai.2024.109901}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109901}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep learning inspired game-based cognitive assessment for early dementia detection}, volume = {142}, year = {2025}, }
• (2025). Dual spectral-spatial residual adaptive network for hyperspectral image classification in the presence of noisy labels. EAAI, 142, 109900. (www)

  @article{ , abstract = {In real-world scenarios, Hyperspectral Image (HSI) datasets introduce potential noise inaccuracies due to multiple annotators. Label noise poses a significant challenge for practical deep learning, yet this issue is largely unexplored. Existing methods, which attempt to clean the noisy labelled data to increase classification accuracy, are computationally expensive and face the risk of removing correctly labelled data. In contrast, other methods that work with noisy labelled data but attempt to minimise the noise impact on classification by formulating a robust loss function lose classification accuracy when the ratio of incorrectly to correctly labelled data is high. This work proposes a Dual Spectral-Spatial Residual Adaptive (DSSRA) network to minimise the noise effect even when the amount of noisy labelled data is high. It offers the following contributions: (1) effective salient feature extraction modules to enhance the discriminatory representation of different classes in the proposed DSSRA network; (2) an adjusted noise tolerance loss (ANTL) function that down-weights the impact of learning with noisy labels. ANTL combines normalised focal loss and reverse cross-entropy to counter label noise; and (3) extensive testing on noisy versions of several benchmark HSI datasets. The results show that our DSSRA model outperforms the state-of-the-art HSI classification methods in handling noisy labels, offering a robust solution for real-world applications.}, archive = {J_EAAI}, author = {Kwabena Sarpong and Mohammad Awrangjeb and Md. Saiful Islam}, doi = {10.1016/j.engappai.2024.109900}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109900}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual spectral-spatial residual adaptive network for hyperspectral image classification in the presence of noisy labels}, volume = {142}, year = {2025}, }
• (2025). Comparative analysis of mechanical behavior prediction models for abdominal aortic wall. EAAI, 142, 109899. (www)

  @article{ , abstract = {This study conducted a comparative analysis of various modeling methods to predict the mechanical behavior of the abdominal aortic wall. Based on the experimental data, datasets were generated using exponential and Gasser-Ogden-Holzapfel (GOH) models. The following datasets were then used to train three models: least squares regression, Markov chain Monte Carlo algorithm, and deep neural network (DNN). The performance of each model was evaluated using the R 2 value, and group k-fold cross-validation was performed to analyze the generalization performance of the DNN. Consequently, the DNN exhibited superior predictive accuracy, which can be attributed to its ability to learn complex data patterns and effectively handle nonlinearity. The exponential model demonstrated higher accuracy than the GOH model under specific conditions, particularly for data with small deformations and rapid stress variations. In contrast, the GOH model exhibited limitations in response to its complex parameter settings and a lack of consistent performance.}, archive = {J_EAAI}, author = {Ji-Hun You and Chiseung Lee}, doi = {10.1016/j.engappai.2024.109899}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109899}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comparative analysis of mechanical behavior prediction models for abdominal aortic wall}, volume = {142}, year = {2025}, }
• (2025). Estimating travel time in transport network with a combined multi-attributed graph convolutional neural network and multilayer perceptron model. EAAI, 142, 109898. (www)

  @article{ , abstract = {In this article, an advanced model for forecasting travel time in road networks is presented, employing a Graph Convolutional Neural Network (GCN) integrated with a Multilayer Perceptron (MLP), focusing on the travel time ratio (TTR). Utilizing data from Poland's Mazovia region, the model analyzes spatial and temporal dependencies in road infrastructure, incorporating road conditions, weather, and time for accurate traffic prediction. The research involved detailed analysis of neuron configurations and learning rates in the GCN-MLP model, focusing on their impact on mean squared error (MSE) and other key performance metrics. The optimized model configurations achieved a Mean Absolute Percentage Error (MAPE) of approximately 8%, outperforming many models in existing literature. The study found that hours, days, weather conditions, and technical road features significantly influence the model's outcomes. Efforts to optimize the model included excluding attributes with minimal impact on MSE, leading to notable improvements in efficiency. A comparison between the GCN-MLP model and traditional MLP models demonstrated the former's superior effectiveness, particularly in achieving lower MSE values and better predictive accuracy. These results highlight the potential and challenges of applying advanced machine learning techniques in traffic management, indicating a significant step forward in this field.}, archive = {J_EAAI}, author = {Igor Betkier}, doi = {10.1016/j.engappai.2024.109898}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109898}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Estimating travel time in transport network with a combined multi-attributed graph convolutional neural network and multilayer perceptron model}, volume = {142}, year = {2025}, }
• (2025). Picture fuzzy compromise ranking of alternatives using distance-to-ideal-solution approach for selecting blockchain technology platforms in logistics firms. EAAI, 142, 109896. (www)

  @article{ , abstract = {The emergence of blockchain technology (BT) has indicated a new era for the supply chain and logistics industries wherein uncertainty significantly affects the establishment of suitable blockchain platforms for logistics firms. Moreover, uncertainty widely occurs in such realistic situations. This study develops a hybrid picture fuzzy information-based framework for assessing and ranking the blockchain platforms in the logistics firm. The experts' significance values are obtained using a distance measure and a rank sum model-based process under the picture fuzzy set (PFS) context. A novel distance measure is offered for PFSs, and its effectiveness over previously developed measures is demonstrated. To estimate the criteria weights, an integrated weighting model is utilized, which is based on the criteria importance through inter-criteria correlation (CRITIC) method for objective weights and the ranking comparison (RANCOM) approach for subjective weights. Moreover, the picture fuzzy compromise ranking of alternatives from distance to ideal solution (CRADIS) approach is utilized to rank the alternatives. To illustrate the superiority and practicality of the model, it is applied to a case study of blockchain platform selection including six alternatives and thirteen criteria to assess and rank the BTs options based on PFSs. The findings demonstrate that “Microsoft (1.0)” emerges as the best choice with the highest compromise measure among a set of six blockchain platforms in the logistics firm. Finally, sensitivity and comparative analyses are conducted to test the robustness and superiority of the developed method.}, archive = {J_EAAI}, author = {Pratibha Rani and Arunodaya Raj Mishra and Ahmad M. Alshamrani and Adel Fahad Alrasheedi and Erfan Babaee Tirkolaee}, doi = {10.1016/j.engappai.2024.109896}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109896}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Picture fuzzy compromise ranking of alternatives using distance-to-ideal-solution approach for selecting blockchain technology platforms in logistics firms}, volume = {142}, year = {2025}, }
• (2025). Adaptive reconstruct feature difference network for open set domain generalization fault diagnosis. EAAI, 142, 109895. (www)

  @article{ , abstract = {Data-driven deep learning methods for intelligent fault diagnosis are highly regarded for their capability to perform end-to-end diagnosis. However, their practical application is often constrained by challenges such as dependency on extensive labeled data, inconsistent data distributions, and the inability to handle unknown test conditions. To address these limitations, this paper introduces an Adaptive Feature Reconstruction Difference Network (AFRDN) designed for open-set domain generalization in the fault diagnosis of rotating machinery. AFRDN achieves robust performance through three key innovations: first, it maps multi-source domain features to a common standard normal distribution, enhancing feature alignment and enabling generalization to unseen target domains. Second, it employs a feature reconstructor to calculate reconstruction differences, adaptively learning decision thresholds for each known health condition category. Finally, the network integrates these components to build a domain generalization model capable of accurately detecting faults, including unknown categories, under previously unseen operating conditions. The effectiveness of the proposed AFRDN was evaluated through extensive experiments on both public and private rotating machinery datasets. The method achieved average accuracies of 90.59% and 85.10%, respectively, outperforming several state-of-the-art open-set domain generalization methods. AFRDN demonstrated superior capabilities in detecting unknown fault categories and improving generalization performance, making it a promising solution for real-world fault diagnosis challenges.}, archive = {J_EAAI}, author = {Mengyu Liu and Zhe Cheng and Yu Yang and Niaoqing Hu and Guoji Shen and Yi Yang}, doi = {10.1016/j.engappai.2024.109895}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109895}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive reconstruct feature difference network for open set domain generalization fault diagnosis}, volume = {142}, year = {2025}, }
• (2025). Artificial intelligence-based approach for water-cut prediction in crude oil processing. EAAI, 142, 109892. (www)

  @article{ , abstract = {The production and handling of oil and gas at an offshore location present significant challenges, and it is equally important to meet statutory requirements and fulfill customer needs. Oil field operators must separate both emulsion phases at faster rates and efficiently to meet the water-cut requirements of customer refineries or transportation barges. Effective separation of the effluent water from the produced fluid results in cleaner crude oil production. The lower the percentage of water in the dispatched crude oil, the lower the penalty imposed by downstream companies, and leading to higher net production and increased profitability. Predicting the product's output qualities, i.e., the water–cut, do well in various ways. It leads to better and optimized use of resources (such as hot utilities and demulsifiers) and a lower corrosion effect. Artificial intelligence (AI) enhances engineering by optimizing processes, improving safety, enabling predictive maintenance, and supporting data-driven decisions in complex industries. The role of AI in the present scenario may reasonably help predict the water–cut parameter in advance. The proposed work aims at achieving a unique integration of some of the nature-inspired algorithms, Genetic Algorithm and Particle Swarm Optimization, into the working framework of Artificial Neural Networks to estimate the performance of the proposed algorithms and in predicting the final water – cut.}, archive = {J_EAAI}, author = {Akash Das and Rahul Sudhanshu and Gaurav Shukla and Nitin Dutt Chaturvedi}, doi = {10.1016/j.engappai.2024.109892}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109892}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligence-based approach for water-cut prediction in crude oil processing}, volume = {142}, year = {2025}, }
• (2025). Knowledge augmented expert finding framework via knowledge graph embedding for community question answering. EAAI, 142, 109891. (www)

  @article{ , abstract = {Expert Finding in Community Question Answering aims to recommend appropriate experts to answer posted questions. However, existing approaches focus on exploiting semantic extraction and authority analysis techniques, failing to realize the latent knowledge-aspect connections between questions and experts. Therefore, the experts recommended for posted questions are limited to text-to-text approximation or domain-independent authority. In this study, we propose a Knowledge Augmented Expert finding framework (KAExpert) that introduces knowledge level information into semantic-based and authority-based expert finding method. A community knowledge graph is firstly constructed by acquiring ternary relations from public databases based on question tags and knowledge-related phrases. Then a flexible knowledge graph embedding is designed to extract the matching relationship between questions and experts at the knowledge level. Along this line, the knowledge-level authority is calculated based on the knowledge graph embedding to optimize the results of domain matching. According to knowledge graph embedding and knowledge-level authority, KAExpert is constructed to optimize the results of domain matching so that the found experts set consists of high-level expert with semantic matching and knowledge matching. Finally, experimental results on two real-world datasets collected from two major commercial question answering web sites show that KAExpert can outperform baseline methods with a significant margin.}, archive = {J_EAAI}, author = {Yue Liu and Zitu Liu and Zhenyao Yu and Qingshan Fu and Weize Tang and Wenxuan Yao and Zhibin Sun}, doi = {10.1016/j.engappai.2024.109891}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109891}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Knowledge augmented expert finding framework via knowledge graph embedding for community question answering}, volume = {142}, year = {2025}, }
• (2025). Development of residual learning in deep neural networks for computer vision: A survey. EAAI, 142, 109890. (www)

  @article{ , abstract = {Deep neural networks (DNNs) have significantly advanced computer vision tasks such as image classification, object detection, and semantic segmentation. Residual learning, a key principle in DNN design, plays a crucial role in optimizing training and improving model performance. By incorporating skip connections, residual learning enables more efficient optimization, resulting in better accuracy during both training and testing. While many surveys have addressed specific deep learning techniques, the role of residual learning in shaping the future of DNNs for computer vision remains underexplored. This survey provides a comprehensive overview of the evolution and impact of skip connections in DNNs. We introduce the core concepts of residual learning, explore theoretical and experimental evidence supporting its effectiveness, and discuss its implications for future research directions. Additionally, we review key papers, models, datasets, and applications of residual learning across image classification, object detection, semantic segmentation, and image reconstruction. We hope this survey will inspire further exploration and innovation in residual learning techniques, both for applications of artificial intelligence in engineering and for theoretical advancements. The project page can be found at ResidualLearningSurvey .}, archive = {J_EAAI}, author = {Guoping Xu and Xiaxia Wang and Xinglong Wu and Xuesong Leng and Yongchao Xu}, doi = {10.1016/j.engappai.2024.109890}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109890}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Development of residual learning in deep neural networks for computer vision: A survey}, volume = {142}, year = {2025}, }
• (2025). An audio-based framework for anomaly detection in large-scale structural testing. EAAI, 142, 109889. (www)

  @article{ , abstract = {FastBlade is a research facility that tests large-scale composite and metal structures. To maximise its throughput by uninterrupted running of experiments, unmanned operation of the site is desired. One of its key enablers is anomaly detection, where microphones are used as a non-specific, affordable, and well-established sensing method. The dataset collected during the operation of the system consists of both normal and anomalous samples, which we need to classify. The problems associated with the dataset involve significant intraclass variability of the normal operation samples, as well as the scarcity of anomalous data, increasing the complexity of the classification problem. In this work, we evaluate the performance of several tools for time–frequency signal analysis, which are used to extract features from the original high-dimensional signal. We choose to apply the wavelet scattering transform (WST) due to its remarkable performance. Based on the findings from the literature review, we first rely on the reconstruction error of the processed WST images to detect anomalous samples. However, due to the nature of the dataset, both the convolutional autoencoder (CAE) and the principal component analysis (PCA) transform turn out to be unsuccessful. We then investigate the hidden layers of the CAE in search of features that can be used to separate normal and anomalous samples. Having identified the most suitable candidates, we discover that applying the normalised cross-correlation (NCC) to measure the similarity of the generic features generated and our dataset results in satisfactory separation. We train a number of classifiers and test the method on unseen data. The model’s accuracy is 99.58%, with a recall of 100% and 92% on normal and anomalous operation samples, respectively. The model’s accuracy and low latency prove the WST’s suitability for robust, real-time detection of different anomaly types. Therefore, the method can be deployed in systems with limited information about the critical assets and can be easily extrapolated to other setups.}, archive = {J_EAAI}, author = {Marek J. Munko and Fergus Cuthill and Miguel A. Valdivia Camacho and Conchúr M. Ó Bradaigh and Sergio Lopez Dubon}, doi = {10.1016/j.engappai.2024.109889}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109889}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An audio-based framework for anomaly detection in large-scale structural testing}, volume = {142}, year = {2025}, }
• (2025). Image reconstruction using machine-learned pseudoinverse in electrical capacitance tomography. EAAI, 142, 109888. (www)

  @article{ , abstract = {Algorithms based on linear approximation are the primary methods for fast online image reconstruction in electrical capacitance tomography (ECT). One-step reconstruction algorithms use certain inverse matrix approximations, such as the normalized transpose matrix or the Moore-Penrose pseudoinverse matrix used in the linear back-projection (LBP) algorithm or mean square optimization, respectively. For the first time, we proposed the use of a machine-learned pseudoinverse matrix (ML-pseudoinverse) as an alternative to these matrices. So far, multivariate linear regression has been used to identify a sensitivity matrix of the forward problem in tomography. Experiments were performed using a cylindrical ECT sensor with sixteen electrodes and flow image patterns typical for two-phase flow monitoring. Training datasets consisting of images of various test objects and their corresponding measurements were generated using numerical simulation of measurements carried out in the MATLAB environment and the ECTsim toolkit. Real measurements were performed using the EVT4 data acquisition system. Reconstructed images were analyzed using selected image quality norms. The reconstruction results obtained for the validation dataset of two-phase flow images showed that the one-step linear reconstruction algorithm achieved better results when using ML-pseudoinverse than LBP normalized transposition or Moore-Penrose pseudoinverse. The proposed approach can be a good solution for a fast reconstruction algorithm for dynamic processes in online monitoring using ECT. Machine-learned pseudoinverse can be applied to numerous engineering problems, including nonlinear inverse problems.}, archive = {J_EAAI}, author = {Damian Wanta and Aleksander Smolik and Waldemar T. Smolik and Mateusz Midura and Przemysław Wróblewski}, doi = {10.1016/j.engappai.2024.109888}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109888}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Image reconstruction using machine-learned pseudoinverse in electrical capacitance tomography}, volume = {142}, year = {2025}, }
• (2025). An efficient large kernel convolution network designed for neural processing unit. EAAI, 142, 109887. (www)

  @article{ , abstract = {As industrial artificial intelligence (AI) applications become increasingly prevalent and the performance of neural processing unit (NPU) is advanced continuously, deploying lightweight networks on NPU has become a favorable option for high real-time industrial scenarios, including autonomous driving, intelligent mine, and more. However, common lightweight networks are not primarily designed for NPU, thereby failing to fully leverage the performance of NPU. In this paper, an NPU lightweight network, called Efficient Large Kernel Convolution Network (ELKCNet) is proposed for downstream tasks, such as semantic segmentation and object detection, which have broad application scenarios and significant potential. By incorporating a portion of large kernel convolutions into the Efficient Large Kernel Convolution (ELKC) block, the network’s receptive field is further enhanced to capture richer contextual information and effectively integrate local and global features. To explore the operational principles of NPU, a design paradigm for NPU lightweight networks is proposed. By testing the inference times of various convolutions and pooling operations on NPU, data support is provided for optimizing and constructing novel NPU lightweight networks. To validate the superiority of the ELKCNet, comparative experiments are conducted on multiple datasets, showing that it outperforms other lightweight networks in object detection and semantic segmentation. Additionally, the ELKCNet is evaluated on the underground mine semantic segmentation dataset to further demonstrate its effectiveness and significant potential in real-time industrial applications.}, archive = {J_EAAI}, author = {Jiawen Wang and Chenfei Liao and Dewei Li and Zhongqi Zhao and Jingchuan Chen and Kehu Yang}, doi = {10.1016/j.engappai.2024.109887}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109887}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient large kernel convolution network designed for neural processing unit}, volume = {142}, year = {2025}, }
• (2025). An advanced physics-informed neural operator for comprehensive design optimization of highly-nonlinear systems: An aerospace composites processing case study. EAAI, 142, 109886. (www)

  @article{ , abstract = {Deep Operator Networks (DeepONet) and their physics-informed variants have shown significant promise in learning mappings between function spaces of partial differential equations (PDEs), enhancing the generalization of traditional neural networks. However, for highly nonlinear real-world applications like aerospace composites processing, existing models often fail to capture underlying solutions accurately and are typically limited to single input functions, constraining rapid process design development. To overcome these limitations this study introduces an advanced physics-informed DeepONet tailored for highly nonlinear systems with multiple input functions. While previous works have primarily focused on single, isolated improvements, our framework is equipped with a comprehensive suite of architectural enhancements including nonlinear decoders, and effective training strategies such as curriculum learning and domain decomposition. The proposed model effectively manages high-dimensional design spaces with substantially enhanced accuracy, surpassing the standard physics-informed DeepONet by two orders of magnitude in training loss and achieving a 30% reduction in prediction error. Its zero-shot prediction capability across a broad design space makes it a powerful tool for accelerating composites process design and optimization, with potential applications in other engineering fields characterized by strong nonlinearity.}, archive = {J_EAAI}, author = {Milad Ramezankhani and Anirudh Deodhar and Rishi Yash Parekh and Dagnachew Birru}, doi = {10.1016/j.engappai.2024.109886}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109886}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An advanced physics-informed neural operator for comprehensive design optimization of highly-nonlinear systems: An aerospace composites processing case study}, volume = {142}, year = {2025}, }
• (2025). A synthetic minority oversampling method with elastic net hypergraph. EAAI, 142, 109885. (www)

  @article{ , abstract = {The standard classifiers perform well on balanced datasets. However, in real-world datasets, most of the data is imbalanced. Therefore, the synthetic minority oversampling technique is an effective and universal solution to this problem. Although synthetic minority oversampling technique and its variants can address the problem of imbalanced classification well in most cases, they cannot fully reflect the high-order information of the data. For this reason, a synthetic minority oversampling method with Elastic Net Hypergraph is proposed in this paper. Firstly, the improved sparse representation Elastic Net is used to establish the hypergraph model, and the noise samples and outlier samples are filtered. Then, the weight of the hyperedge is calculated according to the incidence matrix, and then the oversampling size of each minority sample is obtained. Finally, the Laplacian matrix and Gaussian distribution are calculated to generate new samples and a balanced dataset is obtained. Compared with the popular oversampling methods on 20 datasets, the proposed method shows a significant advantage.}, archive = {J_EAAI}, author = {Pengfei Sun and Zhiping Wang and Peiwen Wang and Kaina Zhao}, doi = {10.1016/j.engappai.2024.109885}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109885}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A synthetic minority oversampling method with elastic net hypergraph}, volume = {142}, year = {2025}, }
• (2025). A cross-modal collaborative guiding network for sarcasm explanation in multi-modal multi-party dialogues. EAAI, 142, 109884. (www)

  @article{ , abstract = {Indirect forms of language, such as sarcasm, are highly prevalent in contemporary human daily communication. While the indirect nature of metaphorical language ensures that speakers can achieve certain pragmatic goals, enabling artificial intelligence to comprehend this way of human communication poses a challenging task. Despite sarcasm detection being an extensively researched area in discourse learning, for dialogue systems to truly understand the underlying meaning of a conversation and generate reasonable responses, mere detection of sarcasm is insufficient. It is crucial to interpret the potential sarcastic meanings to acquire its true essence. Therefore, we propose a cross-modal collaborative guiding network for this study. It takes multimodal (sarcastic) dialogues as input, conducts broadcast mapping on context information from different modalities, enabling cross-modal mutual guidance and understanding. Subsequently, by employing a multi-scale information fusion mechanism to model global relationships within each modality and integrate cross-modal information to complement the overall sarcastic understanding results. Finally, the results are integrated into the bidirectional and auto-regressive transformers framework to generate natural language interpretations of sarcastic dialogues. Notably, through extensive experiments conducted on public datasets, our model achieves top performance across nearly all metrics. Compared to current state-of-the-art multimodal models, our approach demonstrates an average improvement of approximately 2.54% across various evaluation metrics, further validating the advantages of our model over leading methods.}, archive = {J_EAAI}, author = {Xingjie Zhuang and Zhixin Li and Canlong Zhang and Huifang Ma}, doi = {10.1016/j.engappai.2024.109884}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109884}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A cross-modal collaborative guiding network for sarcasm explanation in multi-modal multi-party dialogues}, volume = {142}, year = {2025}, }
• (2025). Spatial–temporal multi-scale interaction for few-shot video summarization. EAAI, 142, 109883. (www)

  @article{ , abstract = {Video summarization is intended to generate a shorter video that conveys the most semantic information of the original video for quick skimming. However, traditional video summarization methods required large amounts of labeled data for training, which may be difficult or costly to obtain, especially in an era where video data are growing at an unprecedented pace. In this paper, a new task namely few-shot video summarization is introduced to address the problem of video summarization from limited labeled data. By requiring only a small number of labeled examples, few-shot video summarization reduced the annotation burden on human annotators. A novel baseline for few-shot video summarization called Spatial-Temporal Multi-scale Interaction (STeMI) is presented to learn models that can generalize well to new tasks or categories with only a few examples. Multi-scale features are extracted from spatial–temporal representations, employing multi-scale interactions in a dual-branch architecture. Several state-of-the-art video summarization methods have been adapted to the few-shot paradigm for equitable evaluation, and several existing datasets have been refined for few-shot segmentation. Extensive experiments on the SumMe and TVSum benchmarks underscored the superiority of STeMI over current video summarization methods. Code is available at https://github.com/zhanzhuxi/STeMI .}, archive = {J_EAAI}, author = {Qun Li and Zhuxi Zhan and Yanchao Li and Bir Bhanu}, doi = {10.1016/j.engappai.2024.109883}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109883}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatial–temporal multi-scale interaction for few-shot video summarization}, volume = {142}, year = {2025}, }
• (2025). Event-triggered predefined-time output feedback fuzzy adaptive control of permanent magnet synchronous motor systems. EAAI, 142, 109882. (www)

  @article{ , abstract = {This paper investigates the event-triggered fuzzy adaptive predefined time control problem for permanent magnet synchronous motor systems with unpredictable states. In control design, fuzzy logic systems (FLSs) approximate unknown nonlinear dynamics, and fuzzy-based state observers estimate unmeasurable states. An event-triggered mechanism utilizing an adaptive backstepping recursion technique is constructed for the controller-to-actuator channel. A predefined time control scheme based on output feedback is proposed, incorporating the tanh function to avoid control singularities. Additionally, a predefined time filter is designed to prevent the explosion of computational complexity. The closed-loop system’s stability is demonstrated using Lyapunov theory and predefined time stability theory, ensuring all signals remain bounded within predefined time intervals. Finally, the effectiveness and feasibility of the proposed adaptive predefined-time event-triggered control method are verified through simulation results.}, archive = {J_EAAI}, author = {Weiqi Liu and Shuai Sui and C.L. Philip Chen}, doi = {10.1016/j.engappai.2024.109882}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109882}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Event-triggered predefined-time output feedback fuzzy adaptive control of permanent magnet synchronous motor systems}, volume = {142}, year = {2025}, }
• (2025). Asymmetric multimodal guidance fusion network for realtime visible and thermal semantic segmentation. EAAI, 142, 109881. (www)

  @article{ , abstract = {In nighttime traffic scenarios, achieving real-time visible and thermal (RGB-Thermal) images semantic segmentation is a pivotal and intricate challenge within autonomous driving systems, demanding a delicate balance between precision and efficiency. However, previous methods have predominantly focused on enhancing model accuracy through increasing the number of parameters while neglecting the burden on inference time. Nevertheless, neglecting the guidance provided by semantic cues during the fusion process inevitably leads to imprecise segmentation. To improve the accuracy and speed of segmentation, this paper introduces a novel Asymmetric Guided Fusion Network (AGFNet) for real-time RGB-Thermal semantic segmentation. Specifically, we design an asymmetric dual-stream encoder to extract RGB and thermal features, thereby decreasing inference time. Afterward, we propose the Multi-scale pooling Cross-select Enhancement Module (MCEM) for mining semantic cues. For feature fusion, we introduce the Guidance Control Fusion Module (GCFM) to leverage semantic cues in guiding the fusion of RGB and thermal features, absorbing complementary relationships between the two modalities. AGFNet achieves 37.4 frames per second (FPS), 57.8% mean Intersection over Union (mIoU), and 48.1 FPS, 56.3% mIoU on the Multi spectral Fusion dataset(MFdataset), achieving a state-of-the-art balance between efficiency and accuracy. Our code will be available at https://github.com/Mask0913/AGFNet .}, archive = {J_EAAI}, author = {Biao Yang and Yuanhui Guo and Rongrong Ni and Yi Liu and Guofa Li and Chuan Hu}, doi = {10.1016/j.engappai.2024.109881}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109881}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Asymmetric multimodal guidance fusion network for realtime visible and thermal semantic segmentation}, volume = {142}, year = {2025}, }
• (2025). A visual data unsupervised disentangled representation learning framework: Contrast disentanglement based on variational auto-encoder. EAAI, 142, 109880. (www)

  @article{ , abstract = {To discover and learn interpretable factors behind the visual data, many approaches use extra regularization terms in learning disentangled representations, which lead to poor results between disentanglement and generative quality. The variational auto-encoder has the ability to learn more semantic information, and the traversal of the generated images along different factor directions shows meaningful and interpretable variations in the latent space. Therefore, we exploit the scalability and training stability of the variational auto-encoder, and focus on meaningful traversal directions in the latent space. We propose contrast disentanglement based on variational auto-encoder, a visual data unsupervised disentangled representation learning framework. Specifically, we explore meaningful interpretable directions in the latent space by constructing the encoder of the typical network module, and then obtain interpretable directions for candidate traversals of target variation images. In this way, the interpretable directions with rich semantic information and disentanglement properties can be obtained. Furthermore, unlike the existing methods, to further improve the ability of interaction and collaborative learning between latent factors and learn more robust and generalized representations, we design the variation space based on disentangled encoders from the contrastive learning perspective to simulate the various variation of the image data, then extract disentangled representations and generate images. Extensive experiments on six disentanglement datasets have demonstrated that the proposed method achieves competing performance on both quantitative metrics and visual quality. Among them, the proposed method achieves better factor variational auto-encoder score and β-variational auto-encoder score (0.97 ± 0.04 and 0.99 ± 0.01, respectively) on the 3Dshapes dataset compared to existing methods.}, archive = {J_EAAI}, author = {Chengquan Huang and Jianghai Cai and Senyan Luo and Shunxia Wang and Guiyan Yang and Huan Lei and Lihua Zhou}, doi = {10.1016/j.engappai.2024.109880}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109880}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A visual data unsupervised disentangled representation learning framework: Contrast disentanglement based on variational auto-encoder}, volume = {142}, year = {2025}, }
• (2025). Dental X-ray image enhancement using a novel evolutionary optimization algorithm. EAAI, 142, 109879. (www)

  @article{ , abstract = {As an optimization problem, the main challenges in the enhancement of dental X-ray images are to perform the tasks of edge detection, noise removal and brightness adjustment in a precise and efficient manner. To overcome these challenges, designing an optimization algorithm that exhibits exploit-exploit behavior in accordance with the geometric structure of the search space of dental images is an important challenge that has not yet been realized. Shortcomings in this area lead to local solution traps and early convergence problems in image enhancement algorithms. This paper introduces the d FDB-LSHADE (dynamic fitness-distance balance-based achievement-history-based adaptive differential evolution with linear population size reduction) algorithm, which is designed for the enhancement of dental X-ray images according to the requirements of the search space of this problem and dynamically changes its exploitation-exploration capabilities. The proposed method is tested on a dataset of 120 periapical images, the most extensive experimental study in the literature. A total of 60 competing algorithms, 53 heuristics and 7 deterministic-based algorithms, were used in the experiments. In the study on dental images, the proposed algorithm has a better Friedman score than all competitors. According to the statistical analysis results obtained from the Wilcoxon pairwise test, the proposed d FDB-LSHADE was able to find better solutions for 23.3% of the images compared to its strongest competitor on the dental image set.}, archive = {J_EAAI}, author = {İbrahim Yildirim and Mustafa Hakan Bozkurt and Hamdi Tolga Kahraman and Sefa Aras}, doi = {10.1016/j.engappai.2024.109879}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109879}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dental X-ray image enhancement using a novel evolutionary optimization algorithm}, volume = {142}, year = {2025}, }
• (2025). Long-term scheduling strategy of hydro-wind-solar complementary system based on chaotic elite selection differential evolution algorithm with death penalty function. EAAI, 142, 109878. (www)

  @article{ , abstract = {Under the main theme of global low-carbon life, clean energy has become an important topic in the market and academic research. However, in order to improve the stability of power generation to the greatest extent possible, we can utilize the good complementary characteristics of multiple energy sources to build a multi-energy complementary power generation system. In order to investigate the long-term scheduling strategy of the hydro-wind-solar complementary system, the scheduling model proposed in this paper takes the maximization of power generation as the optimization objective to optimize the hybrid system including four cascaded hydropower plants of the lower Jinsha River, Daheishan photovolatic power (PV) station and Huidong wind power plant. Through the chaotic elite selection differential evolution algorithm with death penalty function (CEDE-DP) proposed in this paper, the results show that, the monthly water levels of four cascaded hydropower plants fluctuate around the normal storage levels, and the fluctuations of the water levels are concentrated in the months from May to September. Moreover, the complementary characteristics between PV and hydropower are the weakest for wind power in the wet year, especially the Pearson coefficient between PV and Xiluodu/Xiangjiaba Hydropower Station has become almost extinct. From the results, it can be seen that this study provides theoretical basis and practical guidance for realizing efficient scheduling and stable operation of multi-energy systems.}, archive = {J_EAAI}, author = {Yaoyao He and Xiaoyu Hong and Ning Xian}, doi = {10.1016/j.engappai.2024.109878}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109878}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Long-term scheduling strategy of hydro-wind-solar complementary system based on chaotic elite selection differential evolution algorithm with death penalty function}, volume = {142}, year = {2025}, }
• (2025). Intelligent approach for developing a blood product supply chain to mitigate shortages and preclude wastage. EAAI, 142, 109877. (www)

  @article{ , abstract = {Effective management of the Blood Supply Chain (BSC) is crucial for mitigating shortages and preventing wastage, as both issues can have dire consequences for patient health and incur significant financial costs. Challenges such as insufficient access to blood and inefficient utilization underscore the need for a comprehensive evaluation of BSC management alternatives. While various alternatives have been developed, none effectively address all criteria of BSC management, and a systematic assessment to rank these alternatives has been lacking. This study aims to fill this gap by employing a multi-criteria decision-making (MCDM) as an intelligent approach to evaluate and rank BSC management alternatives based on criteria encompassing collection, transportation, production, inventory, and transfusion within a sustainable framework. A novel methodology known as the Fuzzy Weighted Zero Inconsistency (FWZIC) method, based on an interrelationship process and contextualized within a Single-Valued Neutrosophic Fuzzy Set (SvNS), is developed to ascertain the significance of BSC criteria by addressing uncertainties, ambiguity, and interrelationships among criteria. This method is combined with the Grey Correlation Analysis-Vlsekriterijumska Optimizacija I Kaompromisno Resenje (grey-VIKOR) method for ranking alternatives based on seven principal criteria subdivided into 34 sub-criteria. Evaluation of 60 BSC alternatives indicates that the minimum number of donors required to meet demand (C2-7) is the most critical criterion, while minimum transportation cost (C3-1) is the least significant. BSC 41 emerged as the most sustainable alternative, whereas BSC 39 was identified as the least sustainable. These findings highlight the effectiveness of the intilligent approach, supported by systematic ranking, sensitivity analysis, and comparative assessments, thereby underscoring its relevance for enhancing blood supply management.}, archive = {J_EAAI}, author = {Hussein ali khudhyer alhadad and Abdelkarim Elloumi and Hassan A. Alsattar and Sarah Qahtan and Nahia Mourad and Aws Alaa Zaidan and Bilal Bahaa and Vaidyanathan Jayaraman}, doi = {10.1016/j.engappai.2024.109877}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109877}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Intelligent approach for developing a blood product supply chain to mitigate shortages and preclude wastage}, volume = {142}, year = {2025}, }
• (2025). Enhancing generalization in sickle cell disease diagnosis through ensemble methods and feature importance analysis. EAAI, 142, 109875. (www)

  @article{ , abstract = {This work presents a novel approach for selecting the optimal ensemble-based classification method and features with a primarly focus on achieving generalization, based on the state-of-the-art, to provide diagnostic support for Sickle Cell Disease using peripheral blood smear images of red blood cells. We pre-processed and segmented the microscopic images to ensure the extraction of high-quality features. To ensure the reliability of our proposed system, we conducted an in-depth analysis of interpretability. Leveraging techniques established in the literature, we extracted features from blood cells and employed ensemble machine learning methods to classify their morphology. Furthermore, we have devised a methodology to identify the most critical features for classification, aimed at reducing complexity and training time and enhancing interpretability in opaque models. Lastly, we validated our results using a new dataset, where our model overperformed state-of-the-art models in terms of generalization. The results of classifier ensembled of Random Forest and Extra Trees classifier achieved an harmonic mean of precision and recall (F1-score) of 90.71% and a Sickle Cell Disease diagnosis support score (SDS-score) of 93.33%. These results demonstrate notable enhancement from previous ones with Gradient Boosting classifier (F1-score 87.32% and SDS-score 89.51%). To foster scientific progress, we have made available the parameters for each model, the implemented code library, and the confusion matrices with the raw data.}, archive = {J_EAAI}, author = {Nataša Petrović and Gabriel Moyà-Alcover and Antoni Jaume-i-Capó and Jose Maria Buades Rubio}, doi = {10.1016/j.engappai.2024.109875}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109875}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing generalization in sickle cell disease diagnosis through ensemble methods and feature importance analysis}, volume = {142}, year = {2025}, }
• (2025). Deep spiking neural networks based on model fusion technology for remote sensing image classification. EAAI, 142, 109873. (www)

  @article{ , abstract = {The spiking neural network (SNN) based on brain inspiration, as the third-generation neural network, has attracted great research interest due to its ultra-low power event-driven data processing method. How to obtain high-accuracy deep networks has always been a challenge in the field of SNN. At present, there are two main methods for training deep spiking neural networks (SNNs). The first is the spike-based spatiotemporal backpropagation (STBP), and the other indirect is to convert the trained artificial neural network (ANN) into SNN (ANN-SNN). Algorithms that directly train SNNs are usually inefficient, while the ANN-SNN-based models require a long inference time and also suffer from performance losses. The model fusion technology (MFT) proposed in this paper, combined with ANN-SNN and STBP, provides a new training paradigm for obtaining deep SNNs. We propose a bilateral multi-strength integrate-and-fire (BM-IF) spiking neuron for ANN-SNN to simplify the conversion operation. Under the same network architecture and encoding time window, the ResNet34 based on our algorithm achieved State-of-the-art (SOTA) in ImageNet. At the same time, we combined transfer learning methods to achieve SOTA results on three remote sensing scene classification datasets. Our results also indicate that the MFT-based SNN outperforms the optimal inference accuracy provided by ANN-SNN in terms of network architecture and dataset inference time steps reduced by 512-1250 times. It can be seen that this algorithm provides a low-latency and high-precision training scheme for the development of deep SNN.}, archive = {J_EAAI}, author = {Li-Ye Niu and Ying Wei and Liping Zhao and Keli Hu}, doi = {10.1016/j.engappai.2024.109873}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109873}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep spiking neural networks based on model fusion technology for remote sensing image classification}, volume = {142}, year = {2025}, }
• (2025). Linguistic interval-valued polytopic fuzzy set and cloud model based technique for order of preference by similarity to ideal solution. EAAI, 142, 109872. (www)

  @article{ , abstract = {Polytopic fuzzy sets are useful in multi-criteria group decision-making (MCGDM) to capture uncertainty when it occurs in such a way that the sum of squares of positive-membership, neutral-membership and negative-memberships exceeds by 1. This research paper introduces interval-valued polytopic fuzzy set and integrates it with linguistic variable to define linguistic interval-valued polytopic fuzzy set (LIVPOFS) for capturing the uncertainty more efficiently using qualitative assessment. We define basic operations, score function, and accuracy function for LIVPOFS. Using prominent characteristic of cloud model to transform qualitative insights into quantitative data, we propose a LIVPOFS and a trapezium cloud-based technique for order of preference by similarity to ideal solution (TOPSIS) method for MCGDM problems. Proposed MCGDM developed in this study allows user feedback to be translated into quantitative data more efficiently, enhancing decision-making accuracy in qualitative MCGDM. Recognizing the essential role of taxi services in transportation networks, we have applied the proposed MCGDM method to rank taxi services of Ola, Uber, Mega, and Meru in the Delhi region of India. The qualitative phase in implementation of proposed method begins with a structured questionnaire designed to collect user opinions on different taxi services. To streamline computations, users are divided into groups, and their responses are collected via Google Forms. The effectiveness of the proposed LIVPOFS-based trapezium cloud TOPSIS in ranking taxi services is further validated through comparisons with existing MCGDM methods.}, archive = {J_EAAI}, author = {Laxmi Rajput and Sanjay Kumar}, doi = {10.1016/j.engappai.2024.109872}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109872}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Linguistic interval-valued polytopic fuzzy set and cloud model based technique for order of preference by similarity to ideal solution}, volume = {142}, year = {2025}, }
• (2025). Exponential heterogeneous anti-synchronization of multi-variable discrete stochastic inertial neural networks with adaptive corrective parameter. EAAI, 142, 109871. (www)

  @article{ , abstract = {This article departs from the conventional considerations of homogeneous and continuous structures to propose a master–slave heterogeneous frame of time-space discrete inertial neural networks with feedback control at the boundary. The aim of this paper is to consider mean squared exponential anti-synchronization of a discrete-time and -space stochastic heterogeneous inertial neural networks with adaptive corrective parameter. By using the approaches of space–time discrete Lyapunov-Krasovskii functional and linear matrix inequality, some decision theorems for the mean squared exponential anti-synchronization of the above discrete heterogeneous networks are come true based on the designation of a updated law for adaptive parameter in the slave networks and feed-back controller in the boundary. By right of adaptive updated law, the master–slave heterogeneous networks will approximately tend to be the homogeneous networks over the continuous adjustments of the corrective parameter. The findings of this article demonstrate the efficacy of the proposed approach in addressing the challenge of anti-synchronization of time-space discrete heterogeneous networks in the domains of science and engineering. Finally, a numerical example is given to clarify the feasibility of the current work.}, archive = {J_EAAI}, author = {Tianwei Zhang and Yongyan Yang and Sufang Han}, doi = {10.1016/j.engappai.2024.109871}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109871}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exponential heterogeneous anti-synchronization of multi-variable discrete stochastic inertial neural networks with adaptive corrective parameter}, volume = {142}, year = {2025}, }
• (2025). Path planning via reinforcement learning with closed-loop motion control and field tests. EAAI, 142, 109870. (www)

  @article{ , abstract = {Performing evasive maneuvers with highly automated vehicles is a challenging task. The algorithm must fulfill safety constraints and complete the task while keeping the car in a controllable state. Furthermore, considering all aspects of vehicle dynamics, the path generation problem is numerically complex. Hence its classical solutions can hardly meet real-time requirements. On the other hand, single reinforcement learning based approaches only could handle this problem as a simple driving task and would not provide feasibility information on the whole task’s horizon. Therefore, this paper presents a hierarchical method for obstacle avoidance of an automated vehicle to overcome this issue, where the geometric path generation is provided by a single-step continuous Reinforcement Learning agent, while a model-predictive controller deals with lateral control to perform a double lane change maneuver. As the agent plays the optimization role in this architecture, it is trained in various scenarios to provide the necessary parameters for a geometric path generator in a one-step neural network output. During the training, the controller that follows the track evaluates the feasibility of the generated path whose performance metrics provide feedback to the agent so it can further improve its performance. The framework can train an agent for a given problem with various parameters. As a use case, it is presented as a static obstacle avoidance maneuver. the proposed framework was tested on an automotive proving ground with the geometric constraints of the ISO-3888-2 test. The results proved its real-time capability and performance compared to human drivers’ abilities.}, archive = {J_EAAI}, author = {Árpád Fehér and Ádám Domina and Ádám Bárdos and Szilárd Aradi and Tamás Bécsi}, doi = {10.1016/j.engappai.2024.109870}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109870}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Path planning via reinforcement learning with closed-loop motion control and field tests}, volume = {142}, year = {2025}, }
• (2025). Multi-stages attention breast cancer classification based on nonlinear spiking neural p neurons with autapses. EAAI, 142, 109869. (www)

  @article{ , abstract = {Breast cancer (BC) is a prevalent type of malignant tumor in women. Early diagnosis and treatment are vital for enhancing the patients’ survival rate. A challenge is that downsampling may lead to loss of information in deep networks. So we propose a multi-stages attention architecture based on Nonlinear spiking neural P (NSNP) systems neurons with autapses (NSNP-AU) to compensate for the detail and edge information and allow convolutional neural networks to pay more attention to seek the lesion region. Unlike the single-scale attention acquisition methods of existing methods, we set up spatial attention acquisition at each feature map scale of the convolutional network to obtain a fusion global of information on attention guidance. Besides, we introduce a new type of NSNP variant called NSNP-AU neurons. Specifically, NSNP systems are modularized as a feature dimensionality reduction tool, recoding the features extracted from convolutional neural networks as well as the fusion of attention information and preserving the key characteristic elements in feature maps. This ensures the retention of valuable data while gradually transforming high-dimensional complicated info into low-dimensional ones. The proposed method is evaluated on the public dataset BreakHis at various magnifications and classification tasks. It achieves a classification accuracy of 96.32% at all magnification cases, outperforming state-of-the-art methods. Ablation studies are also performed, verifying the proposed model’s efficacy.}, archive = {J_EAAI}, author = {Yaorui Tang and Bo Yang and Hong Peng and Xiaohui Luo}, doi = {10.1016/j.engappai.2024.109869}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109869}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-stages attention breast cancer classification based on nonlinear spiking neural p neurons with autapses}, volume = {142}, year = {2025}, }
• (2025). Data-driven optimization for enhanced excavation efficiency in tunnel construction: A case study. EAAI, 142, 109868. (www)

  @article{ , abstract = {This research presents a hybrid approach to conduct the prediction and optimization of tunnel boring machine (TBM) performance, such as cutter wear and cutter power, where the improved gradient boosting machine (LightGBM) with Bayesian optimization is developed to perform performance prediction of TBM based on different TBM operating parameters with geological conditions. Subsequently, the nondominated sorting genetic algorithm II (NSGA-II) is adopted to solve the multi-objective optimization (MOO) problem for searching the Pareto fronts, followed by applying the order preference by similarity to an ideal solution (TOPSIS) to obtain the best solution. To verify the feasibility and effectiveness of the hybrid approach, a Singapore Mass Rapid Transit (MRT) project is employed as a study case. The results indicate that the improved LightGBM is in good agreement with the actual measurements in predicting cutter power with a mean absolute error (MAE) of 1.2584E-04, a root mean squared error (RMSE) of 1.7237E-04, and R-squared ( R 2 ) of 0.9084 for the cutter wear, while MAE of 31.0081, RMSE of 36.0126, and R 2 of 0.8044 for the cutter power. The developed approach can implement MOO to improve the performance of TBM, where the total cutter wear of the TBM is significantly reduced by 16.80% and cutter power consumption is greatly reduced by 14.91% under the 20% constraint condition. All those results demonstrate that the proposed approach is capable of generating optimal solutions, which can improve the performance of TBM and guide the tunneling operations for facilitating tunnel engineering project construction.}, archive = {J_EAAI}, author = {Jinting Huang and Ankang Ji and Limao Zhang and Xianlei Fu and Xieqing Song}, doi = {10.1016/j.engappai.2024.109868}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109868}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven optimization for enhanced excavation efficiency in tunnel construction: A case study}, volume = {142}, year = {2025}, }
• (2025). Medical image segmentation method based on full perceived dynamic network. EAAI, 142, 109867. (www)

  @article{ , abstract = {In clinical practice, the information access about tissues and lesion areas from the medical images is an important basis for disease diagnosis, treatment planning, and tracking detection. Accurate and automatic medical image segmentation is highly effective in improving diagnostic efficiency and quantitatively evaluating the treatment process. However, due to the diversity of biological tissues and the uncontrollable imaging situation, it is difficult to achieve stable and high-precision segmentation in medical images by the existing methods. In this work, a new medical image segmentation model, full perceived dynamic network (FPD-Net), is proposed to address this issue. First, a more efficient convolution method is designed to efficiently extract feature information by performing comprehensive multidimensional differential convolution operations on the input. Then, a new Transformer block called composite sensing Transformer (CST) is designed, which enhances the perceptual extraction of image detail features and multi-scale information through local detail enhancement and multi-scale encoding. In addition, a deep interaction encoding structure is built that efficiently integrates the advantages of convolutional encoding and autocorrelation encoding, enabling the model to fully perceive and utilize global and local information in the image. Finally, extensive experiments are conducted to validate our method in both tasks: retinal vessel segmentation and liver tumor segmentation, and make comparison with several state-of-the art methods. The results indicate that our FPD-Net outperforms other advanced methods in comprehensive segmentation performance. This shows the great potential of our method in medical assisted diagnosis.}, archive = {J_EAAI}, author = {Wentao Tang and Hongmin Deng and Zhengwei Huang and Yuanjian Jiang}, doi = {10.1016/j.engappai.2024.109867}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109867}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Medical image segmentation method based on full perceived dynamic network}, volume = {142}, year = {2025}, }
• (2025). Low complexity adaptive neural network three-dimensional tracking control for autonomous underwater vehicles considering uncertain dynamics. EAAI, 142, 109860. (www)

  @article{ , abstract = {This paper proposes a low complexity adaptive neural network control scheme for three-dimensional tracking control of autonomous underwater vehicles (AUVs) in the presence of uncertain dynamics and external disturbances. The virtual stabilizing functions are proposed by combining error nonlinear transformation and backstepping control strategy, which ensures that the transient tracking errors are confined in the prescribed range. A low complexity error-driven adaptive control strategy that only employs one neural network is presented to eliminate the uncertain dynamics in different free degrees, thereby effectively reducing the computational complexity of the AUV controller. Theoretical analyses indicate that the tracking error signals of the AUV system are bounded. The comparative simulation studies are conducted to substantiate the effectiveness of our proposed control scheme.}, archive = {J_EAAI}, author = {Wenfeng Xu and Jiapeng Liu and Jinpeng Yu and Yaning Han}, doi = {10.1016/j.engappai.2024.109860}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109860}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Low complexity adaptive neural network three-dimensional tracking control for autonomous underwater vehicles considering uncertain dynamics}, volume = {142}, year = {2025}, }
• (2025). Lightweight deep learning models for aerial scene classification: A comprehensive survey. EAAI, 142, 109859. (www)

  @article{ , abstract = {With the rapid growth of aerial image quantity and quality, the performance of aerial scene classifiers based on deep learning models has also achieved tremendous success. However, this has come at a price of increased computational burden as well as high time consumption. In this context, lightweight deep learning models have emerged recently. This survey presents a comprehensive review of current advances in lightweight deep learning models developed for the task of aerial scene classification. We categorize these models based on their techniques for making the model lighter. The advantages and limitations of each category are discussed here and an in-depth analysis of their performance on benchmark datasets is presented. Furthermore, we identify the current research challenges and opportunities for future work in this field.}, archive = {J_EAAI}, author = {Suparna Dutta and Monidipa Das and Ujjwal Maulik}, doi = {10.1016/j.engappai.2024.109859}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109859}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight deep learning models for aerial scene classification: A comprehensive survey}, volume = {142}, year = {2025}, }
• (2025). Machine learning approaches for predicting and evaluating pause rate at expressway rest areas. EAAI, 142, 109857. (www)

  @article{ , abstract = {An accurate pause rate prediction is crucial for calculating the appropriate size of expressway rest area This study introduces an eXtreme Gradient Boosting (XGBoost) pause rate prediction model that utilizes the tree-structured Parzen estimator for Bayesian optimization of its parameters. The model incorporates 13 input variables selected by considering traffic flow, rest area location and functionality, regional development, and geographical environmental features. These variables were used to predict the hourly pause rate, thereby refining the temporal granularity of the prediction. Compared with the XGBoost model with default parameters, genetic algorithm-based back propagation neural network, and genetic algorithm-based wavelet neural network, the proposed model is more accurate, validated by the root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination values. The confidence intervals of the proposed model were estimated using the bootstrap method. The Shapley Additive explanations model (SHAP) was employed to rank and interpret feature importance, thereby enhancing the interpretability of the model. The traffic volume and the relative location of the rest areas were found to be the most important factors. The proposed methodology offers novel insights and approaches for designing rest area during the planning and construction phases of expressways.}, archive = {J_EAAI}, author = {YouLei Fu and Jiangbi Hu and Lina Gao and Kang Du and Erli Gao and Aolin Yu}, doi = {10.1016/j.engappai.2024.109857}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109857}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning approaches for predicting and evaluating pause rate at expressway rest areas}, volume = {142}, year = {2025}, }
• (2025). Health state assessment based on the Parallel–Serial belief rule base for industrial robot systems. EAAI, 142, 109856. (www)

  @article{ , abstract = {To ensure industrial robots function correctly and implement adequate safety and maintenance strategies, it is crucial to accurately and promptly assess their health. Belief Rule Base (BRB) expert systems excel in modeling complex systems. However, BRB can lead to a “combinatorial explosion” of rules during optimization, reducing interpretability and increasing model complexity. Consequently, this paper introduces a Parallel–Serial Belief Rule Base (PS-BRB) model for assessing health states. Through structural and parameter optimization, the model’s interpretability is enhanced, and its complexity is minimized. To quantify the interpretability of the BRB model, we propose two evaluation criteria: rule simplicity and conflict. To further the modeling performance, we introduce a feature selection strategy and a feature grouping strategy of PS-BRB to construct a BRB model with enhanced interpretability. Finally, a versatile industrial robot system operating in a company’s body-in-white welding production line will be used as a case study to demonstrate the method’s effectiveness.}, archive = {J_EAAI}, author = {Xiaojing Yin and Weidong He and Jidong Wang and Shouxin Peng and You Cao and Bangcheng Zhang}, doi = {10.1016/j.engappai.2024.109856}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109856}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Health state assessment based on the Parallel–Serial belief rule base for industrial robot systems}, volume = {142}, year = {2025}, }
• (2025). Beyond boundaries: Advancements in fire and smoke detection for indoor and outdoor surveillance feeds. EAAI, 142, 109855. (www)

  @article{ , abstract = {This survey paper reviews the challenges and recent advancements in Artificial Intelligence (AI) video-based smoke and fire detection systems, with particular focus on both indoor and outdoor environments. The main problem addressed is the high false alarm rates (ranging from 3.4% to 29.49% across various systems) and the challenges posed by environmental variability, dataset scarcity, and the complexity of real-time detection. The paper critically examines key methodologies, including traditional approaches, deep learning techniques (with accuracy rates reaching up to 98.72% and false alarm rates reduced to as low as 0.61%), hybrid methods, and domain transfer-based tools, highlighting their evolution and current trends. This survey also provides an in-depth analysis of publicly available datasets and evaluation metrics, such as detection accuracy (ranging from 79.66% to 98.72%), robustness to dynamic environments, and real-time processing capabilities (with some systems achieving up to 333 frames per second (FPS). By synthesizing insights from 33 papers published between 2013 and 2024, the survey not only summarizes the current state of the art but also identifies emerging trends, such as the increasing use of automatic feature learning and multi-fusion systems, which have demonstrated significant improvements in detection accuracy. The paper concludes by advocating for future research focused on improving system robustness and reducing false alarms through the integration of visible range cameras and traditional sensors, with the goal of achieving more accurate and reliable fire detection in surveillance systems.}, archive = {J_EAAI}, author = {Rafaqat Alam Khan and Usama Ijaz Bajwa and Rana Hammad Raza and Muhammad Waqas Anwar}, doi = {10.1016/j.engappai.2024.109855}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109855}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Beyond boundaries: Advancements in fire and smoke detection for indoor and outdoor surveillance feeds}, volume = {142}, year = {2025}, }
• (2025). Fast multiplicative fuzzy partition C-means clustering with a new membership scaling scheme. EAAI, 142, 109854. (www)

  @article{ , abstract = {The generalized multiplicative fuzzy C-means (GMFCM) clustering is a novel and potentially competitive fuzzy clustering method. However, the convergence speed of this method is relatively slow. Therefore, a fast generalized multiplicative fuzzy C-means clustering algorithm is proposed in this paper. First, an affinity filtering method based on triangular inequalities is utilized to identify a complete set of non-affinity cluster centers for each sample with low computational complexity. Then, a new multiplicative membership scaling scheme is proposed to modify the multiplicative fuzzy membership between each sample and its non-affinity cluster centers to a non-zero minimum, while increasing the multiplicative fuzzy membership between each sample and its affinity cluster centers. Finally, a new fast algorithm combining these two techniques is proposed to accelerate the convergence process of GMFCM. Extensive experiments on numerical data and images confirm that the proposed algorithm is faster and significantly reduces the number of GMFCM iterations. Specifically, the algorithm reduces the number of iterations by more than 60% compared with the original GMFCM algorithm. At the same time, the running time of the algorithmic process is also greatly reduced.}, archive = {J_EAAI}, author = {Chengmao Wu and Yulong Gao}, doi = {10.1016/j.engappai.2024.109854}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109854}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fast multiplicative fuzzy partition C-means clustering with a new membership scaling scheme}, volume = {142}, year = {2025}, }
• (2025). Novel inverse predictive system integrated with industrial lubricant information. EAAI, 142, 109853. (www)

  @article{ , abstract = {The development of lubricants with the required specifications is a time-consuming and costly process, given the need to explore numerous formulations that represent material types and their combinations for the lubricant. Although machine learning approaches have made significant progress in addressing these challenges, their practical application in industrial plants requires considerable expert knowledge to screen out unfeasible lubricant formulations. Hence, this paper proposes a novel inverse predictive system integrated with industrial information to recommend new and feasible formulations for the lubricant industry. The proposed system integrates a machine-learning-based model for property prediction with an optimization process that incorporates expert knowledge. The developed prediction models demonstrated high performance in predicting the lubricant properties (viscosity at 40 °C, viscosity at 100 °C, and density), with R 2 scores of 0.9839, 0.9779, and 0.9816, respectively. In addition, an optimization process using particle swarm optimization was employed to suggest formulations tailored to the specific requirements of various industries. The recommended formulations were validated using laboratory-scale specimens with errors of 5%–20%. This framework provides promising opportunities for recommending material types and their ratios in various industries.}, archive = {J_EAAI}, author = {Minseong Kim and Chonghyo Joo and Jongkoo Lim and Seungho Yeom and Il Moon and Meng Qi and Junghwan Kim}, doi = {10.1016/j.engappai.2024.109853}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109853}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Novel inverse predictive system integrated with industrial lubricant information}, volume = {142}, year = {2025}, }
• (2025). Frequency domain nuances guided parallel transformer model for industrial anomaly localization. EAAI, 142, 109852. (www)

  @article{ , abstract = {Unsupervised visual anomaly localization research has garnered significant attention in industrial component surface quality inspection tasks, particularly in realistic scenarios characterized by extreme imbalance between positive and negative samples. While several existing approaches endeavor to design hybrid architectures of convolutional neural networks (CNNs) and vision transformers (ViTs) to enhance the performance capability of unsupervised models, their overall effectiveness remains unstable. In this paper, we introduce a novel approach known as frequency domain nuances mining (FDNM), presenting a parallel transformer framework for anomaly localization. This framework leverages the inductive bias property of CNNs for local spatial understanding and the advantageous ViTs for global representation learning, enabling simultaneous capture of local correlations and global information. To further enhance the model’s discriminative ability for subtle differences, we propose a frequency domain decoupling mechanism that exploits the frequency domain properties of images to improve model interpretability. Specifically, multi-granularity differences among frequency domain components serve as prior embeddings in FDNM, augmenting the potential feature representation of local details and global semantics. Furthermore, we synergistically train frequency domain invariant sampling loss and focal loss to balance the differential representation of features in the cross-frequency domain, leading to a more stable training process and more accurate anomaly localization results. Experiments conducted on the challenging 15-class industrial anomaly detection dataset validate the superiority of the proposed method. The image-level area under the receiver operating characteristic curve (AUROC), pixel-level AUROC and average precision (AP) scores of FDNM reach 99.0%, 98.7% and 68.7%, respectively, demonstrating comparable performance to state-of-the-art methods.}, archive = {J_EAAI}, author = {Jun Zhao and Kaixuan Yu and Yu Miao and Yingsen Wang and Yue Ma and Jiawei Zhang and Juanjuan Zhao and Yan Qiang and Bo Pei}, doi = {10.1016/j.engappai.2024.109852}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109852}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Frequency domain nuances guided parallel transformer model for industrial anomaly localization}, volume = {142}, year = {2025}, }
• (2025). Modelling and scheduling distributed assembly permutation flow-shops using reinforcement learning-based evolutionary algorithms. EAAI, 142, 109851. (www)

  @article{ , abstract = {Distributed assembly permutation flowshop scheduling problem (DAPFSP) have aroused extensive interests in both academia and industry due to its wide range of applications in manufacturing systems. However, the existing literature on this topic is comparatively insufficient. This study investigated the considered problem with total flowtime (TF) criterion by developing three evolutionary algorithms (EAs) and their reinforcement learning (RL) based variants. First, two heuristic rules are proposed to initialize population with diversity. Second, six elaborated local search operators are incorporate with the basic EAs considering the problem’s characteristics. Then, the RL algorithms e.g., Q-Learning and Sarsa, are employed to select high-quality local search operators during algorithms' iterations to improve their performance. Three problem characteristic based RL strategies are designed for mapping the RL algorithms and local search operators. Finally, detailed calibration according to 81 large-scale instances is performed to illustrate the effectiveness of the proposed algorithms for solving the DAPFSP.}, archive = {J_EAAI}, author = {Bohan Qiu and Kaizhou Gao and Hui Yu and Ali Sadollah}, doi = {10.1016/j.engappai.2024.109851}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109851}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Modelling and scheduling distributed assembly permutation flow-shops using reinforcement learning-based evolutionary algorithms}, volume = {142}, year = {2025}, }
• (2025). Galaxy image super-resolution reconstruction using diffusion network. EAAI, 142, 109836. (www)

  @article{ , abstract = {The spatial resolution of galaxy images is crucial for galaxy research but is limited by the observation limitations of telescopes. To get a higher resolution, this paper proposes a Galaxy image super-resolution reconstruction method using Diffusion Network (GD-Net), which introduces a global attention mechanism into the Denoising Diffusion Probabilistic Model (DDPM) and improves the feature affine transformation module in the residual structure, effectively improving the reconstruction quality of galaxy images. GD-Net learns the inverse process of adding noises to high-resolution (HR) images, constructing required data samples from the noises. And low-resolution (LR) images serve as conditional guidance for GD-Net to reconstruct super-resolution (SR) images from the noises. To verify the effectiveness of GD-Net, images from the Dark Energy Spectroscopic Instrument (DESI) and the Sloan Digital Sky Survey (SDSS) are used to train the network, and the spatial resolution of the reconstructed images from SDSS is twice that of the original images. The results show that the images reconstructed using the proposed method have high consistency with the HR data provided by DESI. In terms of the Learned Perceptual Image Patch Similarity (LPIPS) metric, GD-Net achieves the least distortion of generated super-resolution galaxy images with an LPIPS value of 0.05 and has superior performance compared to EDSR(Super-Resolution Generative Adversarial Networks), SRGAN(Enhanced Deep Super-Resolution Network), HAT(Activating More Pixels in Image Super-Resolution Transformer) and LDM(Latent Diffusion Model).}, archive = {J_EAAI}, author = {Qian-Qian Shan and Cui-Xiang Liu and Bo Qiu and A-Li Luo and Fu-ji Ren and Zhi-Ren Pan and Yi-Cong Chen}, doi = {10.1016/j.engappai.2024.109836}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109836}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Galaxy image super-resolution reconstruction using diffusion network}, volume = {142}, year = {2025}, }
• (2025). Multi-level domain adaptation for improved generalization in electroencephalogram-based driver fatigue detection. EAAI, 142, 109825. (www)

  @article{ , abstract = {Developing a universal model capable of effective cross-subject fatigue detection is crucial for alerting fatigued drivers in a timely manner and reducing traffic accidents. Although existing domain adaptation methods can reduce distribution differences between subjects, they primarily focus on inter-subject feature (inter-domain) alignment while neglecting category-level feature (intra-domain) alignment. To address this issue, this paper proposes a novel Multi-Level Domain Adaptation (MLDA) algorithm to enhance the model’s generalization ability on unseen data. Specifically, for inter-domain alignment, we employ an improved Wasserstein distance, whose smooth properties more accurately measure inter-domain differences. For intra-domain alignment, we introduce intra-domain contrastive discrepancy, which enhances the discriminability of category features by maximizing inter-class distances and minimizing intra-class distances. The proposed method achieves cross-subject fatigue detection accuracies of 0.942 and 0.843 on the SEED-VIG and SADT public datasets, respectively. Experimental results demonstrate that MLDA offers significant advantages in cross-subject fatigue detection tasks, providing a promising solution for a generalized driver fatigue detection system.}, archive = {J_EAAI}, author = {Fuzhong Huang and Qicong Wang and Lei Chen and Wang Mei and Zhenchang Zhang and Zelong Chen}, doi = {10.1016/j.engappai.2024.109825}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109825}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-level domain adaptation for improved generalization in electroencephalogram-based driver fatigue detection}, volume = {142}, year = {2025}, }
• (2025). Connected vehicle following control based on gated recurrent unit with attention mechanism. EAAI, 142, 109820. (www)

  @article{ , abstract = {A delicate balance between safety, efficiency, and fluidity needs to be carefully maintained in vehicle following, in strict accordance with real-time control imperatives. Achieving efficient vehicle-following operations under safe driving conditions, through smooth behavioral adjustments, presents a significant challenge for data-driven vehicle-following models. In response to this challenge, we have developed a deep neural network based on gated recurrent unit (GRU) with attention mechanism, named AGRUNet model, for artificial intelligence (AI) control of vehicle following behavior. Through training and testing on diverse datasets, the AGRUNet model not only establishes a nonlinear mapping relationship between the following vehicle’s acceleration and the speeds of the leading and following vehicles, their distance, and the control strategy of the leading vehicles but also accurately forecasts the future behaviors of following vehicles in complex vehicle-following scenarios in real-time. This capability enables the following vehicle to optimize its behavior based on the current vehicle-following situation and control requirements, thereby improving safety, efficiency, and smoothness. Rigorous simulations of AGRUNet on the Highway Drone(HighD), Next Generation Simulation(NGSIM), Waymo, and Lyft Level 5(Lyft) datasets demonstrate its superior performance in prediction accuracy and vehicle-following control. Compared to the widely adopted, high-performance Long Short-Term Memory (LSTM) model, AGRUNet achieves prediction accuracy gains of approximately 2%, 7%, 22%, and 3% across these datasets. Extensive testing further indicates that AGRUNet significantly reduces collision rates during sudden emergency braking by the leading vehicle, enhancing safety, and improving the efficiency and smoothness of behavior adjustments, all while ensuring vehicle-following safety.}, archive = {J_EAAI}, author = {Shengjie Wang and Deng Pan and Xianda Chen and Zexin Duan and Zehao Xu}, doi = {10.1016/j.engappai.2024.109820}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109820}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Connected vehicle following control based on gated recurrent unit with attention mechanism}, volume = {142}, year = {2025}, }
• (2025). A looped-type functional for non-fragile fuzzy sampled-data control of doubly fed induction generator-based wind energy conversion systems with failures. EAAI, 142, 109812. (www)

  @article{ , abstract = {In our investigation, by integrating the fuzzy modeling method and looped-type Lyapunov functional in tandem with free-matrix-based integral inequality technique, the nonfragile fault tolerant fuzzy sampled-data control (FTFSDC) problem for doubly fed induction generator (DFIG)-based wind energy conversion systems (WECSs) that are prone to semi-Markovian jump-type actuator failures (SMJAFs) has been examined. First and foremost, fuzzy modeling method, which well-suited for addressing high nonlinearity issues, is exploited to improve the imprecision of DFIG-based WECSs. Subsequently, to delineate the stochastic actuator failures (SAFs), the FTFSDC framework is reconstructed with semi-Markovian jump faulty coefficients undergoing partially unknown transition rates (PUTRs), which has stronger modeling precisely and exhibits the practical scenarios more accurately. More concretely, the PUTRs considered in this article including completely unknown elements and uncertain but bounded elements as two special situations. Thirdly, in a departure from the anterior sampled-data control (SDC) mechanisms, the memory signal τ is subtly embedded into the communication channel, and the sampling patterns in conjunction with the control actions are performed in a sawtooth-characteristic manner. Following that, by adequately capturing the characteristic information of the entire sampling intervals [ t k , t k + 1 ) and [ t k − τ , t k + 1 − τ ) simultaneously, an augmented dual-sided looped-type Lyapunov functional is put forward. The superiority of the newly constructed Lyapunov functional is that more sawtooth structure information of the actual sampling patterns is developed and some symmetric matrices involved are no longer restricted to positive definite, which can significantly enhance the design freedom and flexibility. Afterwards, with the assistance of mathematical deduction and stochastic analysis technique, several certain sufficient conditions such that the underlying plant is asymptotically stable are lastly precisely inferred in the shape of linear matrix inequalities (LMIs). In conclusion, a corresponding numerical simulation is conducted to thoroughly substantiate the efficacy and feasibility of the theoretical outcomes.}, archive = {J_EAAI}, author = {Xiaoqing Li and Kaibo Shi and Liang Han and Jun Cheng and Wei Sun and Zhinan Peng}, doi = {10.1016/j.engappai.2024.109812}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109812}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A looped-type functional for non-fragile fuzzy sampled-data control of doubly fed induction generator-based wind energy conversion systems with failures}, volume = {142}, year = {2025}, }
• (2025). Advancing fault diagnosis in industrial systems: The power of V-nets for managing complex event sequences. EAAI, 142, 109781. (www)

  @article{ , abstract = {This paper introduces V-nets, an innovative formal model developed to enhance fault diagnosis and management of complex Discrete Event Systems (DES), particularly in industrial applications. The V-net model addresses key challenges in traditional models such as Petri nets and automata, including the management of simultaneous events, minimizing false positives, and improving the recognition of partial event sequences. Methods: This study builds upon a comparative analysis of existing fault diagnosis techniques, integrating V-nets with automata to form a robust framework for supervising and diagnosing industrial systems. Through a case study of the Reactor Startup Sequence of a Pressurized Water Reactor (PWR), the research highlights the practical utility of V-nets in real-time fault detection and system monitoring. Findings: The experimental results demonstrate that V-nets significantly reduce false positives and enhance system flexibility when dealing with complex event sequences. By evaluating over 100,000 event sequences using Monte Carlo methods, the study confirms the accuracy and scalability of the V-net model in industrial settings. Novelty: This research pioneers the development of V-nets as a superior alternative to traditional models for fault diagnosis in DES, offering a framework capable of handling simultaneous events and complex time constraints. The proposed model’s potential for real-world implementation opens new avenues for advancing industrial process supervision. The flexibility and scalability of V-nets represent a critical advancement in the formal methods for industrial systems, positioning this model as a promising tool for future industrial applications and automatic V-net generation in real-time scenarios.}, archive = {J_EAAI}, author = {John William Vásquez-Capacho}, doi = {10.1016/j.engappai.2024.109781}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109781}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advancing fault diagnosis in industrial systems: The power of V-nets for managing complex event sequences}, volume = {142}, year = {2025}, }
• (2025). Inferring effort-safety trade off in perturbed squat-to-stand task by reward parameter estimation. EAAI, 142, 109778. (www)

  @article{ , abstract = {In this study, an inverse reinforcement learning (IRL) method is developed to estimate the parameters of a reward function that is assumed to guide the movement of a biological or artificial agent. The workings of the method is shown on the problem of estimating the effort-safety trade-off of humans during perturbed squat-to-stand motions based on their Center of Mass (COM) trajectories. The proposed method involves data generation by reinforcement learning (RL) and a novel data augmentation mechanism followed by neural network training. After the training, the neural network acts as the reward parameter estimator given the Center of Mass (COM) trajectories as input. The performance of the developed method is assessed through systematic simulation experiments, where it is shown that the parameter estimation made by our method is significantly more accurate than the baseline of an optimized template-based IRL approach. In addition, as a proof of concept, a set of human movement data is analyzed with the developed method. The results revealed that most participants acquired a strategy that ensures low effort expenditure with a safety margin, producing COM trajectories slightly away from the effort-optimal.}, archive = {J_EAAI}, author = {Emir Arditi and Tjasa Kunavar and Negin Amirshirzad and Emre Ugur and Jan Babič and Erhan Oztop}, doi = {10.1016/j.engappai.2024.109778}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109778}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Inferring effort-safety trade off in perturbed squat-to-stand task by reward parameter estimation}, volume = {142}, year = {2025}, }
• (2025). Modeling energy efficiency in industrial plants: A novel diagnostic approach. EAAI, 142, 109777. (www)

  @article{ , abstract = {In industrial plants, diagnosing energy efficiency issues is essential to achieve sustainable operations and reduce costs. This paper introduces a novel diagnostic approach using advanced modeling techniques to identify inefficiencies in energy consumption within industrial environments. The proposed method uses discrete event analysis to detect and characterize abnormal energy usage patterns, providing a systematic framework for diagnosing performance issues in complex systems. Two case studies involving high-performance computing (HPC) systems illustrate the practical application of the approach, showcasing its ability to uncover critical inefficiencies and inform energy management strategies. The research addresses a significant gap in current methodologies by providing a detailed diagnostic tool customized to the unique challenges of industrial energy management. This study paves the way for future research into advanced diagnostic techniques, strengthening the importance of precise and actionable information on energy use for industrial stakeholders.}, archive = {J_EAAI}, author = {John William Vásquez Capacho and Carlos Gustavo Perez-Zuñiga and Adalberto Ospino-Castro}, doi = {10.1016/j.engappai.2024.109777}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109777}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Modeling energy efficiency in industrial plants: A novel diagnostic approach}, volume = {142}, year = {2025}, }
• (2025). A propagation-based ranking semantics in explainable bipolar weighted argumentation. EAAI, 142, 109767. (www)

  @article{ , abstract = {The Semantic Web of Things enables exchanging knowledge-based fragments in pervasive computing through machine-to-machine interactions, sustaining collaborative decision-making in networks of smart objects. Abstract argumentation is widely acknowledged, instead, as a general framework for decision-making in multi-agent systems, able to solve negotiation conflicts and evaluate acceptable options for a given scenario. Nevertheless, abstract argumentation frameworks disregard the problem of characterizing the argument structure and their mutual relations. This paper proposes a novel approach integrating Dung-style abstract argumentation with semantic matchmaking in pervasive computing. Object interactions are seen as an argumentative dialogue, where annotations shared by each agent play the role of arguments. A matchmaking scheme, exploiting non-standard non-monotonic inferences, allows the appraisal of argument relations in a bipolar weighted argumentation approach. A propagation-based gradual semantics provides acceptability ranking of arguments with a formal explanation. A fading mechanism is also exploited to take into account limited computational resources. As the proposed approach enables applications in cyber-physical multi-agent systems engineering, early validation and performance experiments are provided by means of a case study on a real-time strategy game.}, archive = {J_EAAI}, author = {Corrado Fasciano and Giuseppe Loseto and Agnese Pinto and Michele Ruta and Floriano Scioscia}, doi = {10.1016/j.engappai.2024.109767}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109767}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A propagation-based ranking semantics in explainable bipolar weighted argumentation}, volume = {142}, year = {2025}, }
• (2025). A topic joint model for knowledge extraction from unstructured maintenance records. EAAI, 142, 109743. (www)

  @article{ , abstract = {Extracting knowledge and constructing domain knowledge graphs from used vehicle maintenance records can reveal the mechanism of faults. However, maintenance records are semi-structured documents containing both structured and unstructured data. Here, we propose a topic joint model to improve the accuracy of knowledge extraction from used vehicle maintenance records. First, we apply the weighted latent Dirichlet allocation method to extract the hidden topic distribution of fault cases. Then, a bidirectional encoder representations from transformers (BERT) -based method is developed to identify the position and category information of fault entities with case-topic features. Finally, we used real data of vehicle maintenance records to verify the efficacy of our proposed method. The results show that when using real-world data, the proposed joint model has a higher accuracy than the standard knowledge extraction method, which is achieved by combining topic features for entity recognition and relationship extraction tasks. The F1 scores of entity recognition and relationship classification were 75.46 and 78.42, respectively.}, archive = {J_EAAI}, author = {Zheyuan Hu and Xu Zhang and Hui Xiong}, doi = {10.1016/j.engappai.2024.109743}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109743}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A topic joint model for knowledge extraction from unstructured maintenance records}, volume = {142}, year = {2025}, }
• (2025). A scalability analysis of a multi-agent framework for solving combinatorial optimization via metaheuristics. EAAI, 142, 109738. (www)

  @article{ , abstract = {This paper addresses the scalability dimension of a multi-agent framework for solving combinatorial optimization problems using metaheuristics. The related literature discusses several properties of multi-agent systems. However, an important property that has received little attention is scalability, which refers to the ability of a system to perform useful work uniformly as the dimension of the system itself increases. This article introduces a methodology for assessing the scalability of multi-agent frameworks using scenarios with one, two, four, eight, ten, fifty, thirty, forty, fifty, and sixty agents. The framework Multi-Agent Architecture for Metaheuristics (AMAM) is the adopted architecture for performing the computational tests, using instances of the Vehicle Routing Problem with Time Windows (VRPTW) and the Unrelated Parallel Machine Scheduling Problem with Sequence-Dependent Setup Times (UPMSP-ST). The proposed methodology uses the obtained values concerning the objective function and the associated runtimes of the instanced problems to perform statistical measurements and tests that revealed an improvement in the framework’s performance concerning the quality of solutions despite the slight increase in runtime for the VRPTW case. The results of a linear regression approach concerning the objective function data showed an adequate representation of agent inclusion but, regarding the runtime values, did not demonstrate such precise adjustments. The final results suggest that the adopted framework is scalable and demonstrates a robust response to the increase in the number of agents in its structure.}, archive = {J_EAAI}, author = {Maria Amélia Lopes Silva and Jardell Fillipe da Silva and Sérgio Ricardo de Souza and Marcone Jamilson Freitas Souza}, doi = {10.1016/j.engappai.2024.109738}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109738}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A scalability analysis of a multi-agent framework for solving combinatorial optimization via metaheuristics}, volume = {142}, year = {2025}, }
• (2025). An innovative combination of deep Q-networks and context-free grammars for symbolic solutions to differential equations. EAAI, 142, 109733. (www)

  @article{ , abstract = {In this research paper, we propose a novel approach that combines deep Q-networks with context-free grammars to solve differential equations symbolically. Our method utilizes the production rules of a context-free grammar to construct valid mathematical expressions. To obtain a solution for a given differential equation, a suitable sequence of production rules should be selected correctly. In our proposed method, a deep Q-network algorithm finds an effective substitution of the production rules so that the obtained expression is a valid and accurate symbolic solution to an equation under consideration. To demonstrate the effectiveness of our innovative method, we evaluate it across three distinct classes of equations arising in engineering and real-world applications: nonlinear ordinary differential equations, nonlinear partial differential equations, and nonlinear integral equations. The experimental results confirm that our proposed approach accurately and efficiently solves a wide range of differential equations.}, archive = {J_EAAI}, author = {Hassan Dana Mazraeh and Kourosh Parand}, doi = {10.1016/j.engappai.2024.109733}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109733}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An innovative combination of deep Q-networks and context-free grammars for symbolic solutions to differential equations}, volume = {142}, year = {2025}, }
• (2025). Data-driven joint multiobjective prediction and optimization for tunnel-induced adjacent bridge pier displacement: A case study in china. EAAI, 142, 109616. (www)

  @article{ , abstract = {To reduce the impact of tunnel construction on adjacent bridge pile foundations and ensure safety during construction, a hybrid intelligent framework combining Bayesian optimization (BO), categorical boosting (CatBoost), and the nondominated sorting genetic algorithm-III (NSGA-III) is proposed in this paper. The nonlinear mapping function relationship between the nine input parameters and the bridge pier vertical and horizontal displacements is established via BO-CatBoost. The key optimization parameters are for interpretability analysis and determined via Shapley additive explanations (SHAP) method. NSGA-III is established with the goal of minimizing pier displacement. The applicability and validity of the proposed method are tested in a case study of the Wuhan Metro. The key findings of this study include the following. (1) The accuracy of the prediction model obtained by the BO-CatBoost algorithm on the basis of the training and simulation of the measured engineering data is high. On the bridge pier horizontal and vertical displacement test sets, the R 2 values are 0.823 and 0.826, the RMSE values are 0.452 and 0.539, and the MAEs are 0.293 and 0.360, respectively. (2) The optimization effect of the two objectives is significant, and the average percentage of improvement stands at 35.54%. When five shield construction parameters are adjusted simultaneously, the optimization effect of the two objectives is the best, and the average improvement percentage is 54.76%. (3) The optimization effect of the developed BO-CatBoost-NSGA-III intelligent algorithm is greater than that of single-objective optimization. Therefore, the intelligent optimization framework proposed in this paper can provide guidance for the optimal control of pier displacement in shield underpass construction engineering.}, archive = {J_EAAI}, author = {Hongyu Chen and Jun Liu and Qiping Geoffrey Shen and Tiejun Li and Yang Liu}, doi = {10.1016/j.engappai.2024.109616}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109616}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven joint multiobjective prediction and optimization for tunnel-induced adjacent bridge pier displacement: A case study in china}, volume = {142}, year = {2025}, }
• (2025). A preprocessing data-driven pipeline for estimating number of clusters. EAAI, 141, 109893. (www)

  @article{ , abstract = {Due to the abundance of information, Artificial Intelligence (AI) research and development is a major focus in academia and industry. As the data volume increases, organizations face challenges in defining, selecting, and prioritizing the most relevant and influential variables for research, growth, and development. In this work, a preprocessing and data-driven pipeline will be demonstrated and implemented in estimating the number of clusters for a given dataset. The pipeline includes two main procedures: 1) a threshold-based method for feature selection, and 2) an estimation of neighborhood size (i.e., the total number of clusters). The procedure uses a risk parameter representing the number of features the user is willing to risk lowering in the overall correlation. This study is significant in the proposed pipeline's ability to predict the number of clusters of a given dataset by controlling the feature correlation and threshold learning of clustering evaluation measures. Using a data-driven methodology, the pipeline estimates the number of desired clusters for a given dataset to understand data distribution before evaluating any clustering procedure. The results showed the pipeline method achieved an improvement of up to 157.805% in the silhouette coefficient. In addition, there were also consistent improvements in other clustering performance metrics.}, archive = {J_EAAI}, author = {Michal Koren and Or Peretz and Oded Koren}, doi = {10.1016/j.engappai.2024.109893}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109893}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A preprocessing data-driven pipeline for estimating number of clusters}, volume = {141}, year = {2025}, }
• (2025). Uncertainty quantification in sequential hybrid deep transfer learning for solar irradiation predictions. EAAI, 141, 109874. (www)

  @article{ , abstract = {Hybrid deep learning model with multi-frequency capabilities is presented for simulating solar irradiation. Utilizing hourly recorded solar irradiation and climate data, model employs Convolutional Neural Network (CNN) to capture spatial and Long Short-Term Memory (LSTM) network to capture temporal characteristics. Initially, local models are developed individually, trained on single datasets. Additionally, global models are trained collectively by fusing multiple datasets. Furthermore, the study delves into transfer learning. To evaluate the uncertainty of prediction, Prediction Intervals (PIs) are computed using the bootstrap method and its results are compared with those of Bayesian Neural Network (BNN). The findings indicate that hybrid CNN-LSTM model surpasses single models in point predictions up to 25% in test. Local models demonstrate efficacy when deployed within their respective regions but exhibit notable errors when data scarcity hampers training. Global models exhibit adaptability to individual locations at expense of training efforts. Pre-training models on comprehensive and diverse source dataset, followed by transfer to a target dataset, generally yields superior performance. Concerning the model's performance in terms of PIs, CNN-LSTM model demonstrates the highest efficacy with average Prediction Interval Coverage Probability (PICP) of 0.82 and average Normalized Mean Prediction Interval Width (NMPIW) of 0.043. Also, in strategies involving limited or no available data for training, CNN-based models exhibit superior performance with lower uncertainty.}, archive = {J_EAAI}, author = {Vahid Nourani and Nazanin Behfar and Martijn J. Booij and Anne Ng and Chunwei Zhang and Sepideh Mohammadisepasi}, doi = {10.1016/j.engappai.2024.109874}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109874}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Uncertainty quantification in sequential hybrid deep transfer learning for solar irradiation predictions}, volume = {141}, year = {2025}, }
• (2025). Fuzzy a∗ quantum multi-stage Q-learning artificial potential field for path planning of mobile robots. EAAI, 141, 109866. (www)

  @article{ , abstract = {Aiming at problem of the efficiency of mobile robots heavily depends on the employed path-planning algorithms that encounter significant challenges from dynamic obstacles and complex environments. This work develops a fuzzy A∗ quantum multi-stage Q-learning artificial potential field approach, which combines fuzzy A∗, quantum multi-stage Q-learning, and artificial potential field (APF) algorithms. The fuzzy A∗ quantum multi-stage Q-learning APF algorithm combines a fuzzy system with an A∗ algorithm to improve the A∗ algorithm. In addition, the fuzzy A∗ quantum multi-stage Q-learning APF algorithm applies quantum computing methods and multi-stage training methods to improve the convergence speed of the Q-learning algorithm. If dynamic obstacles exist in environments and dynamic obstacles can block the global path, the obtained A∗ path points are a dopted as sub-goal points for the APF to plan paths. If mobile robots fall into trap areas, the quantum multi-stage Q-learning is then invoked to plan a path between the local minimum trap and the sub-goal point. Except for escaping the local minimum trap, the proposed fuzzy A∗ quantum multi-stage Q-learning APF has the advantage of consuming less time. The study tests the fuzzy A∗ quantum multi-stage Q-learning APF on environments with narrow passages and traps. The results verify that the proposed fuzzy A∗ quantum multi-stage Q-learning APF can navigate beyond complex environment situations and break free from local minimum traps.}, archive = {J_EAAI}, author = {Likun Hu and Chunyou Wei and Linfei Yin}, doi = {10.1016/j.engappai.2024.109866}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109866}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzy a∗ quantum multi-stage Q-learning artificial potential field for path planning of mobile robots}, volume = {141}, year = {2025}, }
• (2025). A modified adaptive large neighborhood search algorithm for serial-batching machines scheduling considering changeover time and rate-modifying activities. EAAI, 141, 109865. (www)

  @article{ , abstract = {Optimizing the scheduling of computer numerical control (CNC) machine tools is crucial for enhancing the efficiency of automotive component processing. This paper addresses the scheduling problem for CNC machine tools by simultaneously considering several critical factors for the first time in this research area: job changeover time, non-identical job sizes, deterioration effects, rate-modifying activities, and identical serial-batch processing machines. Using a mixed-integer programming model, our objective is to minimize the total completion time. We derive several structural properties that inform the design of two heuristic algorithms and establish lower bounds. These algorithms are further integrated with the First-fit Decreasing and Shortest Processing Time (FFD-SPT) and Best-fit Decreasing and Shortest Processing Time (BFD-SPT) algorithms to create a hybrid heuristic strategy. Given the NP-hard nature of the problem, we propose a modified adaptive large neighborhood search (MALNS) algorithm that leverages the hybrid heuristic strategy to achieve high-quality solutions within a reasonable timeframe. Extensive computational experiments on various problem instances demonstrate the effectiveness and robustness of the proposed MALNS algorithm compared to an exact solver and state-of-the-art metaheuristic algorithms. Collectively, our findings offer significant improvements in solving complex scheduling problems in automotive manufacturing, ensuring enhanced efficiency and productivity.}, archive = {J_EAAI}, author = {Shaojun Lu and Chiwei Hu and Min Kong and Amir M. Fathollahi-Fard and Binyun Wu}, doi = {10.1016/j.engappai.2024.109865}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109865}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A modified adaptive large neighborhood search algorithm for serial-batching machines scheduling considering changeover time and rate-modifying activities}, volume = {141}, year = {2025}, }
• (2025). Robust automatic modulation classification using asymmetric trilinear attention net with noisy activation function. EAAI, 141, 109861. (www)

  @article{ , abstract = {Nowadays, automatic modulation classification (AMC) plays an essential role in the cognitive radio based non-cooperative wireless communication system. Although numerous deep learning models have been developed for AMC, it remains challenging to effectively recognize various modulation schemes in complicated signal-to-noise ratio (SNR) conditions. In this paper, we introduce a robust AMC method based on the asymmetric trilinear attention net (Tri-Net) with noisy activation function. In Tri-Net, the asymmetric trilinear representation module is developed to deal with various channels of received radio signals and extract rich features to improve the generalization ability. Then a hybrid coding module with squeeze and excitation (SE) blocks-based attention mechanism is designed to help the model adapt to fluctuating SNRs. Finally, the predicted modulation schemes can be output through the classification module. During the training process, the noisy rectified linear unit (ReLU) is proposed guiding the model to explore the convergence position closer to the global optimum. Extensive experiments on practical and simulation communication applications demonstrate that Tri-Net achieves superior classification performance compared with a series of state-of-the-art deep learning models, especially at low SNRs.}, archive = {J_EAAI}, author = {Qinghe Zheng and Xinyu Tian and Zhiguo Yu and Mingqiang Yang and Abdussalam Elhanashi and Sergio Saponara}, doi = {10.1016/j.engappai.2024.109861}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109861}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Robust automatic modulation classification using asymmetric trilinear attention net with noisy activation function}, volume = {141}, year = {2025}, }
• (2025). Real-time pricing strategy considering carbon emissions and time coupling in smart grid: A binary integer bilevel optimization model with decision-making. EAAI, 141, 109858. (www)

  @article{ , abstract = {A real-time pricing strategy considering carbon emissions and time coupling in smart grid is proposed and the binary integer bilevel optimization model with decision-making is established in this paper. In the model, users' loads are divided into two types, i.e., interruptible loads and non-interruptible loads; besides, renewable energy resources generation and their uncertainties, and unplanned electricity outage are also taken into account. The startup time slot of users' non-interruptible loads is determined by the current prices and the future prices, which is forecasted by the exponential smoothing method. Finally, the reinforcement learning algorithm is designed to solve the proposed model, which is distributed and protects each user's privacy. The results show that prices forecast is not only helpful for users to determine the startup time slot to save electricity payment and increase average welfare, but also increases the power supplier’ s average profit and average social welfare; ladder-type carbon prices incorporated into the real-time electricity prices and outage compensation are effective to lead users to consume more electricity that is from renewable energy sources generation, which can effectively reduce carbon emissions by approximately 22.44% compared with only the real-time prices mechanism; multi-energy generation systems reduce the cost of fossil fuels, the total cost of power generation and improve the economic efficiency of the system.}, archive = {J_EAAI}, author = {Yiling Luo and Yan Gao}, doi = {10.1016/j.engappai.2024.109858}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109858}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time pricing strategy considering carbon emissions and time coupling in smart grid: A binary integer bilevel optimization model with decision-making}, volume = {141}, year = {2025}, }
• (2025). Objective transformation-based and niche-based many-objective evolutionary algorithm with a two-step coordination mechanism. EAAI, 141, 109850. (www)

  @article{ , abstract = {Evolutionary algorithms have emerged as powerful tools for optimization. However, striking a balance between convergence and diversity in many-objective optimization remains a significant challenge. To address this gap, we propose TSEA-OTN, an objective transformation-based and niche-based many-objective evolutionary algorithm with a two-step coordination mechanism. Uniquely, TSEA-OTN operates without relying on relaxed Pareto dominance, reference vectors, or additional indicators. Instead, it utilizes prior knowledge about the curvature of the PF (Pareto optimal front) to transform the objectives of the population and establish niches. Additionally, a niche-assisted density estimation method is designed to measure the distribution of individual. The environmental selection process incorporates a two-step mechanism: in the former step, the niche-assisted density evaluation method identifies crowded individuals to prioritize diversity; in the latter step, the Euclidean distance among transformed individuals and convergence evaluation criteria are used to eliminate individuals within the same niche for promoting convergence. Finally, TSEA-OTN is evaluated against six state-of-the-art algorithms on DTLZ (Deb-Thiele-Laumanns-Zitzler), MaF (Many-objective function), WGF (Walking Fish Group) benchmark suites, as well as an engineering case study. Experimental results demonstrate the competitive performance of TSEA-OTN in solving many-objective optimization problems. This research not only advances the field of evolutionary computation but also provides novel solutions for real-world optimization.}, archive = {J_EAAI}, author = {Jiale Luo and Qinghua Gu and Xuexian Li and Lu Chen}, doi = {10.1016/j.engappai.2024.109850}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109850}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Objective transformation-based and niche-based many-objective evolutionary algorithm with a two-step coordination mechanism}, volume = {141}, year = {2025}, }
• (2025). Understanding aircrew circadian resynchronization following time zone flights. EAAI, 141, 109849. (www)

  @article{ , abstract = {The human body has evolved to maintain its biological functions in sync with the approximately 24-h environmental cycle caused by the Earth's rotation, forming a circadian rhythm akin to the day–night cycle. However, flight across time zones can disrupt this rhythm. Aircrew experience more severe jet lag symptoms compared to athletes or visitors who travel across time zones non-regularly, meaning it is crucial to address the occupational health issues arising from flights across time zones. Alleviating jet lag symptoms in aircrew necessitates exploration of the circadian adaptation following such flights. Therefore, it is essential to parameterise and validate a model to estimate the resynchronisation of aircrew's biological clocks with the day–night cycle in new time zones during stopovers after flights across time zones and after completing round-trip flights across different time zones. Empirical sleep data were collected from 127 pilots. Sleep metric data were gathered using wrist activity monitors during real commercial flight operations. By analysing those sleep metric data, the study validated the research hypothesis that circadian resynchronisation in aircrew after crossing time zones follows an exponential function. It also verified the parameters in the model of circadian adaptation: Aircrew's circadian rhythm phase adjustment rate is faster after flying westward rather than eastward across time zones, and there are two modes of ‘advance or delay’ in aircrew's circadian rhythm phase adjustment after flying eastward across time zones. Resultantly, this research offers a more accurate mathematical model of pilots' circadian adaptation rate, which can be used to predict and estimate the time course of aircrew's circadian resynchronisation with the day–night cycle in new time zones after flights across time zones. Moreover, this research can be used for the ‘re-synchronization of circadian rhythms of flight crews after flying across time zones’ by using artificial intelligence and intelligent computer program systems, such as expert systems. The proposed model can also predict and evaluate changes in the recovery of circadian disorders of pilots after flying across time zones, which can help the development of dynamic algorithms for real-time adjustments of work and rest schedules, providing immediate feedback and suggestions for adjustments to help aircrew adapt to new time zones more quickly. Further, the results can be integrated into the flight safety management system to collaborate with the health, safety, and operations management department and promote the application and development of artificial intelligence for health management and intelligent decision-making.}, archive = {J_EAAI}, author = {Sun Junya and Liao Yang and Wang Cong and Jia Hongbo}, doi = {10.1016/j.engappai.2024.109849}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109849}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Understanding aircrew circadian resynchronization following time zone flights}, volume = {141}, year = {2025}, }
• (2025). Ship course-keeping in waves using sample-efficient reinforcement learning. EAAI, 141, 109848. (www)

  @article{ , abstract = {Maintaining a steady course in waves is important for ships for various reasons such as safety, fuel efficiency, and comfort. This has commonly been addressed by adopting conventional control algorithms. Reinforcement learning (RL) methods, on the other hand, have demonstrated successful performance in a wide range of control problems. In this work, the performance of two RL agents (model-free and model-based) in comparison to a linear-quadratic regulator (LQR) is investigated in a numerical environment. The model-free RL agent performed better than the LQR with respect to keeping its course and minimizing the rudder usage. By applying model-based RL, the low sample efficiency and consequent long training times that typically complicate model-free RL were mitigated. As a result, the training time of the course-keeping agent was reduced by more than an order of magnitude. Moreover, the model-based agent learned to exclusively react to the low-frequency yaw motion while ignoring the first-order wave disturbances, thereby reducing the rudder usage considerably.}, archive = {J_EAAI}, author = {Justin Greep and Afşin Baran Bayezit and Bart Mak and Douwe Rijpkema and Ömer Kemal Kınacı and Bülent Düz}, doi = {10.1016/j.engappai.2024.109848}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109848}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ship course-keeping in waves using sample-efficient reinforcement learning}, volume = {141}, year = {2025}, }
• (2025). Online monitoring of weld cross-sectional shape using optical emission spectroscopy and neural network during laser dissimilar welding. EAAI, 141, 109847. (www)

  @article{ , abstract = {Laser dissimilar welding (LDW) has received substantial interest in manufacturing due to its advantages in enhancing product performance. However, offline quality monitoring remains prevalent in LDW manufacturing, resulting in high costs and extended production times. In this study, a novel deep neural network (DNN) model was developed to predict the cross-sectional shape of the weld in online, allowing for an intuitive assessment of the welding state during aluminum/copper (Al/Cu) LDW. The weld shape prediction (WSP) training model utilized three parallel DNNs. In each DNN, the mean squared error of the output was determined via forward propagation, and the weights and biases between DNN layers were adjusted through backpropagation of the error. The resilient backpropagation algorithm was selected as the optimization method for training. Optical signals collected by optical emission spectroscopy were used as DNN input. Principal component analysis (PCA), major emission line (MEL), and line intensity ratio (LIR) methods were used to preprocess input data. Representative weld cross-sectional images, obtained via singular value decomposition, served as DNN output for model training. For PCA and MEL, the WSP model using input data with Cu information (89.5% for PCA and 88.3% for MEL) showed higher average prediction accuracy (PA) compared to the case using input data without Cu information (81.2% for PCA and 82.6% for MEL) because Cu emission lines better reflected LDW-associated physical phenomena. The highest average PA (89.8%) was achieved using input data preprocessed by the LIR method. These results demonstrate the applicability of the proposed WSP to online LDW process monitoring with high accuracy and robustness.}, archive = {J_EAAI}, author = {SeungGu Kang and Sangmoo Jeon and Kihwan Ryu and Joonghan Shin}, doi = {10.1016/j.engappai.2024.109847}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109847}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Online monitoring of weld cross-sectional shape using optical emission spectroscopy and neural network during laser dissimilar welding}, volume = {141}, year = {2025}, }
• (2025). An integrated feature extraction framework of linear multi-layer perceptron to reduce computation complexity for remaining useful life prediction. EAAI, 141, 109846. (www)

  @article{ , abstract = {Recently, there has been a growth in deep learning-based solutions for RUL prediction, although these increasingly complex models have significantly improved prediction performance, these studies typically overlook the computational and storage resources required for model deployment. Thus, we attempt to construct a lightweight model based on a simple linear multi-layer perceptron (MLP) that achieves prediction performance comparable to complex models, while ensuring easier deployment on resource-constrained edge devices. Firstly, a feature reconstruction method based on unsupervised clustering is proposed, which uses the K-means algorithm to perform unsupervised clustering on the variable operating condition data, and then standardization is conducted according to the mean and variance of each class, so as to separate the degradation features from the operating condition. Then, we propose a time-series linear extractor (TiLE) architecture for extracting degradation features from multi-sensor data. This lightweight framework achieves the advantage of linear computational scalability, which improves the inference efficiency of the model. The feature recalibration mechanism of TiLE is designed to reduce the interference of random factors, which is conducive to improving the prediction accuracy. Experimental results on the NASA turbine engine dataset show that the TiLE-based model outperforms state-of-the-art methods while achieving superior computational complexity and inference efficiency.}, archive = {J_EAAI}, author = {Hui Gao and Qingwen Guo and Zhizheng Zhang and Yibin Li}, doi = {10.1016/j.engappai.2024.109846}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109846}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated feature extraction framework of linear multi-layer perceptron to reduce computation complexity for remaining useful life prediction}, volume = {141}, year = {2025}, }
• (2025). Multi-modal multi-sensor feature fusion spiking neural network algorithm for early bearing weak fault diagnosis. EAAI, 141, 109845. (www)

  @article{ , abstract = {To address the problem of diagnosing early weak faults in bearings, a multi-modal multi-sensor feature fusion Spiking Neural Network algorithm is proposed in this paper. At the feature level, Principal Component Analysis is utilized to merge the vibration signals in the horizontal and vertical directions from multiple sensors, so as to fully utilize the feature information of multi - modal signals. Subsequently, Moving Average denoising is adopted to highlight the information features of weak faults. Signal augmentation is achieved by serially connecting the fused vibration signals of multiple sensors, and the feature extraction capability is further enhanced by using the Continuous Wavelet Transform to extract the time frequency modal features, finally constructing a Spiking Neural Network intelligent diagnostic model. Experimental results show that this algorithm is superior to other feature extraction and diagnostic algorithms in accurately diagnosing early weak faults in bearings.}, archive = {J_EAAI}, author = {Zhenzhong Xu and Xu Chen and Jiangtao Xu}, doi = {10.1016/j.engappai.2024.109845}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109845}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-modal multi-sensor feature fusion spiking neural network algorithm for early bearing weak fault diagnosis}, volume = {141}, year = {2025}, }
• (2025). Comprehensive fault diagnosis of lithium-ion batteries: An innovative approach based on hybrid coding and genetic search. EAAI, 141, 109842. (www)

  @article{ , abstract = {The performance and lifespan of lithium-ion batteries are significantly impacted by various faults. In particular, concurrent faults result in complex crossover and coupling issues, which present considerable challenges to fault diagnosis. To address these issues, this paper proposes a comprehensive fault diagnosis method utilizing hybrid coding and genetic search. The Lyapunov index between predicted and faulty battery states is applied to calculate trajectory divergence rates, facilitating the detection of abnormal battery conditions. Fault modes are uniformly characterized using a hybrid code, and a population is generated for genetic search optimization, from which the most suitable solution is selected. Experimental results indicate that, compared to the original approach, the improved Lyapunov index enhances detection accuracy and robustness in handling abnormal data. Furthermore, the hybrid coding and genetic search-based algorithm effectively isolates and classifies faults in multi-fault scenarios.}, archive = {J_EAAI}, author = {Chunhui Ji and Guang Jin and Ran Zhang}, doi = {10.1016/j.engappai.2024.109842}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109842}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Comprehensive fault diagnosis of lithium-ion batteries: An innovative approach based on hybrid coding and genetic search}, volume = {141}, year = {2025}, }
• (2025). Agile combination of advanced booking models for short-term railway arrival forecasting. EAAI, 141, 109841. (www)

  @article{ , abstract = {Accurate forecasts are essential for allocating perishable resources. As arrivals are the accumulation of reservations, utilizing reservation records to construct forecasting models is novel but challenging. This study focuses on agilely combining individual alternatives via a complexity-based selecting procedure. Real records over a year from a railway corporation are collected to form booking curves. Eleven complexity indicators are proposed to describe the characteristics of booking curves and five advanced booking models are constructed. Then five machine learning classifiers are applied to map the relationships between complexity and model suggestion. The empirical results show promising outcomes when a booking point is distant from the service day and cautiously selecting forecasting alternatives for combination is critical to predictive accuracy. More specifically, the Support Vector Machine combination achieves a 10% performance improvement compared with the all-inclusive average method. Additionally, the proposed scheme can achieve equal or better performance compared with individual models.}, archive = {J_EAAI}, author = {Tsung-Hsien Tsai}, doi = {10.1016/j.engappai.2024.109841}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109841}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Agile combination of advanced booking models for short-term railway arrival forecasting}, volume = {141}, year = {2025}, }
• (2025). TSPCS-net: Two-stage pavement crack segmentation network based on encoder-decoder architecture. EAAI, 141, 109840. (www)

  @article{ , abstract = {Crack segmentation is of great significance in automatic pavement crack detection based on image recognition. Although recent convolutional neural network (CNN)-based segmentation methods have shown promising performance, accurate pavement crack segmentation still faces some challenges, such as various crack sizes, class imbalance issues, and background interference. To overcome these challenges, a compact two-stage pavement crack segmentation network based on encoder-decoder architecture (TSPCS-Net) is proposed, which includes a classification network and a segmentation network. The classification network, consisting of a feature extraction module transferred from the segmentation network and a lightweight feature fusion module, is used to quickly classify and eliminate crack-free images that existed in large numbers in actual pavement image datasets. The segmentation network is constructed based on an encoder-decoder architecture for precise pixel-level segmentation of the samples determined as crack images. Specifically, to extract multi-scale crack features, a novel multi-scale encoder module is designed by combining dilated convolution and residual structure. Then, a left-side path (LSP) is designed to alleviate the influence of class imbalance on feature extraction. Finally, an attention module with high-dimensional features guiding low-dimensional features (AM-HGL) is proposed to focus on crack-relevant features and suppress interference information. The effectiveness of the proposed TSPCS-Net is validated on a self-made unmanned aerial vehicles pavement crack (UAVPC) dataset and two public pavement distress datasets, and extensive experiments show that the proposed method outperforms current state-of-the-art methods in terms of segmentation performance and efficiency, which can meet the needs of pavement crack segmentation in practical application scenarios.}, archive = {J_EAAI}, author = {Biao Yue and Jianwu Dang and Qi Sun and Yangping Wang and Yongzhi Min and Feng Wang}, doi = {10.1016/j.engappai.2024.109840}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109840}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {TSPCS-net: Two-stage pavement crack segmentation network based on encoder-decoder architecture}, volume = {141}, year = {2025}, }
• (2025). Tow train routing in narrow aisles: A grid-based approach with blocking constraints. EAAI, 141, 109839. (www)

  @article{ , abstract = {We address the problem of routing a fleet of tow trains in a manufacturing plant or warehouse with narrow aisles. In a narrow-aisles workspace, tow trains cannot reverse in place, and may block each other’s movements. Our approach divides the workspace into a grid of pixels. This allows for the detection of blocking vehicles. The shortest paths between workstations in this grid-based layout are determined using Dijkstra’s algorithm. The problem of routing the vehicles is formulated as a generalized vehicle routing problem (GVRP). We propose a mathematical model for this problem and develop a straightforward heuristic to solve it. The heuristic uses a simulated annealing framework and several problem-specific neighborhood operators. A blocking detection and avoidance algorithm is implemented to remove detected vehicle collisions. A computational study confirms the ability of the heuristic to find efficient solutions to this challenging problem.}, archive = {J_EAAI}, author = {Gül Gündüz Mengübaş and Kenneth Sörensen and Muhammed Kotan}, doi = {10.1016/j.engappai.2024.109839}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109839}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Tow train routing in narrow aisles: A grid-based approach with blocking constraints}, volume = {141}, year = {2025}, }
• (2025). BAIoT-EMS: Consortium network for small-medium enterprises management system with blockchain and augmented intelligence of things. EAAI, 141, 109838. (www)

  @article{ , abstract = {The rapid adoption of Augmented Intelligence of Things (AIoT) in enterprise management (EM) presents significant challenges in securely managing and exchanging information. This study introduces a blockchain-based platform, BAIoT-EMS, designed to enhance security and efficiency in AIoT-enabled EM systems. The platform leverages a consortium network and InterPlanetary File Storage (IPFS) for secure storage and transaction management, supported by smart contracts to automate and safeguard processes like device registration. A novel multi-proof-of-work consensus mechanism is implemented to analyze, validate, and verify AIoT transactions while minimizing resource consumption. Simulation results demonstrate a 63.51% improvement in performance and an 11.75% reduction in computational power usage, highlighting the effectiveness of the proposed framework.}, archive = {J_EAAI}, author = {Abdullah Ayub Khan and Jing Yang and Asif Ali Laghari and Abdullah M. Baqasah and Roobaea Alroobaea and Chin Soon Ku and Roohallah Alizadehsani and U. Rajendra Acharya and Lip Yee Por}, doi = {10.1016/j.engappai.2024.109838}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109838}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {BAIoT-EMS: Consortium network for small-medium enterprises management system with blockchain and augmented intelligence of things}, volume = {141}, year = {2025}, }
• (2025). Efficient extended-graph multi-matrices subspace learning: Toward meta-similarity and efficiency promotion. EAAI, 141, 109835. (www)

  @article{ , abstract = {Multi-view learning (MVL) has become a hot topic due to it can view a problem from different perspectives and settle it comprehensively. The consensus and complementarity principles provide important guidances for MVL. A category of MVL methods, called multi-matrices learning (MML), can excavate additional geometry information from each sample, and shows superior performance. However, existing MVL models primarily follow two types of sample relationships, each for one principle, whereas, there are actually three types of relationships. In this paper, we propose an efficient extended-graph multi-matrices subspace learning (EEGMMSL) approach based on MML, incorporating three types of relationships guaranteeing these two principles. The diagonal terms of our extended-graph can maintain the intro-view sample-similarity-relationships, embed samples of each view to a easier-separated-subspace, realizing the complementarity principle. The off-diagonal items of our extended-graph sustain the cross-view sample-similarity-relationships, and keep these embedded subspaces consensus, guaranteeing the consensus principle from sample level. By further maintaining the meta-similarity in our extended-graph, it can rectify the relationships in these subspaces, so that guarantees consensus principle from the feature level. Our approach can unify most of the relation-based MML methods and is more flexible in solving different problems. Focusing on the efficiency, we also proposed an new M-separate optimization algorithm to more efficiently optimize our objective function, instead of its traditional heuristic-gradient-descent optimization. This greatly reduces the computational complexity of MML. Experiments show the superior performance of EEGMMSL, and the remarkable time advantage of the M-separate optimization algorithm. The ideas in our paper can also be extended to other MVL methods.}, archive = {J_EAAI}, author = {Maoye Ren and Zhe Wang and Xinhai Yu}, doi = {10.1016/j.engappai.2024.109835}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109835}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Efficient extended-graph multi-matrices subspace learning: Toward meta-similarity and efficiency promotion}, volume = {141}, year = {2025}, }
• (2025). Temporal stability and non-unique solution of reacting eyring powell flows over shrinking wedges using neural networks. EAAI, 141, 109828. (www)

  @article{ , abstract = {In this paper, author presents an innovative artificial intelligence techniques based on deep learning simulation algorithms due to wide range of applications in science, robotics and engineering. The deep learning simulation algorithm using the Levenberg-Marquardt Scheme with Back Propagation Neural Networks (LMS-BPNN) is evaluated in the flow investigation of the chemically reacting non-Newtonian fluid. An efficient similarity variable is applied to change Partial Differential Equations (PDEs) of a considered flow problem into dimensionless Ordinary Differential Equations (ODEs). An error is found to be 10 − 4 with function fit for scenarios 1–5, while performance in terms of mean squared error is found to be 10 − 10 . It is seen that flow response output i.e. f ′ ( η ), θ ( η ) and ϕ ( η ) meet boundary requirements for different scenarios 1–5 with the deep learning-based technique LMS-BPNN. The dual solution is evaluated for flow response output parameter i.e. C f x , N u x and S h x for different parameters with the proposed LMS-BPNN. The dual nature of C f x is calculated for three cases of scenario 1, and the critical value is found to be − 0.96195 , − 1.0581 , and − 1.11 . The dual solution of S h x is calculated for various values of K c = 0.2 , 0.4 , 0.6 , and it is found that λ c = − 1.1175 remains the same. The perturbation scheme is applied to the boundary layer problem to obtain the eigenvalues problem. The unsteady solution f ( η , τ ) converges to steady solution f o ( η ) for τ → ∞ when γ ≥ 0 . However, an unsteady solution f ( η , τ ) diverges to a steady solution f o ( η ) for τ → ∞ when γ < 0 . It is found that the boundary layer thickness for the second (lower branch) solution is higher than the first (upper branch) solution. This investigation is the evidence that the first (upper branch) solution is stable and reliable.}, archive = {J_EAAI}, author = {M.I. Khan and A. Zeeshan and M.B. Arain and A.S. Alqahtani and M.Y. Malik}, doi = {10.1016/j.engappai.2024.109828}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109828}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Temporal stability and non-unique solution of reacting eyring powell flows over shrinking wedges using neural networks}, volume = {141}, year = {2025}, }
• (2025). Automated dimensional quality inspection of super-large steel mesh using fixed-spacing detection transformer and improved oriented fast and rotated brief. EAAI, 141, 109827. (www)

  @article{ , abstract = {With the advance of automation in constructing simply supported box girders for high-speed railways, traditional manual binding of steel mesh has been transmitted into automated welding. Nevertheless, dimensional quality inspection of steel mesh is still stagnant in manual operation, which is highly cost-ineffective and time-consuming. Current research lacks real-time and accurate detection methods for super-large steel mesh up to 32 m × 10 m. This paper proposes an efficient and precise dimensional quality inspection method for super-large steel mesh. Firstly, an automated image acquisition framework based on customized devices and method is proposed to synchronize equipment processing and inspection. The geometric features of the steel bar spacing are then explored, leading to the design of a fixed-spacing deformable attention module for detecting rebar images. Additionally, a “tic-tac-toe” template, replacing the traditional “discrete circle” template, is designed to enhance the accuracy of oriented fast and rotated brief. Experimental validation indicates a 0.156 improvement in the average precision of the fixed-spacing deformable attention module and a 39.6% increase in the accuracy of oriented fast and rotated brief. Field testing in a high-speed railway girder yard in China demonstrates second-level efficiency and millimeter-level accuracy of the proposed method. This study improves the deformable detection transformer, oriented fast, rotated brief and introduces a dimensional quality inspection method suitable for super-large steel mesh. It contributes significantly to artificial intelligence via replacing manual labor, promoting automation in high-speed railway construction, and thus benefiting practical engineering.}, archive = {J_EAAI}, author = {Xinfei Guo and Yimiao Huang and Shaopeng Zhang and Guowei Ma}, doi = {10.1016/j.engappai.2024.109827}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109827}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automated dimensional quality inspection of super-large steel mesh using fixed-spacing detection transformer and improved oriented fast and rotated brief}, volume = {141}, year = {2025}, }
• (2025). A novel spatial–temporal graph convolution network based on temporal embedding graph structure learning for multivariate time series prediction. EAAI, 141, 109826. (www)

  @article{ , abstract = {The prediction of multivariate time series is a pivotal research area in data mining, offering extensive practical applications in many real-world scenarios, including transportation, finance, energy systems, the Internet of Things. Accurately predicting multivariate time series is challenging due to the complex temporal and spatial dependencies among variables. To tackle this challenge, this study proposed a deep learning model utilizing graph neural networks for predicting multivariate time series. Specifically, a multivariate time series is modeled as a graph, with nodes representing variables, edges indicating their interdependencies, and the time series data serving as node attributes. We leverage the temporal convolutional network to construct a graph structure learning module that captures the underlying dependencies between variables through the learned adjacency matrix. The prediction model was built by integrating Long Short-Term Memory networks and graph neural networks, enabling the simultaneous capture of temporal and spatial dependencies in multivariate time series data. Additionally, to mitigate the issue of over-smoothing in graph neural networks, we incorporated the Laplacian sharpening technique into our model. The proposed method is generalizable for handling multivariate time series data, as it does not require a pre-defined adjacency matrix among variables. We empirically evaluated the performance of our method through extensive experiments conducted on six real-world datasets, the experimental results indicated that our method could effectively improve the accuracy of multivariate time series prediction.}, archive = {J_EAAI}, author = {Tianyang Lei and Jichao Li and Kewei Yang and Chang Gong}, doi = {10.1016/j.engappai.2024.109826}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109826}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel spatial–temporal graph convolution network based on temporal embedding graph structure learning for multivariate time series prediction}, volume = {141}, year = {2025}, }
• (2025). A fast recognition framework for identifying damage levels in rotating and small target solar greenhouse under complex scenarios. EAAI, 141, 109824. (www)

  @article{ , abstract = {Rapid identification, localization, and accurate counting are essential to prevent the abandonment of agricultural infrastructure like solar greenhouses, thereby avoiding the waste of agricultural land resources. However, detection difficulties include limited feature information from rotating small-target greenhouses, complex background environments, and inaccurate quantification of errors between predicted and actual rotating frames. To address these challenges, this study introduces an innovative framework for accurately locating and counting greenhouses with various damage levels, particularly suited for complex scenarios with multiple damage levels. Solar greenhouses are classified based on quantitative damage indicators, utilizing existing post-disaster building classification criteria from vertical remote sensing imagery. The proposed model utilizes MobileNetV3 as the backbone feature extraction network to reduce model parameters and maintain a lightweight structure. The lightweight operator and rotation target enhancement module are integrated into the head and neck layers of the network to form the Carafe-EVCBlock (CE) module, ultimately improving the loss function to enhance the feature representation and detection accuracy of rotation targets. Comprehensive experimental investigations and performance comparisons validate the superiority of this framework for detecting small, rotating greenhouse targets in dense scenarios. Furthermore, a PyQt5-based visualization system was developed for real-time detection and display of remote sensing images, demonstrating high accuracy in detecting and classifying damage levels in solar greenhouses.}, archive = {J_EAAI}, author = {Jinhao Zhang and Danni Jia and Bin He and Wenwen Li and Xinyue Ren and Cailong Cheng and Quan Fan}, doi = {10.1016/j.engappai.2024.109824}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109824}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A fast recognition framework for identifying damage levels in rotating and small target solar greenhouse under complex scenarios}, volume = {141}, year = {2025}, }
• (2025). Group decision-making strategy based on aggregation operators of linguistic confidence interval neutrosophic numbers in a linguistic neutrosophic multivalued scenario. EAAI, 141, 109823. (www)

  @article{ , abstract = {In the scenario of linguistic neutrosophic multivalued sets (LNMSs), LNMS can capture true, false, and indeterminate linguistic multivalued information and effectively represent group decision making information in multiple criteria group decision-making (MCGDM) issues. Therefore, the objective of this study is to develop an MCGDM strategy using linguistic confidence interval neutrosophic number (LCINN) trigonometric aggregation operators to effectively tackle MCGDM problems with periodicity in an LNMS scenario. To do so, this study first presents a transformation approach from LNMS to the linguistic confidence interval neutrosophic set to ensure the confidence level of linguistic term sequences (LTSs) in LNMS from the perspective of probability estimation and to address the operation issue between different LTS lengths. Next, the linguistic sine t-norm, cosine t-conorm, and trigonometric operation laws of LCINNs are defined to include the periodic features of the sine and cosine functions. Then, the LCINN trigonometric weighted average and geometric operators are constructed to provide periodic aggregation tools for LCINNs in the LNMS scenario. Furthermore, an MCGDM strategy using the proposed aggregation operators is constructed to effectively tackle MCGDM problems with periodicity in LNMS scenarios. Finally, the constructed MCGDM strategy is applied to a choice case of energy storage technologies, and then its validity of the ranking results is presented by comparing it with the existing MCGDM methods in the scenarios of linguistic neutrosophic uncertain sets and LNMSs. In addition, the proposed strategy (no learning process) is simpler in the decision-making process than machine learning or artificial intelligence decision-making algorithms.}, archive = {J_EAAI}, author = {Jun Ye}, doi = {10.1016/j.engappai.2024.109823}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109823}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Group decision-making strategy based on aggregation operators of linguistic confidence interval neutrosophic numbers in a linguistic neutrosophic multivalued scenario}, volume = {141}, year = {2025}, }
• (2025). A priori uncertainty quantification of reacting turbulence closure models using bayesian neural networks. EAAI, 141, 109821. (www)

  @article{ , abstract = {While many physics-based closure model forms have been posited for the sub-filter scale (SFS) in large eddy simulation (LES), vast amounts of data available from direct numerical simulations (DNS) create opportunities to leverage data-driven modeling techniques. Albeit flexible, data-driven models still depend on the dataset and the functional form of the model chosen. Increased adoption of such models requires reliable uncertainty estimates both in the data-informed and out-of-distribution regimes. In this work, we employ Bayesian neural networks (BNNs) to capture both epistemic and aleatoric uncertainties in a reacting flow model. In particular, we model the filtered progress variable scalar dissipation rate which plays a key role in the dynamics of turbulent premixed flames. We demonstrate that BNN models can provide unique insights about the structure of uncertainty of the data-driven closure models. We also propose a method for the incorporation of out-of-distribution information in a BNN, which can be used for out-of-distribution query detection. The efficacy of the model is demonstrated by a priori evaluation on a dataset consisting of a variety of flame conditions and fuels.}, archive = {J_EAAI}, author = {Graham Pash and Malik Hassanaly and Shashank Yellapantula}, doi = {10.1016/j.engappai.2024.109821}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109821}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A priori uncertainty quantification of reacting turbulence closure models using bayesian neural networks}, volume = {141}, year = {2025}, }
• (2025). REINFORCE with bound-guided gradient estimator for the traveling salesman problem toward scale generalization. EAAI, 141, 109819. (www)

  @article{ , abstract = {The Traveling Salesman Problem (TSP), a foundational combinatorial optimization challenge, has significant theoretical and practical relevance across domains. With the rise of Deep Reinforcement Learning (DRL), researchers have increasingly applied DRL-based solutions to the TSP. However, most studies have focused on DRL network design, with less emphasis on algorithmic refinement. In this paper, We propose a knowledge-inspired REINFORCE algorithm called REINFORCE with Bound-guided Gradient Estimator (RIDGE), helping train the model towards global optimum and scale generalization for TSP. By incorporating Beardwood–Halton–Hammersley Theorem as prior knowledge to ascertain the theoretical optimal path bound for the TSP, we design an adaptive baseline for gradient estimator during model training to approximate the theoretical optimal bound, effectively guiding the model towards global optimization. This approach enables it to learn the underlying theoretical principles of the TSP problem and helps prevent the model from getting trapped in local optima thereby enhancing the ability of scale generalization. Notably, RIDGE’s network-agnostic structure makes it highly adaptable across combinatorial optimization frameworks. Extensive experimentation on randomly generated and publicly available benchmarks demonstrates that our approach yields performance on small-scale TSP instances comparable to state-of-the-art (SOAT) DRL methods, while exhibiting noticeable improvements in scale generalization capabilities. For example, when trained on a dataset with 20 nodes, the model trained using the RIDGE method outperforms the baseline model on 95.25% of the TSPLIB instances. Additionally, we found that reducing training set size moderately enhances generalization capability.}, archive = {J_EAAI}, author = {Haopeng Duan and Kaiming Xiao and Lihua Liu and Haiwen Chen and Hongbin Huang}, doi = {10.1016/j.engappai.2024.109819}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109819}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {REINFORCE with bound-guided gradient estimator for the traveling salesman problem toward scale generalization}, volume = {141}, year = {2025}, }
• (2025). Optimizing resource and power consumption in a cloud environment via consolidation and placement investigation: A survey. EAAI, 141, 109818. (www)

  @article{ , abstract = {Network Function Virtualization ( N F V ) is the latest technological innovation in which a new structure has been introduced to make a network more flexible, more controllable, and more cost-effective. This technology ensures the sharing of physical resources while reducing losses. The N F V technology enables the simultaneous execution of a variety of network functions over a shared network. Additionally, it imposes various challenges, such as placement, orchestration, and consolidation. Despite the considerable importance of the use of N F V s , several technical problems remain to be identified and addressed. The consolidation of Virtual Network Functions ( V N F ) can be considered one of the main challenges for the better operation and management of V N F . This consolidation consists of allocating certain virtual resources to the same physical machine or between two virtual resources to balance the exploitation of resources and minimize energy consumption. In this article, we present a detailed and comprehensive comparative study of consolidation problems by presenting a new and original classification of the main approaches proposed in the literature. Our goal is to give scholars and practitioners more experience from the literature on how to consolidate and place elements to optimize energy consumption and resources in a cloud environment.}, archive = {J_EAAI}, author = {Wided Khemili and Jalel Eddine Hajlaoui and Mohamed Nazih Omri}, doi = {10.1016/j.engappai.2024.109818}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109818}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing resource and power consumption in a cloud environment via consolidation and placement investigation: A survey}, volume = {141}, year = {2025}, }
• (2025). Modal disentangled generative adversarial networks for bidirectional magnetic resonance image synthesis. EAAI, 141, 109817. (www)

  @article{ , abstract = {Magnetic resonance imaging (MRI) is commonly used both in clinical diagnosis and scientific research. Owing to the high cost, time constraints, and limited application of multi-contrast MRI images obtained from metallic implants, it incurs low throughput and misses a specific modality. Medical image cross-modal synthesis based on Artificial Intelligence (AI) technologies is proposed to synthesize the desired missing modal images. However, it still suffers from low expandability, invisible latent representations, and poor interpretability. We thus propose modal disentanglement generative adversarial networks for bidirectional T1-weighted (T1-w) and T1-weighted (T2-w) medical image synthesis with controllable cross-modal synthesis and disentangled interpretability. Firstly, we construct a cross-modal synthesis model to achieve bidirectional generation between T1-w and T2-w MRI images, which can be easily extended for adaptive modality synthesis without training multiple generators and discriminators. Then, we use an easily trained deep network to disentangle deep representations in latent space and map representations in latent space into pixel space to visualize morphological images and yield multi-contrast MRI images with controllable feature generation. Besides, we construct an easy-to-interpret deep structure by incorporating morphology consistency to preserve edge contours and visualize deep representations in latent space to enable interpretability, which is critical for artificial intelligence oriented to engineering applications and clinical diagnostics. The experiments demonstrate that ours outperforms recent state-of-the-art methods with average improvements of 15.8% structural similarity (SSIM), 12.7% multiscale structural similarity (MSIM), 38.2% peak signal-to-noise ratio (PSNR) and 5.2% visual information fidelity (VIF) on benchmark datasets.}, archive = {J_EAAI}, author = {Liming Xu and Yanrong Lei and Jie Shao and Xianhua Zeng and Weisheng Li}, doi = {10.1016/j.engappai.2024.109817}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109817}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Modal disentangled generative adversarial networks for bidirectional magnetic resonance image synthesis}, volume = {141}, year = {2025}, }
• (2025). Simplified and intelligent controllers for multi-input multi-output processes. EAAI, 141, 109816. (www)

  @article{ , abstract = {Designing a controller to serve a process like a multi-input multi-output (MIMO) system is complicated due to unknown parameters, unmodeled nonlinearities, and unknown multiple-input couplings. Several controllers, such as the proportional-integral-derivative (PID) controller, have been introduced for the MIMO system to enhance its dynamic response due to their simplicity in design. This study aims to improve and develop the PID controller to be simpler and more intelligent. This paper proposes the design and verification of three controllers based on the multi-degree of freedom (MDOF) concept: the simplified optimum intelligent PID (SOI-PID) controller, the simplified universal intelligent PID (SUI-PID) controller, and the MDOF fuzzy logic controller (MDOF-FLC) for a MIMO system. A comparison study with different algorithms like the fuzzy logic controller (FLC) and the two degree of freedom (DOF)-PID controller was implemented to test and evaluate the proposed controllers under two MIMO systems: a proposed simple MIMO system and the four DOF robot arm manipulator. Simulation and experimental results show that the proposed SOI-PID and SUI-PID controllers respond faster with a better response than the MDOF-FLC, FLC, and two DOF-PID controller. The proposed controllers convert the PID controller into an intelligent controller via a simple design compared with intelligent algorithms.}, archive = {J_EAAI}, author = {Mohamed Fawzy El-Khatib and Feaka M.H. Khater and Essam Hendawi and Mohamed I. Abu El-Sebah}, doi = {10.1016/j.engappai.2024.109816}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109816}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Simplified and intelligent controllers for multi-input multi-output processes}, volume = {141}, year = {2025}, }
• (2025). A novel local enhanced channel self-attention based on transformer for industrial remaining useful life prediction. EAAI, 141, 109815. (www)

  @article{ , abstract = {Remaining useful life (RUL) prediction is a foundational technique for predictive maintenance (PdM) and is critical to ensuring the reliability and safety of complex industrial machines. Recently, while advanced deep learning architectures like recurrent neural network (RNN), convolutional neural network (CNN) and self-attention (SA) have been widely used for RUL prediction, existing methods still face difficulties in simultaneously processing global long-term dependencies and local contextual information of sequence units as well as the spatial correlations of industrial multi-sensors. In this article, we propose local enhanced channel self-attention based on Transformer (LECformer), a novel deep RUL prediction method to overcome these issues. LECformer can more effectively capture the long-term dependencies by Transformer architecture compared with RNN/CNN-based methods. Moreover, LECformer proposes a novel local enhanced channel self-attention (LECSA) mechanism to replace the traditional SA of vanilla Transformer, which can adaptively extract both long-term dependencies and local contextual information, while dynamically weighting the importance of different channels to improve predictive performance. Two widely used turbofan engine datasets and a bearing dataset are applied to validate the effectiveness of the proposed method. Experimental results show that the LECformer significantly outperforms the state-of-the-art RUL prediction methods.}, archive = {J_EAAI}, author = {Zhizheng Zhang and Wen Song and Qiong Wu and Wenxu Sun and Qiqiang Li and Lei Jia}, doi = {10.1016/j.engappai.2024.109815}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109815}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel local enhanced channel self-attention based on transformer for industrial remaining useful life prediction}, volume = {141}, year = {2025}, }
• (2025). MCAGU-net: A model for composite fault diagnosis of multi-sensor node networks. EAAI, 141, 109814. (www)

  @article{ , abstract = {Diagnosing sensor node faults into specific types is beneficial for selecting targeted fault handling strategies and recovering fault data. However, current diagnostic models for multi-sensor nodes ignore the potential interactions between sensing units within nodes, resulting in a low diagnosis accuracy and the failure to clearly identify faulty sensing units, hindering effective fault recovery. To address this, this paper proposes, for the first time, a comprehensive solution for sensing unit level composite fault diagnosis in multi-sensor nodes. The approach utilizes graph structures to uniformly model the signal features of multiple sensing units and their correlations, thereby assisting the diagnostic model in effectively capturing composite fault characteristics. Furthermore, a multi-channel attention graph U-Net (MCAGU-Net) diagnostic model is introduced, which reduces channel redundancy and adopts an attention mechanism to enhance multi-channel collaboration, effectively handling the diversity and complexity of composite fault patterns. Empirical results on six datasets show that this approach outperforms the baseline methods on most metrics. Particularly, the diagnostic accuracy reaches 96.9–98.4%, regardless of whether the datasets are class-balanced or imbalanced. Compared to the baseline methods, this approach improves recall rate by over 6.2% on balanced datasets and by over 10.25% on imbalanced datasets.}, archive = {J_EAAI}, author = {Kangshuai Zhang and Quancheng Zhang and Qi Liu and Yang Yang and Yunduan Cui and Lei Peng}, doi = {10.1016/j.engappai.2024.109814}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109814}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MCAGU-net: A model for composite fault diagnosis of multi-sensor node networks}, volume = {141}, year = {2025}, }
• (2025). Triplet trustworthiness validation with knowledge graph reasoning. EAAI, 141, 109813. (www)

  @article{ , abstract = {Knowledge graph is widely used in intelligent analysis and graph structure applications, which uses triplets to describe the relations and facts in the real world. In the process of incorporating external triplets into existing knowledge graph, it is vital to verify the trustworthiness level of triplet knowledge for building a comprehensive knowledge graph. In this paper, we establish a model for evaluating knowledge graph trustworthiness based on a triple strategy. The model quantifies the meaning of entities and relations expressed in the knowledge graph and obtains the quantification of triple trustworthiness measurement. And it integrates the internal semantic information of triplets and the global information of the knowledge graph, including entity-level, relation-level, and graph-level trustworthiness measurement, and finally uses multi-layer perceptron fusion to obtain the final score. This paper analyzes the effectiveness of the model output trustworthiness values and conducts error detection experiments in five real-world knowledge graph datasets. Experimental results show that compared with other models, our model has achieved significant effects.}, archive = {J_EAAI}, author = {Geng Zhang and Yu-Jie Xiong and Jian-Peng Hu and Chun-Ming Xia}, doi = {10.1016/j.engappai.2024.109813}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109813}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Triplet trustworthiness validation with knowledge graph reasoning}, volume = {141}, year = {2025}, }
• (2025). A novel radial basis function neural network classifier based on three-way decisions. EAAI, 141, 109811. (www)

  @article{ , abstract = {The radial basis function neural network (RBFNN) has gained widespread adoption due to its simple network topology and excellent learning capability. However, the performance of empirical formula-based RBFNN tends to deteriorate when faced with sample imbalance issues. In order to fully harness the potential of RBFNN classifier, inspired by three-way decisions (TWDs) conforming to the logic of human behavior, this paper investigates a novel classifier named TWDs-RBFNN that integrates TWDs and RBFNN. This novel artificial intelligence approach efficiently handles imbalanced samples in imbalanced classification tasks while achieving enhanced performance within a more streamlined topology. TWDs-RBFNN incorporates a TWDs module into the RBFNN framework, which dynamically stores and efficiently trains difficult-to-classify samples for reinforcement learning of intricate sample information and network topology optimization through parameter updates. Drawing upon current advancements in neuroscience and granule computing theory, we introduce gradient harmonizing mechanism to balance samples at the loss function level and handle outliers to avoid parameter bias and improve training efficiency. We also introduce information granules and establish a novel conditional probability formula to safeguard against information loss. The results of numerous experiments on imbalanced datasets demonstrate that TWDs effectively enhances the compactness of the network structure and improve its generalization ability in RBFNN. Specifically, we provide theoretical proof for the convergence of TWDs-RBFNN to analyze their effectiveness.}, archive = {J_EAAI}, author = {Tengbiao Li and Junsheng Qiao}, doi = {10.1016/j.engappai.2024.109811}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109811}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel radial basis function neural network classifier based on three-way decisions}, volume = {141}, year = {2025}, }
• (2025). A hybrid ensemble model to detect bitcoin fraudulent transactions. EAAI, 141, 109810. (www)

  @article{ , abstract = {Fraudulent transactions in the Bitcoin ecosystem wield substantial influence over both the economy and the level of trust within a blockchain network. The identification of fraudulent transactions is a crucial task within the financial system. However, existing detection methods predominantly rely on either an individual model or a single bagging or boosting ensemble model, leading to inadequate prediction accuracy and limited interpretability. To address this limitation, a hybrid ensemble model that integrates bagging (random forest (RF)) and boosting (categorical boosting (CatBoost)) is proposed, in which strong classifiers -CatBoost is used to replace the weak classifiers in RF, thus effectively improving the prediction performance. Empirical findings demonstrate the proposed hybrid ensemble framework can consistently yields the highest accuracy. Furthermore, we introduce extreme gradient boosting as a surrogate model that can obtain more accurate relations between actual labels and the predictions, compensating for the challenges of understanding complex models. Then, the visualization package based on Shapley additive explanations (SHAP) value is adopted for the interpretability analysis on the surrogate model, which contributes to the field of fraud detection by revealing the potential influencing factors behind the predicted results and providing unique insights into how fraud detection behavior can be detected.}, archive = {J_EAAI}, author = {Lifang Zhang and Ye Xuan and Zhenkun Liu and Zhiyuan Du and Shuai Wang and Jianzhou Wang}, doi = {10.1016/j.engappai.2024.109810}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109810}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid ensemble model to detect bitcoin fraudulent transactions}, volume = {141}, year = {2025}, }
• (2025). Detection of breath sounds in speech: A deep learning approach. EAAI, 141, 109808. (www)

  @article{ , abstract = {Breath sound detection from speech recordings has wide-ranging applications, from high-quality audio recordings to medical diagnostics. However, perceptual recognition of breath sounds for annotation is prone to errors, and breath sounds typically occupy only 5% of speech recordings, leading to significant class imbalance. Additionally, the limited availability of annotated data makes the application of deep learning (DL) methods challenging. This paper proposes the use of thermal and normal videos, alongside speech data, to mitigate annotation errors in breath sound detection. To address class imbalance, we leverage self-supervised learning (SSL), employing a jigsaw puzzle solver as a pretext task to augment training data and enhance model performance. The jigsaw puzzle solver helps address the class imbalance by creating a balanced task for pretraining, improving the performance of the downstream task. This work also uses convolutional neural networks (CNN) and bidirectional long short-term memory (BiLSTM) models to locate breath sounds in speech recordings accurately. The proposed SSL implementation achieves an F1-score of 96% in a speaker-independent configuration. The proposed algorithm has also been tested on publicly available audio recordings from YouTube 1 and the BiLSTM version is available for testing on Hugging Face. 2}, archive = {J_EAAI}, author = {K. Mohamed Ismail Yasar Arafath and Aurobinda Routray}, doi = {10.1016/j.engappai.2024.109808}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109808}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detection of breath sounds in speech: A deep learning approach}, volume = {141}, year = {2025}, }
• (2025). The lead-bismuth eutectic corrosion rate prediction and composition optimization of ferritic/martensitic steels by physics-guided neural network. EAAI, 141, 109807. (www)

  @article{ , abstract = {In developing safer lead-bismuth cooled nuclear power plants, accurately predicting corrosion rates of ferritic/martensitic (F/M) steel, used as structural materials, is crucial. Incorporated with prior physics knowledge, a physics-guided neural network (PGNN) model was developed for this purpose. By integrating physical knowledge about oxide layer development over time, the proposed PGNN model outperforms conventional models like Artificial Neural Network, Random Forest, eXtreme Gradient Boosting, and Support Vector Machine. The PGNN model demonstrated excellent physical consistency in terms of input feature interpretability and output result trends. Furthermore, with the particle swarm optimization (PSO) algorithm, an application on F/M steel composition design was conducted to enhance the lead-bismuth corrosion resistance with the proposed model. The short-term static lead-bismuth corrosion tests were conducted to assess the anti-corrosion performance of the optimized F/M steel, which showed consistency with the PGNN model's predictions.}, archive = {J_EAAI}, author = {Shaowu Feng and Xingyue Sun and Gang Chen and Xu Chen}, doi = {10.1016/j.engappai.2024.109807}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109807}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The lead-bismuth eutectic corrosion rate prediction and composition optimization of ferritic/martensitic steels by physics-guided neural network}, volume = {141}, year = {2025}, }
• (2025). An efficient perceptual video compression scheme based on deep learning-assisted video saliency and just noticeable distortion. EAAI, 141, 109806. (www)

  @article{ , abstract = {Perceptual Video Compression (PVC) is one of the most promising approaches to enhancing compression efficiency. Saliency and Just Noticeable Distortion (JND) are the most common visual models in PVC. Most methods solely rely on one of them to construct PVC schemes, while a few methods attempt to integrate both. However, the latter are often too simplistic and coarse, failing to fully exploit them to reduce bitrate. To address this issue, we propose an efficient PVC scheme based on Deep Learning-Assisted Saliency and JND (SJ-PVC), which is implemented on the latest Versatile Video Coding (VVC). Specifically, we first design a Structurally Simplified Network (SS-Net) for video saliency prediction, which removes the redundancy of multi-scale models and maintains excellent accuracy. Then, an Adaptive Quantization Parameter (QP) selection algorithm based on Saliency and JND (SJAQP) is proposed. It can dynamically adjust QP offsets according to the characteristics of the encoding unit and effectively integrate the offset effects of saliency and JND. Finally, we design a Rate-Distortion Optimization based on Saliency and JND (SJRDO), which incorporates JND to obtain perceptual distortion terms and adjusts rate–distortion balance through saliency to achieve more rational bitrate allocation. This scheme fully considers the properties of both visual models and further eliminates visual redundancy. Extensive experiments demonstrate that, when maintaining subjective visual quality, SJ-PVC can save 22.87% of the bitrate compared to VVC while also shortening encoding time, significantly enhancing video coding efficiency.}, archive = {J_EAAI}, author = {Yunzuo Zhang and Tian Zhang and Shuangshuang Wang and Puze Yu}, doi = {10.1016/j.engappai.2024.109806}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109806}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An efficient perceptual video compression scheme based on deep learning-assisted video saliency and just noticeable distortion}, volume = {141}, year = {2025}, }
• (2025). Applying deep learning and automated machine learning for enhanced state monitoring and health assessment of high-pressure heater in thermal power units. EAAI, 141, 109805. (www)

  @article{ , abstract = {High-pressure heaters are essential components in thermal power plants, where effective monitoring can significantly enhance safety and cost efficiency. This study proposes a novel method for monitoring and state prediction of high-pressure heaters, addressing challenges such as rapid load variations and extended low-load conditions that affect model accuracy. Using historical data for feature selection, the system employs a Convolutional Neural Network-bidirectional Long Short-term Memory (CNN-BiLSTM) network for state prediction. Hyperparameters are optimized using Bayesian methods, and automated machine learning enhances the model’s adaptability. Additionally, an improved Support Vector Data Description (SVDD) algorithm is used for both precise anomaly detection and health assessment. Validated with real-world data from a coal-fired power plant, the model achieved mean squared errors (MSE) of 1.3260 for the shell-side drain temperature and 1.4968 for the tube-side outlet temperature. The self-updating capability further reduced MSE by up to 56.98%, and anomaly detection accuracy reached 96.63%. These results highlight the model’s adaptability and effectiveness in improving monitoring and forecasting for high-pressure heaters.}, archive = {J_EAAI}, author = {Guoxiong Zhu and Yang Hu and Xiaoning Zhang and Jiyu Chen and Jizhen Liu}, doi = {10.1016/j.engappai.2024.109805}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109805}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Applying deep learning and automated machine learning for enhanced state monitoring and health assessment of high-pressure heater in thermal power units}, volume = {141}, year = {2025}, }
• (2025). A physics-informed deep neural network based beam vibration framework for simulation and parameter identification. EAAI, 141, 109804. (www)

  @article{ , abstract = {Physics-Informed Neural Networks (PINNs) integrate Neural Network (NN) models with physical phenomena governed by differential equations. This study proposes a PINN framework that can be used to simulate transverse vibrations in beams and determine their dynamic parameters. Previous studies have shown that it is difficult for neural networks to learn high-frequency dynamics. Therefore, a Neural Tangent Kernel (NTK) is used to manage the spectral bias phenomenon for high-frequency learning. The proposed model is used to simulate cases with different boundary conditions, and the results are validated using the finite element solutions. For the forward problem solutions, the error is found to be approximately O ( 1 0 − 5 ) at lower frequencies and O ( 1 0 − 1 ) at higher frequencies. For the inverse problem, the system parameters are identified with an error of 1.41%. Furthermore, physical experiments are used to demonstrate the validity of the proposed framework. The differences between the PINN results and the analytical and measured data were less than O ( 1 0 − 1 ) . The results show that PINNs have the potential to approximate the solutions of high-order partial differential equations and identify the dynamic parameters of structures governed by them.}, archive = {J_EAAI}, author = {Cem Söyleyici and Hakkı Özgür Ünver}, doi = {10.1016/j.engappai.2024.109804}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109804}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A physics-informed deep neural network based beam vibration framework for simulation and parameter identification}, volume = {141}, year = {2025}, }
• (2025). Object-size-driven design of convolutional neural networks: Virtual axle detection based on raw data. EAAI, 141, 109803. (www)

  @article{ , abstract = {As infrastructure ages, the need for efficient monitoring methods becomes increasingly critical. Bridge Weigh-In-Motion (BWIM) systems are crucial for cost-effective determination of loads and, consequently, the residual service life of road and railway infrastructure. However, conventional BWIM systems require additional sensors for axle detection, which must be installed in potentially inaccessible locations or places that interfere with bridge operation. This study presents a novel approach for real-time detection of train axles using sensors arbitrarily placed on bridges, providing an alternative to dedicated axle detectors. The developed Virtual Axle Detector with Enhanced Receptive Field (VADER) has been validated on a single-track railway bridge using only acceleration measurements, detecting 99.9% of axles with a spatial error of 3.69 cm. Using raw data as input outperformed the state-of-the-art spectrogram-based method in both speed and memory usage by 99%, thereby making real-time application feasible for the first time. Additionally, we introduce the Maximum Receptive Field (MRF) rule, a novel approach to optimise hyperparameters of Convolutional Neural Networks (CNNs) based on the size of objects. In this context, the object size relates to the fundamental frequency of a bridge. The MRF rule effectively narrows the hyperparameter search space, overcoming the need for extensive hyperparameter tuning. Since the MRF rule can theoretically be applied to all unstructured data, it could have implications for a wide range of deep learning problems, from earthquake prediction to object recognition.}, archive = {J_EAAI}, author = {Henrik Riedel and Steven Robert Lorenzen and Clemens Hübler}, doi = {10.1016/j.engappai.2024.109803}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109803}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Object-size-driven design of convolutional neural networks: Virtual axle detection based on raw data}, volume = {141}, year = {2025}, }
• (2025). Defect detection in fused deposition modelling using lightweight convolutional neural networks. EAAI, 141, 109802. (www)

  @article{ , abstract = {Early detection of defects in additive manufacturing (AM) processes has significant benefits such as reducing material waste and improving production time, which results in an increase in overall productivity. Existing methods for defect detection in AM lacks in generalizability, takes more time for training, and finds difficulty in identifying complex defects in the vertical plane in real-time. In addition, there are no datasets freely available for active research. Considering this, in this article, we propose an algorithm that uses simple convolutional neural networks (CNNs), which is an artificial intelligence (AI) technique, to detect major defects in layers during fused deposition modelling (FDM) in real-time. It analyses AM infill patterns to identify irregularities such as staircase, overfill, and void defects. The proposed model is trained using the dataset that has been collected manually and augmented at different scales for building a robust model. The results show that the proposed model is very effective and provides over 97.77% accuracy on real-time images. Furthermore, our proposed model uses fewer convolution layers than popular models, such as visual geometry group (VGG) 19, mobile neural network (MobileNet) V2, residual network (ResNet) 50, and densely connected convolutional network (DenseNet) 121. In addition, we open source the custom-generated datasets that contain staircase, overfill, and void defects images. For future research, we plan to expand dataset diversity and employ real-time adaptive learning.}, archive = {J_EAAI}, author = {Basil Kuriachen and Rathinaraja Jeyaraj and Deepak Raphael and P. Ashok and P. Shanmuga Sundari and Anand Paul}, doi = {10.1016/j.engappai.2024.109802}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109802}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Defect detection in fused deposition modelling using lightweight convolutional neural networks}, volume = {141}, year = {2025}, }
• (2025). Crime scene classification from skeletal trajectory analysis in surveillance settings. EAAI, 141, 109800. (www)

  @article{ , abstract = {Video anomaly analysis is a core task to the field of computer vision, particularly for crime detection in surveillance footage. In this work, we address the task of human-related crime classification using skeletal joint trajectories extracted from surveillance video frames. First, we emphasize the need to enhance the ground truth labels for the Human-Related Crime dataset (HR-Crime) and propose supervised and unsupervised methodologies to generate trajectory-level labels. Next, based on the trajectory-level labels, we introduce a trajectory-based crime classification framework, evaluating different architectures and feature fusion strategies for representing human trajectories. Our experiments validate the approach and open avenues for future research on this topic.}, archive = {J_EAAI}, author = {Alina-Daniela Matei and Estefanía Talavera and Maya Aghaei}, doi = {10.1016/j.engappai.2024.109800}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109800}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Crime scene classification from skeletal trajectory analysis in surveillance settings}, volume = {141}, year = {2025}, }
• (2025). High-performance remaining useful life prediction for aeroengine based on combining health states and trajectory similarity. EAAI, 141, 109799. (www)

  @article{ , abstract = {Aeroengine is a kind of highly complex and precise thermal machinery with various data features. Therefore, the selections of appropriate key features and effective utilization of data are the challenges and focal points in the prediction of remaining useful life (RUL) for the aeroengine. This paper proposes a high-precision method for predicting the RUL of aeroengines based on health states and trajectory similarity. Firstly, with a comprehensive understanding of the domain knowledge, the K-means clustering method is employed to categorize different health states of aeroengines and construct the aeroengine life database accordingly. This effectively reduces the false prediction caused by overlapping curves in the life model library. Secondly, by introducing a segmented similarity measurement method, the trajectory similarity of the Health Index (HI) curve between test data and life library can be better matched. Furthermore, incorporating a multiple weighted combination of L best-matched HI curves further improves the prediction accuracy. Finally, the validity of this method is verified by the simulation data set of turbofan aeroengines provided by National Aeronautics and Space Administration (NASA). Compared with other two similar algorithms, the accuracy increases by 4% and 6% respectively, in which the penalty Score of the proposed method decreases 20.82% and 69.17% respectively, and the lowest root mean square error (RMSE) is obtained.}, archive = {J_EAAI}, author = {Peng Peng and Yonghua Li and Zhongyi Guo}, doi = {10.1016/j.engappai.2024.109799}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109799}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {High-performance remaining useful life prediction for aeroengine based on combining health states and trajectory similarity}, volume = {141}, year = {2025}, }
• (2025). RepCrack: An efficient pavement crack segmentation method based on structural re-parameterization. EAAI, 141, 109791. (www)

  @article{ , abstract = {The detection of pavement cracks is a significant challenge in the field of road maintenance. In response to this challenge, the industry has witnessed a growing trend towards the utilization of deep learning-based methods for crack detection. However, existing approaches utilize complex network structures which, while improving accuracy, compromise detection efficiency and hinder practical application. This paper proposes a novel pixel-level pavement crack segmentation model, RepCrack, which addresses the current limitations in accuracy and efficiency in pavement crack detection tasks. The proposed model is based on large kernel convolution and multi-scale feature extraction, in order to extract more detailed pavement crack information during the training phase. The model building process incorporates structural re-parameterization techniques, including the conversion of multi-branch structures to single on-path structures and the shortening of long sequence structures, which effectively reduces the complexity of the model. Furthermore, to enhance the model's capacity to extract features, an improved coordinate attention mechanism is incorporated to augment the network's focus on the crack region. To validate the detection effect, publicly available datasets, including the Crack Forest Dataset (CFD), German Asphalt Pavement Distress (GAPs) dataset, and Crack Tree dataset, were employed. The experimental results demonstrate that our proposed method exhibits a notable superiority over existing crack detection techniques. For instance, RepCrack, with a parameter count of 5.8 million, attains an Intersection over Union (IoU) of 63.8% on the CFD dataset and attains a state-of-the-art methodology (SOTA) performance, exhibiting a 5% enhancement in accuracy and a 68.7% reduction in inference latency in comparison to SegNext.}, archive = {J_EAAI}, author = {Minglun Ni and Lei Chen and Peixin Shi and Ruiqi Ren}, doi = {10.1016/j.engappai.2024.109791}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109791}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {RepCrack: An efficient pavement crack segmentation method based on structural re-parameterization}, volume = {141}, year = {2025}, }
• (2025). Point cloud semantic segmentation network based on graph convolution and attention mechanism. EAAI, 141, 109790. (www)

  @article{ , abstract = {Point cloud data provides rich three-dimensional spatial information. Accurate three-dimensional point cloud semantic segmentation algorithms enhance environmental understanding and perception, with wide-ranging applications in autonomous driving and scene analysis. However, Graph Neural Networks often struggle to retain semantic relationships among neighboring points during feature extraction, potentially leading to the omission of critical features during aggregation. To address these challenges, we propose a novel network, the Feature-Enhanced Residual Attention Network. This network includes an innovative graph convolution module, the Neighborhood-Enhanced Convolutional Aggregation Module, which utilizes K-Nearest Neighbor and Dilated K-Nearest Neighbor techniques to construct diverse dynamic graphs and aggregate features, thereby prioritizing essential information. This approach significantly enhances the expressiveness and generalization capabilities of the network. Additionally, we introduce a new spatial attention module designed to capture semantic relationships among points. Experimental results demonstrate that the Feature-Enhanced Residual Attention Network outperforms benchmark models, achieving an average intersection ratio of 61.3% and an overall accuracy of 86.7% on the Stanford Large-Scale Three-dimensional Indoor Spaces dataset, thereby significantly improving semantic segmentation performance.}, archive = {J_EAAI}, author = {Nan Yang and Yong Wang and Lei Zhang and Bin Jiang}, doi = {10.1016/j.engappai.2024.109790}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109790}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Point cloud semantic segmentation network based on graph convolution and attention mechanism}, volume = {141}, year = {2025}, }
• (2025). A study on the automation of borehole image auto-recognition based on instance segmentation. EAAI, 141, 109789. (www)

  @article{ , abstract = {To mitigate the time consumption and inefficiency of manual borehole image data recognition, this paper introduces a instance segmetation-based algorithm model called borehole image auto-recognition (BIAR). The algorithm initially employs the instance segmentation module of the you only look once version 8(YOLOv8)deep learning model to automatically detect and classify structures like dike filling and fractures in borehole wall images. Polynomial fitting formulas are then applied to the segmented mask coordinates to generate structural curves of the detected features. The dip and dip direction of these structures are calculated from the extrema of the fitted curves. Finally, the algorithm outputs the processed borehole images. A comparison between automatic and manual recognition using a dataset of actual borehole images demonstrates the algorithm's ability to quickly and accurately identify fractures and discontinuities, even in the presence of complex borehole wall information. The model's accuracy and robustness are also validated. Validation indicates that the BIAR algorithm can identify and map structural surface features in borehole images at a depth of 100 m within 30 s, achieving over 90% accuracy in identifying structural surfaces.This algorithm provides a highly efficient and accurate method for automatic borehole image analysis in engineering applications.}, archive = {J_EAAI}, author = {Tong Jiang and Fanke Meng and Li Wan}, doi = {10.1016/j.engappai.2024.109789}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109789}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A study on the automation of borehole image auto-recognition based on instance segmentation}, volume = {141}, year = {2025}, }
• (2025). Graph dual-stream convolutional attention fusion for precipitation nowcasting. EAAI, 141, 109788. (www)

  @article{ , abstract = {Accurate precipitation nowcasting is crucial for applications such as flood prediction, disaster management, agriculture optimization, and transportation management. While many studies have approached this task using sequence-to-sequence models, most focus on single regions, ignoring correlations between disjoint areas. We reformulate precipitation nowcasting as a spatiotemporal graph sequence problem. Specifically, we propose Graph Dual-stream Convolutional Attention Fusion, a novel extension of the graph attention network. Our model’s dual-stream design employs distinct attention mechanisms for spatial and temporal interactions, capturing their unique dynamics. A gated fusion module integrates both streams, leveraging spatial and temporal information for improved predictive accuracy. Additionally, our framework enhances graph attention by directly processing three-dimensional tensors within graph nodes, removing the need for reshaping. This capability enables handling complex, high-dimensional data and exploiting higher-order correlations between data dimensions. Depthwise-separable convolutions are also incorporated to refine local feature extraction and efficiently manage high-dimensional inputs. We evaluate our model using seven years of precipitation data from Copernicus Climate Change Services, covering Europe and neighboring regions. Experimental results demonstrate superior performance of our approach compared to other models. Moreover, visualizations of seasonal spatial and temporal attention scores provide insights into the most significant connections between regions and time steps.}, archive = {J_EAAI}, author = {Lóránd Vatamány and Siamak Mehrkanoon}, doi = {10.1016/j.engappai.2024.109788}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109788}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph dual-stream convolutional attention fusion for precipitation nowcasting}, volume = {141}, year = {2025}, }
• (2025). Deep forest with SHapley additive explanations on detailed risky driving behavior data for freeway crash risk prediction. EAAI, 141, 109787. (www)

  @article{ , abstract = {Freeway crash risk prediction is a critical component of traffic safety management, yet existing crash risk models often fail to capture complex driving behaviors and lack interpretability. This study introduces a novel freeway crash risk prediction framework based on the Deep Forest (DF) algorithm, considering the detailed risky driving behavior data. The DF model integrates multi-grained scanning and cascade forest layers, enabling it to capture the complex relationship between risky driving behavior features. SHapley Additive Explanations (SHAP) are applied to interpret the model's predictions, including both SHAP summary and interaction results. Additionally, ablation studies are conducted to evaluate the contributions of key components like multi-grained scanning and cascade structures to the model's performance. The experimental results demonstrate that the DF model outperforms traditional machine learning models. The DF model achieves an area under the receiver operating characteristic curve of 0.825, with a balanced Sensitivity of 0.75 and Specificity of 0.816, surpassing other models. The ablation studies show that removing multi-grained scanning, cascade layers, or completely random tree forest leads to performance declines, confirming the importance of each component. The SHAP analysis highlights that sharp acceleration and braking behaviors have the most significant impact on crash risk, offering clear, interpretable insights into how driving behaviors contribute to risk. Overall, the DF model's superior performance and SHAP-based interpretability provide a powerful tool for traffic safety management. These findings emphasize the value of incorporating both driving behavior intensity and model interpretability into crash risk prediction, offering practical applications for reducing crash rates.}, archive = {J_EAAI}, author = {Xiaochi Ma and Zongxin Huo and Jian Lu and Yiik Diew Wong}, doi = {10.1016/j.engappai.2024.109787}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109787}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep forest with SHapley additive explanations on detailed risky driving behavior data for freeway crash risk prediction}, volume = {141}, year = {2025}, }
• (2025). Multi-scale latent space sequential fusion of images for pose estimation of underground tunneling machinery. EAAI, 141, 109786. (www)

  @article{ , abstract = {Recently, the underground unmanned tunneling technology has received increasing attention, in which the pose estimation of the tunneling machinery occupies a prominent position. However, the autonomous positioning of the tunneling machinery presents significant challenges due to the harsh underground environments and the extremely complicated working conditions, which severely restricts the development and application of underground intelligent tunneling technology. Therefore, this paper investigates a machine vision and deep learning based pose estimation method for the underground tunneling machinery. Considering that both effective features and interference features of time-series images exhibit spatiotemporal correlations across multiple scales, we propose a multi-scale latent space sequential fusion (MSLSSF) model to integrate information from time-sequence images and further estimate the pose variables of the tunneling machine. The multi-scale variational autoencoder (MSVAE) is employed in the proposed model to obtain multi-scale latent space representations for each image in the time sequence. The proposed MSLSSF strategy consists of two steps. The first step involves fusing the latent representations of the time-series images at each scale using long short-term memory (LSTM) models, as specific correlations may exist in the representations of the sequential images across various scales. The second step employs an attention mechanism to adaptively fuse the results from the first step across all scales and timesteps. The presented information fusion methods provide a way for the model to take full advantage of the multi-scale features of time-sequence images. Experiments are conducted on our custom-made tunneling machine visual localization dataset, where the proposed MSLSSF based pose estimation method outperforms the advanced comparison methods in both accuracy and robustness, thereby validating the efficacy of the proposed approach. Additionally, the feasibility of the MSLSSF strategy is confirmed through the visualization results of the proposed model.}, archive = {J_EAAI}, author = {Hongzhuang Wu and Cheng Cheng and Deyi Zhang and Hongzhi Zhou}, doi = {10.1016/j.engappai.2024.109786}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109786}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale latent space sequential fusion of images for pose estimation of underground tunneling machinery}, volume = {141}, year = {2025}, }
• (2025). A dynamic graph convolutional network-based framework for the unsteady operating states recognition of multi-product pipeline systems. EAAI, 141, 109785. (www)

  @article{ , abstract = {Considering that the existing methods lack spatial and temporal information mining of pipeline multidimensional operation data, it is unable to accurately recognize the unsteady operation conditions among pipeline stations. In this study, a dynamic graph convolutional network classification model is proposed for the recognition of unsteady operating states in multi-product pipeline systems. Firstly, dynamic graph convolutional network of multi-pipeline system (DPipeNet) is constructed based on the visibility graph algorithm, mutual information and long and short-term memory network model. Secondly, static graph convolutional network of multi-pipeline system (SPipeNet) is constructed by using the real geographic location information of each station of multi-pipeline. Then, the input subgraph of the graph convolutional network is used to construct the multi-pipeline system operational state relationship network (OSRN), and the vulnerable state nodes of the system are evaluated using complex network centrality metrics. Finally, the proposed model is applied to real operational data of a multi-pipeline system in China. The results show that in the two-classification scenario, both DPipeNet and SPipeNet have higher accuracies, but DPipeNet has a lower missed rate. In the multi-classification scenario, DPipeNet has the highest precision, which can reach more than 85%, and the recall rate is improved by 13%–25% compared with the neural network models in recent literature and SPipeNet. In the vulnerability analysis scenario, the intermediate station pump startup/stoppage of multi-pipeline has higher vulnerability. The proposed method also provides decision support for managers in pipeline system operation and maintenance management.}, archive = {J_EAAI}, author = {Li Zhang and Lin Fan and Jianjun Liu and Dingyu Jiao and Yuxuan He and Jing Zhou and Karine Zeitouni and Huai Su and Jinjun Zhang}, doi = {10.1016/j.engappai.2024.109785}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109785}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dynamic graph convolutional network-based framework for the unsteady operating states recognition of multi-product pipeline systems}, volume = {141}, year = {2025}, }
• (2025). An ensemble based lightweight deep learning model for the prediction of cardiovascular diseases from electrocardiogram images. EAAI, 141, 109782. (www)

  @article{ , abstract = {Cardiovascular Disease (CVD) has become a serious reason of death all over the world. According to the Australian Institute of Health and Welfare report, CVD was the underlying cause of 42,700 deaths (25% of all) in 2021. Therefore, accurate detection of CVD is crucial for early treatment. An Electrocardiogram (ECG) signal is used to measure the rhythms and electrical activity of the heart. Traditionally, paper-based ECG images are used to detect CVD which may not be accurate always. The existing pre-trained models such as transfer learning can be used for automatic diagnosis. However, they fall short of multiclass classification due to complex variations of the ECG signals. For this research, four cardiac abnormalities including Abnormal Heartbeats, Myocardial Infarction, History of Myocardial Infarction, and Normal Heartbeats are predicted using the combination of machine learning and deep learning techniques. The deep learning method is used for feature extraction and the machine learning algorithm is used for classification. A lightweight Convolutional Neural Network (CNN) model is designed to extract the prominent features from ECG images and use these features for the prediction of CVDs. However, only a classifier may not fully utilize extracted features for classification. Therefore, the proposed method developed an optimized weighted average ensemble model for classifier selection. The proposed ensemble-based lightweight model showed its ability to extract informative features when tested on real ECG datasets and outperformed the baseline methods by achieving the highest accuracy of 99.29%.}, archive = {J_EAAI}, author = {Md Nahid Hasan and Md Ali Hossain and Md Anisur Rahman}, doi = {10.1016/j.engappai.2024.109782}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109782}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An ensemble based lightweight deep learning model for the prediction of cardiovascular diseases from electrocardiogram images}, volume = {141}, year = {2025}, }
• (2025). A novel maximum power point tracking approach based on fuzzy logic control and optimizable gaussian process regression for solar systems under complex environment conditions. EAAI, 141, 109780. (www)

  @article{ , abstract = {Photovoltaic (PV) systems have multiple peaks in the power–voltage (P–V) curve due to partial shading conditions (PSC). Multiple peaks make determining and tracking the global maximum power point (GMPP) more complex. More powerful algorithms and controller structures are needed to track GMPP in PV systems operating in complex environmental conditions.Therefore, this paper introduces a machine learning based fuzzy logic controller (MLBFLC) method to determine and track GMPP. MLBFLC is proposed to determine the optimal duty cycle of the DC–DC converter used in PV systems operating under PSCs. In order to test this method, real-time temperature and irradiance data for one month (February, May, August and November) from different seasonal conditions were used. The reference voltage values at the maximum power point (MPP) were obtained from the hyperparameter optimized Gaussian Process Regression (GPR) method. The Fuzzy Logic Controller (FLC) method was used to determine the optimum duty cycle of the converter. The proposed method was compared with the metaheuristic optimization algorithms such as particle swarm optimization (PSO) and the flying squirrel search optimization (FSSO) algorithm, for four different scenarios, using real-time temperature and irradiance data. Consequently, it is observed that the proposed MLBFLC method successfully tracks the GMPP with higher speed and higher accuracy for all scenarios. Under different PSCs determined in the scenarios, an efficiency value of 99.916% was achieved with the MLBFLC method and it was observed that it successfully followed the MPP with a tracking time of 0.123 s.}, archive = {J_EAAI}, author = {Mehmet Yılmaz and Muhammed Reşit Çorapsız and Muhammed Fatih Çorapsız}, doi = {10.1016/j.engappai.2024.109780}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109780}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel maximum power point tracking approach based on fuzzy logic control and optimizable gaussian process regression for solar systems under complex environment conditions}, volume = {141}, year = {2025}, }
• (2025). Blast furnace control by hierarchical similarity search in historical data. EAAI, 141, 109779. (www)

  @article{ , abstract = {In iron and steel industries, the temperature of the hot metal in blast furnaces impacts the quality of products as well as the safety and efficiency of the production process. The control of such equipment is challenging due to the large number of variables associated with the temperature and the complex interactions among them. Therefore, this work proposes a novel blast furnace control system for controlling its temperature that is based on hierarchical nearest neighbor search in historical data collected from human operators. This approach generates controls with similar patterns to those realized by the operators. The system first searches for a past control sequence that (i) led to the desired temperature variation, (ii) starts from a furnace state similar to the current one and (iii) is consistent with actions performed recently. Then, a new control plan is produced based on the chosen sequence. To evaluate the system in a safe environment, classical machine learning regression models (Linear Regression, Partial Least Squares, Random Forest, Extreme Gradient Boosting and Light Gradient-Boosting Machine) were built to simulate the blast furnace. Such evaluation in a learning-based simulated environment is only possible since the proposed control system does not rely on a blast furnace model. Experimental results show that our system can effectively control the temperature for different target values, initial furnace states and several of the evaluated blast furnace models. Experiments also show that the control system can perform multi-step long-term planning using the evaluated models.}, archive = {J_EAAI}, author = {Lucas L. Amorim and Filipe Mutz and Thiago M. Paixão and Vinicius Rampinelli and Alberto F. De Souza and Claudine Badue and Thiago Oliveira-Santos}, doi = {10.1016/j.engappai.2024.109779}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109779}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Blast furnace control by hierarchical similarity search in historical data}, volume = {141}, year = {2025}, }
• (2025). Variational learning to rank for test case prioritization via prioritizing metric inspired differentiable loss. EAAI, 141, 109776. (www)

  @article{ , abstract = {The Test Case Prioritization (TCP) plays a vital role in software testing by strategically ordering test cases to expose faults at the earliest feasible stages. Recent TCP studies have embraced learning-to-rank techniques to tackle the challenges of complexity and scalability in test ranking. However, the measurements of TCP, such as the Average Percentage of Fault Detection (APFD), are non-differentiable and incompatible with gradient-based optimizations. Moreover, learning models are susceptible to overfitting due to their complex natures of embedding capabilities. The aim of this study was to investigate differentiable APFD-inspired objectives and examine the consistency in priority embedding space, to jointly improve the model generalization. Therefore, we proposed a novel variational learning based TCP model, named APFD-Net. The model was constructed upon a simplified Set-Transformer and optimized by combining three losses, namely, the variational likelihood loss weighted by APFD difference, the differentiable soft APFD loss and the prior loss, thus making APFD applicable to direct optimizations. Industry datasets were used to verify the validity of our approach. Experiments results demonstrate that the APFD-Net achieves best effectiveness and relatively improved at least 4.8% in APFD compared to state-of-art methods, indicating its potential for industrial applications.}, archive = {J_EAAI}, author = {Peng Tang and Junfeng Wang and Mingxing Liu}, doi = {10.1016/j.engappai.2024.109776}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109776}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Variational learning to rank for test case prioritization via prioritizing metric inspired differentiable loss}, volume = {141}, year = {2025}, }
• (2025). A bilevel coevolution framework with knowledge transfer for large-scale optimization and its application in multiperiod economic dispatch. EAAI, 141, 109775. (www)

  @article{ , abstract = {Complex systems typically consist of multiple components and serve requirements across multiple periods. Their optimization involves large-scale parameters. If all parameters are considered at one time, the high-dimensional searching space will present great challenge. Otherwise, if parameters are considered partially, the global fittest solutions can hardly be found. Keep these in mind, in this paper, a novel bilevel coevolution framework with knowledge transfer (BiKT) is introduced for large-scale multiobjective optimization. Specifically, in this framework, the optimization problem is decomposed to several low-dimensional subproblems, establishing a bilevel structure. Then, the original problem and the subproblems are regarded as two tasks, their coevolution is realized by converting searching agents between the upper and lower optimization workflows, fulfilling exploration in global optimization and exploitation in local areas. Meanwhile, a knowledge transfer strategy is studied to adapt the search directions and accelerate convergence speeds. The superiority of the novel framework has been verified by experimental studies on large-scale benchmark problems, and the effectiveness of knowledge transfer has been discussed through ablation experiments. In the end, the proposed method is employed to tackle a large-scale real-world challenge, the multiperiod economic dispatch problem in the power system. After problem formulation and analysis, the proposed method can perfectly solve the application.}, archive = {J_EAAI}, author = {Anqi Pan and Haifeng Liu and Yinghao Shan and Bo Shen}, doi = {10.1016/j.engappai.2024.109775}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109775}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A bilevel coevolution framework with knowledge transfer for large-scale optimization and its application in multiperiod economic dispatch}, volume = {141}, year = {2025}, }
• (2025). Risk identification of listed companies violation by integrating knowledge graph and multi-source risk factors. EAAI, 141, 109774. (www)

  @article{ , abstract = {The regulatory compliance supervision of listed enterprises is of great significance for ensuring the stable operation of financial markets. However, existing graph propagation algorithms used to identify corporate violations are limited by their inherent randomness. At the same time, current methods consider a relatively narrow range of risk dimensions, making it difficult to accurately distinguish the different characteristics of enterprises. To this end, this paper designs a Propagation of Violation Risks for Listed Enterprises (PVR-LE) algorithm, which reduces the risk of false propagation through a risk weight decay mechanism and dynamic updating of risk propagation patterns. Furthermore, a Multi-source Risk Fusion Neural Network (MRFNN) for corporate violation prediction tasks is proposed, which fuses the liquidity risk characteristics between enterprises, clustering characteristics, and the enterprise’s risk characteristics to endow the violating enterprise nodes with more distinctive and characteristic vector features, thereby identifying whether there are violations in the company. At the same time, a generative adversarial network is used to generate samples of violating enterprises to solve the problem of class imbalance. Experiments are conducted on a real dataset constructed from information on Chinese listed companies, and this method improves the accuracy, recall, the weighted harmonic mean of precision and recall(F1-score), and geometric mean(G-mean) metrics by 2.12%, 3.19%, 2.14%, and 2.79%, respectively, compared to the best performance of existing models. The experimental results show that the proposed method effectively improves the accuracy of identifying corporate violations and helps promote the development of intelligent regulatory work for listed enterprises.}, archive = {J_EAAI}, author = {Jinlong Wang and Pengjun Li and Yingmin Liu and Xiaoyun Xiong and Yuanyuan Zhang and Zhihan Lv}, doi = {10.1016/j.engappai.2024.109774}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109774}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Risk identification of listed companies violation by integrating knowledge graph and multi-source risk factors}, volume = {141}, year = {2025}, }
• (2025). PerSceptoMed 1.0: An advanced machine learning model for demographic-based antibiotic susceptibility prediction in pseudomonas aeruginosa infections. EAAI, 141, 109773. (www)

  @article{ , abstract = {Background Antimicrobial resistance caused by Pseudomonas aeruginosa is a pressing global health issue impacting both hospitals and communities. Methodology We analyzed demographic and susceptibility data from 1087 samples, and tested them against five different antibiotics at the Institute of Medical Sciences & SUM Hospital. Eight machine learning algorithms (linear: multiple linear regression, generalized linear model-elastic net regularization, support vector machine-linear kernel, linear discriminant analysis; non-linear: support vector machine-radial basis function kernel, decision tree, k-nearest neighbors, kernel k-nearest neighbors), were evaluated to develop the PerSceptoMed 1.0 model for susceptibility prediction. Performance metrics such as accuracy, Matthews Correlation Coefficient, Hamming loss, 0/1 loss, and Receiver Operating Characteristic curves were computed. Statistical analysis was conducted to compare model performance. Results Gender disparities were observed, with male patients showing higher infection rates, potentially due to occupational exposure, microbiome differences, comorbidities, and healthcare-seeking behaviors. age, clinical status, and antibiotic susceptibility were significant factors, complicating treatment strategies. Infections extended to outpatient departments, indicating a shift from hospital to community settings. Kernel k-nearest neighbors exhibited strong predictive capabilities, improving accuracy from 84.39% pre-balancing to 91.01% post-balancing, and an Area Under the Curve of 0.92. Statistical tests showed significant performance variations among models. Conclusion The study highlights the importance of gender-sensitive approaches in infection management and the superior performance of nonlinear models in decoding complex data patterns. PerSceptoMed 1.0 demonstrates promise in enhancing antimicrobial resistance management by providing actionable insights based on demographic data.}, archive = {J_EAAI}, author = {Sweta Padma Routray and Saswati Mahapatra and Debasish Swapnesh Kumar Nayak and Swayamprabha Sahoo and Bhabani Shankar Das and Enketeswara Subudhi and Rajashree Panigrahy and Tripti Swarnkar}, doi = {10.1016/j.engappai.2024.109773}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109773}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {PerSceptoMed 1.0: An advanced machine learning model for demographic-based antibiotic susceptibility prediction in pseudomonas aeruginosa infections}, volume = {141}, year = {2025}, }
• (2025). Developing guidelines for functionally-grounded evaluation of explainable artificial intelligence using tabular data. EAAI, 141, 109772. (www)

  @article{ , abstract = {Explainable Artificial Intelligence (XAI) techniques are used to provide transparency to complex, opaque predictive models. However, these techniques are often designed for image and text data, and it is unclear how fit-for-purpose they are when applied to tabular data. As XAI techniques are rarely evaluated in the context of tabular data, the applicability of existing evaluation criteria and methods are also unclear and needs re-examination. For example, some works suggest that evaluation methods may unduly influence the evaluation results when using tabular data. This lack of clarity on evaluation procedures can lead to reduced transparency and ineffective use of XAI techniques in real world settings. In this study, we examine literature on XAI evaluation to derive guidelines on functionally-grounded assessment of local, post hoc XAI techniques. We identify 20 evaluation criteria and associated evaluation methods, and derive guidelines on when and how each criterion should be evaluated. We also identify key research gaps to be addressed by future work. Our study contributes to the body of knowledge on XAI evaluation through in-depth examination of functionally-grounded XAI evaluation protocols, and has laid the groundwork for future research on XAI evaluation.}, archive = {J_EAAI}, author = {Mythreyi Velmurugan and Chun Ouyang and Yue Xu and Renuka Sindhgatta and Bemali Wickramanayake and Catarina Moreira}, doi = {10.1016/j.engappai.2024.109772}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109772}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Developing guidelines for functionally-grounded evaluation of explainable artificial intelligence using tabular data}, volume = {141}, year = {2025}, }
• (2025). Integrating belief-desire-intention agents with large language models for reliable human–robot interaction and explainable artificial intelligence. EAAI, 141, 109771. (www)

  @article{ , abstract = {This paper presents the development of an innovative communication interface between humans and robots, designed for human-in-the-loop interaction with a user interface through natural language. The novelty of the presented approach lies in integration of a Belief-Desire-Intention agent, communicating directly to a robot and ensuring safety properties and verifiable decision making, with large language models that excel in language understanding and generation. We establish a framework that leverages the strengths of both paradigms by allowing users to formulate commands in natural language, using the ability of large language models to interpret and ground them into actionable goals, with the proven ability of Belief-Desire Intention agents to perform verifiable reasoning and goal management. In addition, we utilize the ability of this agent to represent and store its mind state, including its belief base, plan library, and history of selected events and actions, to allow large language models based explanations of the system behavior. Our findings demonstrate that this architecture provides several benefits in terms of performance and safety, without impacting efficiency or requiring extensive integration effort. This research contributes a novel perspective on the combination of large language models with symbolic approaches for explainable Artificial Intelligence and aims at inspiring new developments in human-centered Artificial Intelligence systems.}, archive = {J_EAAI}, author = {Laurent Frering and Gerald Steinbauer-Wagner and Andreas Holzinger}, doi = {10.1016/j.engappai.2024.109771}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109771}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrating belief-desire-intention agents with large language models for reliable human–robot interaction and explainable artificial intelligence}, volume = {141}, year = {2025}, }
• (2025). Combining normalizing flows with decision trees for interpretable unsupervised outlier detection. EAAI, 141, 109770. (www)

  @article{ , abstract = {Outlier detection is critical for ensuring data integrity across various domains, from fraud detection in finance to anomaly identification in healthcare. Despite the importance of anomaly detection, most methods focus on performance, with interpretability remaining underexplored in unsupervised learning. Interpretability is essential in contexts where understanding why certain data points are classified as outliers is as important as the detection itself. This study introduces an interpretable approach to unsupervised outlier detection by combining normalizing flows and decision trees. Normalizing flows transform complex data distributions into simpler, tractable forms, allowing precise density estimation and the generation of pseudo-labels that differentiate inliers from outliers. These pseudo-labels are subsequently used to train a decision tree, offering both a structured decision-making process and interpretability in an unsupervised context, thereby addressing a key gap in the field. Our method was evaluated against 23 established outlier detection algorithms across 17 datasets using Precision, Recall, F1 Score, and Matthews Correlation Coefficient (MCC). The results showed that our approach ranked 4th in F1 Score, 6th in MCC, 3rd in Precision, and 19th in Recall. While it performed strongly on some datasets and less so on others, this variability is likely due to dataset-specific characteristics. Post-hoc statistical significance testing demonstrated that interpretability in unsupervised outlier detection can be achieved without significantly compromising performance, making it a valuable option for applications that require transparent and understandable anomaly detection.}, archive = {J_EAAI}, author = {Vasilis Papastefanopoulos and Pantelis Linardatos and Sotiris Kotsiantis}, doi = {10.1016/j.engappai.2024.109770}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109770}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Combining normalizing flows with decision trees for interpretable unsupervised outlier detection}, volume = {141}, year = {2025}, }
• (2025). Ship fuel consumption prediction based on transfer learning: Models and applications. EAAI, 141, 109769. (www)

  @article{ , abstract = {Data-driven fuel consumption rate (FCR) prediction models largely depend on the amount of training data, which can be scarce for new ships with limited operating time. To tackle this issue, we implement three transfer learning strategies to leverage knowledge from another seven container ships to construct artificial neural network (ANN)-based FCR prediction models for a target ship with limited data. Numerical experiments reveal that the ANN models incorporating the three transfer strategies outperform the model trained solely on the target ship data, reducing mean absolute percentage error by 12.57%, 6.44%, and 16.03%, respectively. This study also investigates the impacts of target dataset size on the performance of transfer strategies using ship FCR prediction as an example, revealing that the smaller amount of available data, the greater improvement in prediction accuracy using the transfer strategy. These insights contribute to the development of effective operational solutions for enhancing ship energy efficiency and promoting sustainable shipping practices.}, archive = {J_EAAI}, author = {Xi Luo and Mingyang Zhang and Yi Han and Ran Yan and Shuaian Wang}, doi = {10.1016/j.engappai.2024.109769}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109769}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ship fuel consumption prediction based on transfer learning: Models and applications}, volume = {141}, year = {2025}, }
• (2025). A novel approach to model-based diagnosis with multiple observations. EAAI, 141, 109768. (www)

  @article{ , abstract = {Model-based diagnosis (MBD) with multiple abnormal observations utilizes inconsistencies between actual and expected observations of systems to localize system faults. Current state-of-the-art algorithms still require solvers to consider a substantial number of suspected faulty components. To address this challenge, we introduce the Dual Principles with Decision Node (DPDN) algorithm. The first part of DPDN consists of two innovative principles: the Output Dependency Judgment Principle (ODJP) and the Deep Propagation Dependency Principle (DPDP). These principles are designed to identify and categorize a larger portion of components as ‘normal’, thereby broadening the idea of filtered nodes. By increasing the number of components classified as ‘normal’, the diagnostic process becomes more efficient as fewer components need to be diagnosed. The second part of DPDN integrates a newly defined decision node guided by its Decision Node Principle (DNP). This decision node, along with its corresponding principle, further bolsters the diagnostic process by classifying additional components as normal. With DPDN, complex real-world systems can reduce the number of components considered during the diagnostic process by eliminating those that are functioning normally, thereby decreasing the time required to obtain diagnoses. We conduct comparative experiments to evaluate the efficiency of our algorithm against other prevalent methods. Empirical results distinctly underscore DPDN’s superior performance in relation to other state-of-the-art algorithms.}, archive = {J_EAAI}, author = {Ran Tai and Dantong Ouyang and Weiting Liu and Luyu Jiang and Liming Zhang}, doi = {10.1016/j.engappai.2024.109768}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109768}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel approach to model-based diagnosis with multiple observations}, volume = {141}, year = {2025}, }
• (2025). XNILMBoost: Explainability-informed load disaggregation training enhancement using attribution priors. EAAI, 141, 109766. (www)

  @article{ , abstract = {In the ongoing energy transition, characterized by increased reliance on distributed renewable sources and smart grid technologies, the need for advanced and trustworthy artificial intelligence (AI) in energy management systems is crucial. Non-intrusive load monitoring (NILM), a method for inferring individual appliance energy consumption from aggregate smart meter data, has gained prominence for enhancing energy efficiency. However, advanced deep neural network models used in NILM, while effective, raise transparency and trust concerns due to their complexity. This paper introduces a novel explainability-informed NILM training framework, specifically designed for low-frequency NILM. Our approach aligns with principles for trustworthy AI, focusing on human agency and oversight, technical robustness, and transparency, incorporating explainability directly into the training phase of a NILM model. We propose a novel iterative, explainability-informed NILM training algorithm that uses attribution priors to guide model optimization, including implementation and evaluation of the framework across multiple state-of-the-art NILM architectures, namely, convolutional, recurrent, and dilated causal layers. We introduce a novel Robustness-Trust metric to measure joint improvement in predictive and explainability performance, utilizing explainability metrics of faithfulness, robustness and effective complexity while analyzing model predictive performance against NILM-specific regression and classification metrics. Results broadly show that robust models achieve better explainability, while explainability-enhanced models can lead to improved model robustness. Together, our results demonstrate significant improvements in robustness and transparency of NILM systems across various appliances, model architectures, measurement scales, types of buildings, and energy usage patterns. This work paves the way for more transparent and trustworthy deployments in AI-driven energy systems.}, archive = {J_EAAI}, author = {Djordje Batic and Vladimir Stankovic and Lina Stankovic}, doi = {10.1016/j.engappai.2024.109766}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109766}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {XNILMBoost: Explainability-informed load disaggregation training enhancement using attribution priors}, volume = {141}, year = {2025}, }
• (2025). Long-term self-supervised learning for accelerometer-based sleep–wake recognition. EAAI, 141, 109758. (www)

  @article{ , abstract = {Sleep is a crucial health metric linked to various health problems. Accurate analysis of sleep duration relies on identifying sleep and wake phases. While Polysomnography is considered the gold standard for sleep measurements, it is impractical for large population-based studies. Accelerometers offer an alternative but face challenges in accurate sleep–wake recognition (SWR). Recent advances in self-supervised learning (SSL) have shown promise in related fields. This study introduces a contribution to the field of artificial intelligence by proposing a new SSL model, long-term accelerometer to vector (LTA2V), for SWR. LTA2V learns context information during pre-training using global positional encoding and long-term sequences. Experiments on three datasets show that LTA2V outperforms other methods, achieving an average F1-score of 73.8%, outperforming the best supervised method by 3.4% and the best alternative SSL model by 1.7%. Analysis of Variance and Tukey’s Honest Significant Difference tests show that SSL methods significantly outperform rule-based and traditional machine learning methods and are on par with supervised deep learning. Despite improvements, SWR performance enhancements through SSL are limited compared to other fields, suggesting that there is limited scope for further major improvements in accelerometer-based SWR. Applying LTA2V in population-based studies can lead to more reliable data on sleep duration across populations, improving our understanding of how sleep impacts health.}, archive = {J_EAAI}, author = {Aleksej Logacjov and Kerstin Bach and Paul Jarle Mork}, doi = {10.1016/j.engappai.2024.109758}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109758}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Long-term self-supervised learning for accelerometer-based sleep–wake recognition}, volume = {141}, year = {2025}, }
• (2025). Reinforcement learning adaptive network for nonaffine discrete-time control systems: Managing implicit zero-gain. EAAI, 141, 109757. (www)

  @article{ , abstract = {This paper investigates the phenomenon of implicit zero control gain in nonaffine discrete-time systems caused by dead-zone regions, utilizing an experimental setup focused on motor torque control. It challenges the conventional assumptions in adaptive controller designs, which typically require a nonzero control gain. An adaptive controller is subsequently developed using an actor–critic architecture integrated with fuzzy-rule networks and a customized surfacing reward function dealing both tracking error and control effort. The proposed controller achieves near-optimal tracking performance, improves control energy efficiency, and exhibits robustness against unknown disturbances. Learning laws are derived to optimize the controller’s performance without imposing a minimum control gain requirement. The controller’s effectiveness in addressing the dead-zone problem is validated through experimental and comparative results, especially under disturbance amplitudes of approximately 30% of the desired output trajectory, including high-frequency components.}, archive = {J_EAAI}, author = {C. Treesatayapun}, doi = {10.1016/j.engappai.2024.109757}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109757}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reinforcement learning adaptive network for nonaffine discrete-time control systems: Managing implicit zero-gain}, volume = {141}, year = {2025}, }
• (2025). Data-driven spiking neural networks for intelligent fault detection in vehicle lithium-ion battery systems. EAAI, 141, 109756. (www)

  @article{ , abstract = {Electric vehicles (EVs) powered by high-energy batteries are anticipated to be a primary avenue for achieving energy decarbonization in future societies. However, the high energy density of lithium batteries poses significant safety risks under complex conditions and sudden environmental changes. Thus, safety and fault issues of high-energy batteries during vehicle operation have gained much attention. This study proposes an intelligent fault diagnosis algorithm based on spiking neural networks (SNNs) using a data-driven approach. Time series data are stacked and fed into the SNN for learning, capturing temporal characteristics and forming a stable kernel representation. The predicted output is compared with the actual output to generate a residual signal for fault diagnosis. The algorithm is validated on a battery management experimental platform with injected internal and external short circuit faults. Results show the algorithm effectively and swiftly detects abnormal signals and has some transferability.}, archive = {J_EAAI}, author = {Penghao Wu and Engang Tian and Hongfeng Tao and Yiyang Chen}, doi = {10.1016/j.engappai.2024.109756}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109756}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven spiking neural networks for intelligent fault detection in vehicle lithium-ion battery systems}, volume = {141}, year = {2025}, }
• (2025). Input-optimized physics-informed neural networks for wave propagation problems in laminated structures. EAAI, 141, 109755. (www)

  @article{ , abstract = {Accurate prediction of wave propagation characteristics in laminated structures is crucial for engineering applications, such as ultrasonic examination and composite material optimization. This study develops a novel framework based on extended physics-informed neural networks (XPINNs) to enable the analysis and prediction of wave propagation in laminated structures. The XPINNs with domain decomposition are extended to model multilayered laminated structures, with each sub-PINN governing an individual component layer. An innovative hybrid-handed coordinate system is proposed to address compatibility conditions among layers and unify all inputs of sub-PINNs, developing an input-optimized framework named unified-input XPINNs (Uni-XPINNs). Three types of errors, including the relative root mean square error, the relative absolute error, and the mean square error, are utilized to assess the performance of the proposed framework against analytical solutions and finite element simulations. Based on the obtained prediction models, the study investigates the impact of body forces on wave propagations in laminated structures. Numerical results demonstrate high accuracy in predicting wave propagation in laminate structures, with a maximum error of 3.063% for all cases discussed. It is observed that body forces significantly affect wave propagation when they exceed a specific threshold. Below this threshold, their impacts are minimal. This research explores the application of machine learning methods to solve wave propagation problems, offering alternatives to traditional theoretical methods and numerical simulations.}, archive = {J_EAAI}, author = {Liangteng Guo and Shaoyu Zhao and Jie Yang and Sritawat Kitipornchai}, doi = {10.1016/j.engappai.2024.109755}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109755}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Input-optimized physics-informed neural networks for wave propagation problems in laminated structures}, volume = {141}, year = {2025}, }
• (2025). Deep-learning-empowered visual ship detection and tracking: Literature review and future direction. EAAI, 141, 109754. (www)

  @article{ , abstract = {Visual object detection and tracking is a fundamental task in artificial intelligence and plays a crucial role in various maritime applications, leading to enhanced maritime traffic safety and security. Although significant progress has been made in object detection and tracking in recent years, it still remains a challenge due to the image quality decline and object appearance changes caused by complex maritime traffic environments. This article aims to provide a comprehensive review of the current state of research in the field. We have examined the latest advancements in various aspects and proposed some directions for future research. Firstly, we delve into four main challenges in ship detection and tracking, including environmental variations, ship scale, ship occlusion, and lightweight models. Subsequently, we review the methods employed to address these challenges, and key contents about ship detection and tracking, include datasets, evaluation metrics, experimental results from mainstream methods. Finally, we offer a discussion and some future research directions that warrant attention, which will become potential research effort in the future.}, archive = {J_EAAI}, author = {Boxing Zhang and Jingxian Liu and Ryan Wen Liu and Yanhong Huang}, doi = {10.1016/j.engappai.2024.109754}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109754}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep-learning-empowered visual ship detection and tracking: Literature review and future direction}, volume = {141}, year = {2025}, }
• (2025). Zero-faulty sample machinery fault detection via relation network with out-of-distribution data augmentation. EAAI, 141, 109753. (www)

  @article{ , abstract = {Zero-faulty sample machinery fault detection is extremely challenging. However, it is common issue that fault samples are extremely difficult to obtain, which limits the effectiveness of traditional data-driven machinery fault detection methods in practical application. To address the mentioned limitations, the paper proposes a relation network with out-of-distribution data augmentation for zero-faulty sample machinery fault detection. Specifically, the out-of-distribution data generator is referenced to generate out-of-distribution (OOD) samples for the normal data, assisting in mining of intrinsic state information from normal samples. Then, the feature splicing mechanism is devised to create feature pairs between normal state samples and OOD samples. The relation network with similarity evaluation capability is built by learning in the feature pairs. Finally, the relation network determines the type of fault state by evaluating the similarity between it and the normal state. The effectiveness of the proposed method is substantiated through experiments conducted on propulsion shaft system, industrial robot, and wind turbine gearbox. The results demonstrate that the proposed method can accurately detect and locate machinery fault positions under both single and compound fault modes, with 5% greater accuracy and strong generalizability than current approaches.}, archive = {J_EAAI}, author = {Zuoyi Chen and Hong-Zhong Huang and Jun Wu and Yuanhang Wang}, doi = {10.1016/j.engappai.2024.109753}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109753}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Zero-faulty sample machinery fault detection via relation network with out-of-distribution data augmentation}, volume = {141}, year = {2025}, }
• (2025). Special issue on: “Emerging industrial digitalisation and machine learning applications in maintenance engineering, asset, and operations management”. EAAI, 141, 109752. (www)

  @article{ , archive = {J_EAAI}, author = {Christos Emmanouilidis and Ajith Parlikad and Giacomo Barbieri and David Romero}, doi = {10.1016/j.engappai.2024.109752}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109752}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Special issue on: “Emerging industrial digitalisation and machine learning applications in maintenance engineering, asset, and operations management”}, volume = {141}, year = {2025}, }
• (2025). Automatic speech recognition for moroccan dialect in noisy traffic environments. EAAI, 141, 109751. (www)

  @article{ , abstract = {In this study, we explored the impact of traffic noise on speech recognition performance for the Moroccan Fessi dialect at varying noise levels. To achieve this, we developed a highly configurable, open-source speech recognition platform, designed to identify optimal parameters for handling the dialect’s unique characteristics. Our approach employs a stochastic hidden Markov model combined with Gaussian mixture models, specifically tailored for the low-resource Moroccan dialect. The novelty of this approach lies in its ability to enable parameter learning and adaptation with minimal prior knowledge of the dialect’s linguistic rules. We further analyzed the noise performance of the system in relation to the syllabic structure of words, assessing its effect on automatic speech recognition accuracy. Experimental results show a substantial degradation in recognition performance under noisy conditions, with a word error rate of 96.95% in noisy environments compared to 8.14% in clean settings. Additionally, we observed that specific syllables in the Moroccan dialect significantly influence the recognition system’s performance.}, archive = {J_EAAI}, author = {Abderrahim Ezzine and Naouar Laaidi and Hassan Satori}, doi = {10.1016/j.engappai.2024.109751}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109751}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic speech recognition for moroccan dialect in noisy traffic environments}, volume = {141}, year = {2025}, }
• (2025). An adaptive differential evolution with dynamic perturbation and dimensional bidirectional crossover mechanism for diversity enhancement. EAAI, 141, 109750. (www)

  @article{ , abstract = {Differential Evolution (DE) is a robust population-based evolutionary algorithm for global optimization. It has seen significant success in both evolutionary computation competitions and practical engineering applications. However, traditional DE encounters issues like early convergence and stagnation due to a decrease in population diversity when addressing complex, high-dimensional optimization problems. To address these issues, this paper proposes an adaptive differential evolution algorithm with a dynamic perturbation and dimensional bidirectional crossover mechanism. The key contributions are as follows: Firstly, a new hybrid adaptive parameter control strategy is introduced, improving the generation of the scale factor and crossover rate by dividing the evolutionary process into two stages, thereby enhancing the efficiency of parameter adaptation. Secondly, a donor vector dynamic perturbation strategy is proposed, which enhances the generation ability of donor vectors and improves population diversity by dynamically adjusting the perturbation intensity. Thirdly, a dimensional bidirectional crossover mechanism is introduced to address stagnation and maintain population diversity throughout the evolutionary process. The performance of the proposed algorithm is validated on 72 universal benchmarks and two real-world problems, and the results support its superiority.}, archive = {J_EAAI}, author = {Wencan Zhou and Zhenyu Meng}, doi = {10.1016/j.engappai.2024.109750}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109750}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An adaptive differential evolution with dynamic perturbation and dimensional bidirectional crossover mechanism for diversity enhancement}, volume = {141}, year = {2025}, }
• (2025). Structure aware transfer function network for low light image enhancement. EAAI, 141, 109749. (www)

  @article{ , abstract = {Low-light images are prevalent in surveillance, photography, and underwater imaging, but suffer from poor visibility. It is demanding to develop effective techniques for enhancing low-light images. Compared with the methods based on Retinex decomposition, the transfer function provides a simpler pipeline for low light image enhancement. However, the previous method has limitations in estimating the transfer function and the final enhancement. In this work, we present a framework for low light image enhancement, wherein we give a novel supervised approach for learning the transfer function from the low-light image. We call the proposed model S tructure A ware T ransfer function Net work (SATNet) , comprising two sub-networks: the S tructure-based T ransfer net work ( S T n e t ), and the R efinement net work ( R n e t ). The S T n e t learns the transfer function from the low-light input image, then a coarsely enhanced image is obtained simply by multiplying the low-light input by the estimated transfer function. Then the R n e t eliminates the noise in the coarsely enhanced image. Extensive experiments on seven datasets demonstrate that the proposed method outperforms baselines in brightening low-light images and preserving details.}, archive = {J_EAAI}, author = {Xiaofang Li and Weiwei Wang and Yu Han and Xiangchu Feng}, doi = {10.1016/j.engappai.2024.109749}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109749}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Structure aware transfer function network for low light image enhancement}, volume = {141}, year = {2025}, }
• (2025). Adaptive model-agnostic meta-learning network for cross-machine fault diagnosis with limited samples. EAAI, 141, 109748. (www)

  @article{ , abstract = {Deep learning-based methods have been extensively studied in rotating machinery defect diagnosis. However, training an accurate and robust diagnostic model is still a challenge under severe domain bias and limited samples. For this reason, a new adaptive model-agnostic meta-learning (AMAML) is proposed for cross-machine fault diagnosis with limited samples. First, a novel adaptive feature encode network is built, incorporating lightweight spatial-bilateral channel attention. This enables the network to extract critical fault information in multiple dimensions adaptively within limited samples, which improves the learning efficiency of generalized diagnostic knowledge. Then, an adaptive loss computation (ALC) method is devised, which inventively realizes the interaction between loss computation and model performance. The underfitting and overfitting dilemmas under few-shot conditions are tackled by ALC. Finally, an adaptive meta-optimization strategy is proposed for dynamically adapting the update strategy of the base learner, so that the model is always optimized in the direction of strong generalizability while obtaining high performance. Six cross-machine diagnosis tasks are conducted to verify the effectiveness of AMAML. The average diagnostic accuracy of the AMAML under the 5-shot setting reached 97.42%. Experiments confirm that AMAML is superior to other prevailing methods and is potentially promising for engineering applications.}, archive = {J_EAAI}, author = {Mingzhe Mu and Hongkai Jiang and Xin Wang and Yutong Dong}, doi = {10.1016/j.engappai.2024.109748}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109748}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive model-agnostic meta-learning network for cross-machine fault diagnosis with limited samples}, volume = {141}, year = {2025}, }
• (2025). Machine learning -driven predictions of lattice constants in ABX3 perovskite materials. EAAI, 141, 109747. (www)

  @article{ , abstract = {In pursuing accelerated material design, predictive modeling of lattice constants in perovskite materials has become essential for semiconductors, optoelectronics, and thermoelectrics applications. This study introduces machine learning models—Support Vector Regression (SVR), Gaussian Process Regression (GPR), Artificial Neural Networks (ANN), and Ensemble Regression Trees (ERT) trained on a comprehensive dataset incorporating ionic radii, electronegativity, density, and atomic number of perovskite constituents. Each model was optimized via the Bayesian optimization method to enhance accuracy in predicting lattice parameters. Notably, the GPR model achieved the highest precision, with an R 2 of 100% in training and 99% in testing, underscoring its capacity to capture intricate structural correlations. Results indicate that machine learning, specifically GPR, can provide an efficient and scalable alternative to traditional experimental methods, positioning these models as invaluable tools for high-throughput screening in material discovery. This approach presents a promising pathway for advancing computational material science, enabling rapid and precise lattice constant predictions to facilitate innovations across diverse technological domains.}, archive = {J_EAAI}, author = {Abdulgafor Alfares and Yusuf Abubakar Sha'aban and Ahmed Alhumoud}, doi = {10.1016/j.engappai.2024.109747}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109747}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning -driven predictions of lattice constants in ABX3 perovskite materials}, volume = {141}, year = {2025}, }
• (2025). Mention distance-aware interactive attention with multi-step reasoning for document-level relation extraction. EAAI, 141, 109746. (www)

  @article{ , abstract = {Document-level relation extraction (DocRE) is widely used in various natural language processing and artificial intelligence applications, aiming to identify relations between entities in a document that may appear multiple times or span multiple sentences. Previous methods have relied on global document features to generate a unified representation for each entity, which may result in the loss of entity-specific information and fail to capture the nuanced relations between different entity pairs. In this paper, we propose the M ention D istance-aware I nteractive A ttention with M ulti-step R easoning (MDIAMR) model, which leverages attention mechanisms and graph neural networks to address these challenges. Specifically, we introduce an entity interaction attention module to explicitly model the distance information between mentions and dynamically aggregate all mentions of each entity. This approach highlights the importance of different mentions to the entity, thereby enhancing the entity’s representation. Additionally, we design a multi-step reasoning module based on graph attention networks (GAT), which iteratively updates entity pair representations by considering information from neighboring pairs. We also incorporate a cosine similarity loss function to help the model differentiate between entity pairs that are related or unrelated to the target entity pair. Experimental results demonstrate that our model outperforms competitive baseline models on two publicly available datasets.}, archive = {J_EAAI}, author = {Fu Zhang and Jiapeng Wang and Huangming Xu and Honglin Wu and Jingwei Cheng and Weijun Li}, doi = {10.1016/j.engappai.2024.109746}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109746}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Mention distance-aware interactive attention with multi-step reasoning for document-level relation extraction}, volume = {141}, year = {2025}, }
• (2025). A deep reinforcement learning-based optimization approach for containerized microservice scheduling in hybrid Fog/Cloud environments. EAAI, 141, 109745. (www)

  @article{ , abstract = {The deployment of microservices in Hybrid Fog/Cloud (HFC) environments for Internet of Things (IoT) applications presents a significant challenge in efficiently scheduling containerized services across distributed resources. While existing studies have explored microservice scheduling, a comprehensive approach that considers resource constraints, workflow dependencies, and dynamic hybrid environments remains elusive. This paper introduces a novel Deep Reinforcement Learning-based Algorithm (DRLA) for containerized microservice scheduling in HFC environments. DRLA utilizes a multi-constrained Binary Quadratic Program (BQP) model to optimize execution time, resource consumption, and occupancy rates while considering microservice dependencies and resource capabilities. The algorithm leverages two Deep Reinforcement Learning (DRL) agents, DQN and REINFORCE, to learn and adapt to the dynamic nature of the HFC federation. Experimental evaluations using five real-world Business Process (BP) use cases demonstrate that DRLA outperforms existing scheduling approaches such as default Kubernetes, Reward Sharing Deep Q-Learning (RSDQL), and Deep Reinforcement Learning (DRL) schedulers. Compared to existing schedulers, ours delivers optimal or near-optimal solutions, demonstrating significant improvements in key performance values. Indeed, DRLA achieves average optimality gaps of just 0.16% and 0.21%, significantly outperforming the Kubernetes scheduler, which exhibits a gap of 2.88%. It is worth noting that other similar algorithms see a far higher increase in optimality gaps. This highlights DRLA’s excellent performance and ability to schedule containerized microservices in hybrid fog/cloud environments, resulting in near-optimal solutions for a variety of use cases.}, archive = {J_EAAI}, author = {Ameni Kallel and Molka Rekik and Mahdi Khemakhem}, doi = {10.1016/j.engappai.2024.109745}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109745}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep reinforcement learning-based optimization approach for containerized microservice scheduling in hybrid Fog/Cloud environments}, volume = {141}, year = {2025}, }
• (2025). Retraction notice to “Optimal infrastructure in microgrids with diverse uncertainties based on demand response, renewable energy sources and two-stage parallel optimization algorithm” [Eng. appl. artif. intell. 123 (2023) 106233]. EAAI, 141, 109744. (www)

  @article{ , archive = {J_EAAI}, author = {Yue Yu and Laleh Shahabi}, doi = {10.1016/j.engappai.2024.109744}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109744}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Retraction notice to “Optimal infrastructure in microgrids with diverse uncertainties based on demand response, renewable energy sources and two-stage parallel optimization algorithm” [Eng. appl. artif. intell. 123 (2023) 106233]}, volume = {141}, year = {2025}, }
• (2025). U-HPAUSM: Mining high probability average utility sequences in uncertain quantitative sequential databases. EAAI, 141, 109742. (www)

  @article{ , abstract = {Mining high utility sequences in quantitative sequence databases (QSDBs) is an important problem in data mining with various real-life applications. However, one drawback of this problem is that it often generates numerous long sequences, leading to difficulty for users in analyzing the result set. Furthermore, traditional QSDBs do not incorporate uncertain information, such as probability, which is common in real-world scenarios. To address this issue, an average utility function named a u is used by taking the length of the sequence into account. This paper addresses the problem of mining high probability and average-utility sequences (HPAUSs) in uncertain QSDBs, of which each item is additionally associated with a weight or probability. One significant challenge in mining HPAUSs arises from the fact that a u lacks the anti-monotonic property. In response, researchers have proposed upper bounds or weak upper bounds on a u . These bounds are designed to satisfy anti-monotonic or anti-monotone-like properties. However, the well-known bounds tend to be much larger than a u , leading to a considerably vast search space. To overcome this challenge, this paper proposes a new upper bound and the tightest weak upper bound on a u . Additionally, it introduces corresponding pruning strategies, along with a novel data structure and an algorithm named U-HPAUSM, designed to efficiently discover HPAUSs. Extensive experiments on both real-life and synthetic uncertain QSDBs demonstrate the U-HPAUSM's efficiency in terms of runtime, memory usage, and scalability. Notably, U-HPAUSM can run up to three orders of magnitude faster than the current state-of-the-art algorithms for discovering HPAUSs.}, archive = {J_EAAI}, author = {Hai Duong and Tin Truong and Tien Hoang and Bac Le}, doi = {10.1016/j.engappai.2024.109742}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109742}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {U-HPAUSM: Mining high probability average utility sequences in uncertain quantitative sequential databases}, volume = {141}, year = {2025}, }
• (2025). A multi-scale feature fusion network based on semi-channel attention for seismic phase picking. EAAI, 141, 109739. (www)

  @article{ , abstract = {In the field of seismic data processing, deep learning technologies have been widely used for seismic phase picking. However, it is difficult to take full advantage of the features extracted at different stages in existing models. In this paper, a multi-scale feature fusion network was proposed for seismic phase picking to address this problem. In the stage of feature extraction, semi-channel attention is introduced. It improves the representation ability of the model by efficiently utilizing the feature information extracted from the encoder. In the stage of decoding, a channel compression module is designed to reduce the number of feature channels. It improves the receptive field of channels. Additionally, a multi-feature fusion module is presented to integrate features at multiple scales. It reduces the loss of useful information and improves the accuracy of phase picking. The effectiveness of our network is validated on Stanford earthquake dataset, where the picking errors for phase picking are 2 ms. The parameter of our network is only 52,100. Compared with earthquake transformer, it has 42.1% fewer time costs to process 12,656 test samples on Graphics Processing Unit.}, archive = {J_EAAI}, author = {Shuguang Zhao and Jiang Wang and Ping Huang and Fa Zhao and Fudong Zhang and Yadongyang Zhu}, doi = {10.1016/j.engappai.2024.109739}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109739}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-scale feature fusion network based on semi-channel attention for seismic phase picking}, volume = {141}, year = {2025}, }
• (2025). Deep interval type-2 generalized fuzzy hyperbolic tangent system for nonlinear regression prediction. EAAI, 141, 109737. (www)

  @article{ , abstract = {Recently, due to the rapid rise of artificial intelligence (AI), considerable progress has been made in the field of nonlinear regression prediction. However, many existing methods suffer from the issues of rule and parameter explosion and poor accuracy, particularly for high-dimensional data with uncertainty. To address these limitations, this paper proposes a deep interval type-2 generalized fuzzy hyperbolic tangent system (DIT2GFHS). First, a novel neural network-based implementation of the interval type-2 fuzzy generalized fuzzy hyperbolic tangent system (IT2GFHS) is introduced to improve the efficiency of system parameter updates and optimization. Then, using a hierarchical and block-based framework, multiple IT2GFHSs are stacked layer by layer from bottom to top to construct the DIT2GFHS, with each layer’s fuzzy subsystems being independent of the others. Additionally, DIT2GFHS incorporates optimization algorithms and the Adam optimizer for training, thereby avoiding the tedious manual parameter tuning process. The detailed analysis of the construction manner and internal mechanisms for DIT2GFHS indicates that it features a reduced number of parameters, a transparent and clear structure, strong capability in handling uncertainty, and favorable accuracy. Notably, the small number of parameters and the explicit structure reduce computational and hardware burdens while maintaining interpretability. Finally, extensive experimental studies on both relatively low-dimensional and high-dimensional datasets are conducted. The results demonstrate that DIT2GFHS achieves excellent performance with fewer parameters compared to many deep-structured models, including deep fuzzy systems and deep learning models. This highlights its potential impact in addressing practical nonlinear regression problems.}, archive = {J_EAAI}, author = {Jianjian Zhao and Tao Zhao}, doi = {10.1016/j.engappai.2024.109737}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109737}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep interval type-2 generalized fuzzy hyperbolic tangent system for nonlinear regression prediction}, volume = {141}, year = {2025}, }
• (2025). Fuzzy normalization-based multi-attributive border approximation area comparison. EAAI, 141, 109736. (www)

  @article{ , abstract = {In e-commerce, robust multi-criteria decision analysis is essential for accurately aligning rankings with customer preferences. Addressing this need, the Fuzzy Normalization-based Multi-Attributive Border Approximation Area Comparison (FN-MABAC) method introduces an advanced extension of the traditional MABAC model. The primary objective is to mitigate the ranking reversal paradox while enhancing decision-makers’ preference representation. To achieve these goals, FN-MABAC employs a fuzzy normalization approach, providing a structured framework that minimizes reversal risks and aligns with reference rankings. Experimental analyses compared FN-MABAC with established methods, including Technique of Order Preference Similarity to the Ideal Solution (TOPSIS) and Characteristic Object METhod (COMET), using datasets on graphics card evaluations. Results demonstrate that FN-MABAC achieves higher concordance with reference rankings and exhibits superior resistance to ranking reversals compared to traditional approaches. Specifically, FN-MABAC achieved a Weighted Similarity (WS) coefficient match level of 0.97522 with the reference ranking, while STFN-MABAC reached a level of 0.99599, highlighting the high potential of these methods to adapt to preferences in the presented e-commerce example. Conclusions drawn indicate FN-MABAC’s potential to offer precise, stable rankings essential for e-commerce decision-making. Additionally, the concept of Stochastic Fuzzy Normalization-based MABAC (STFN-MABAC) is introduced as a re-identification approach, further aligning ranking outputs with customer preferences, particularly in scenarios with incomplete data. This study underscores FN-MABAC’s significance as an effective, resilient tool in multi-criteria analysis, contributing to improved decision-making within e-commerce contexts.}, archive = {J_EAAI}, author = {Bartłomiej Kizielewicz and Jakub Więckowski and Wojciech Sałabun}, doi = {10.1016/j.engappai.2024.109736}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109736}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fuzzy normalization-based multi-attributive border approximation area comparison}, volume = {141}, year = {2025}, }
• (2025). From density to geometry: Instance segmentation for reverse engineering of optimized structures. EAAI, 141, 109732. (www)

  @article{ , abstract = {This paper introduces You Only Look Once v8 for Topology Optimization (YOLOv8-TO), a novel approach for reverse engineering topology-optimized structures into interpretable geometric parameters using the YOLOv8 instance segmentation model. Density-based topology optimization methods require post-processing to convert the optimal density distribution into a parametric representation for design exploration and integration with computer-aided design tools. Traditional methods such as skeletonization struggle with complex geometries and require manual intervention. YOLOv8-TO addresses these challenges by training a custom YOLOv8 model to automatically detect and reconstruct structural components from binary density distributions. The model is trained on a diverse dataset of both optimized and random structures generated using the Moving Morphable Components method. A custom reconstruction loss function based on the dice coefficient of the predicted geometry is used to train the new regression head of the model via self-supervised learning. The method is evaluated on test sets generated from different topology optimization methods, including out-of-distribution samples, and compared against a skeletonization approach. Results show that YOLOv8-TO significantly outperforms skeletonization in reconstructing visually and structurally similar designs. The method showcases an average improvement of 13.84% in the Dice coefficient, with peak enhancements reaching 20.78%. The method demonstrates good generalization to complex geometries and fast inference times, making it suitable for integration into design workflows using regular workstations. Limitations include the sensitivity to non-max suppression thresholds. YOLOv8-TO represents a significant advancement in topology optimization post-processing, enabling efficient and accurate reverse engineering of optimized structures for design exploration and manufacturing.}, archive = {J_EAAI}, author = {Thomas Rochefort-Beaudoin and Aurelian Vadean and Sofiane Achiche and Niels Aage}, doi = {10.1016/j.engappai.2024.109732}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109732}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {From density to geometry: Instance segmentation for reverse engineering of optimized structures}, volume = {141}, year = {2025}, }
• (2025). Blind sonar image quality assessment via machine learning: Leveraging micro- and macro-scale texture and contour features in the wavelet domain. EAAI, 141, 109730. (www)

  @article{ , abstract = {In subsea environments, sound navigation and ranging (SONAR) images are widely used for exploring and monitoring infrastructures due to their robustness and insensitivity to low-light conditions. However, their quality can degrade during acquisition and transmission, where standard SONAR image processing techniques can hardly produce high-quality outcomes. An effective image quality assessment (IQA) method can assess their usefulness and aid to develop refinement techniques by identifying the degradation issues, ensuring the reliability of SONAR data. Existing methods often fail to account for degradations from noise, distortion, and resolution changes simultaneously. To address this challenge, we propose a new blind quality assessment method that measures the overall quality of SONAR images by quantifying both the perceptual and utility qualities using the micro- and macro-scale texture and contour features derived from the wavelet domain. By combining the local binary pattern (LBP) micro-scale texture features with the proposed histograms of Schmid Gabor-like edge maps as macro-scale features, a support vector regression model is learned to map from these features to subjective quality scores. Extensive experiments have demonstrated the superiority of our method over existing SONAR IQA techniques on distorted and reconstructed super-resolution side-scan, acoustic lens, and forward-looking SONAR images. Specifically, our method achieves Pearson’s and Spearman’s correlation metrics of 0.8616 and 0.8541, respectively, for distorted SONAR images, demonstrating improvements of 4.69% and 4.8%. For reconstructed super-resolution SONAR images, our method attains correlation metrics of 0.9415 and 0.9408, reflecting improvements of 0.8% and 1.6% over the second-best method, respectively. To facilitate ease of access, a comprehensive list of key abbreviations and their full names is provided in Table A.9 in the Appendix section. The source code of the proposed method will be shared at https://github.com/hfarhaditolie/BSIQA .}, archive = {J_EAAI}, author = {Hamidreza Farhadi Tolie and Jinchang Ren and Rongjun Chen and Huimin Zhao and Eyad Elyan}, doi = {10.1016/j.engappai.2024.109730}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109730}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Blind sonar image quality assessment via machine learning: Leveraging micro- and macro-scale texture and contour features in the wavelet domain}, volume = {141}, year = {2025}, }
• (2025). Classification model for blast furnace status based on multi-source information. EAAI, 141, 109728. (www)

  @article{ , abstract = {The blast furnace (BF) is the key equipment for smelting, which has a significant impact on the sustainable development of global environment and energy. Therefore, it will be helpful for the operators if accurate evaluation to the status of the BF is realized. The classification model based on multi-source information is proposed in this paper, where the kernel fisher (KF) algorithm is enhanced to simplify the complexity both in algorithm and data dimensionality. The proposed model can realize the classification of both the distribution of gas flow (GF) and the geometric information of the burden surface (BS), and this paper also proposes a dynamic process adaptive kernel fisher (DPAKF) algorithm to enhance the adaptive ability of the classifier, where the online algorithm updating mechanism is designed. The experimental results demonstrate that the classification accuracy was increased from 90% to 95%, and the time expense of DPAKF was 236 s, which is prior compared with the existing support vector machine (SVM), k-nearest neighbor (KNN) and genetic algorithm (GA) models. The results of the student's t-distribution show that the statistics between DPAKF and KF, SVM, KNN and GA were 1.25, 2.81, 2.84 and 2.96, respectively, which demonstrate that there are significant statistical differences between DPAKF and SVM, KNN, and GA. This research provides a potential solution to the classification problem of the BF.}, archive = {J_EAAI}, author = {Kai Guo and YaXian Zhang and Sen Zhang and WenDong Xiao}, doi = {10.1016/j.engappai.2024.109728}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109728}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Classification model for blast furnace status based on multi-source information}, volume = {141}, year = {2025}, }
• (2025). Train a real-world local path planner in one hour via partially decoupled reinforcement learning and vectorized diversity. EAAI, 141, 109726. (www)

  @article{ , abstract = {Deep Reinforcement Learning (DRL) has exhibited efficacy in resolving the Local Path Planning (LPP) problem. However, its practical application remains significantly constrained due to its limited training efficiency and generalization capability. To address these challenges, we propose a solution termed Color, which includes an Actor-Sharer-Learner (ASL) training framework designed to improve efficiency, and a fast yet diverse simulator named Sparrow aimed at elevating both efficiency and generalization. Specifically, the ASL employs a Vectorized Data Collection (VDC) mode to enhance data collection, decouples the model optimization from data collection to expedite data consumption, and partially connects the two procedures with a Time Feedback Mechanism (TFM) to evade data underuse or overuse. Meanwhile, the Sparrow simulator utilizes a 2-Dimensional (2D) grid-based world, simplified kinematics, matrix operation, and conversion-free data flow to achieve a lightweight design. The lightness facilitates vectorized diversity, allowing for rapid and diversified simulation across numerous copies of the vectorized environments, thereby significantly enhancing both efficiency and generalization capacity. Comprehensive experiments demonstrate that with merely one hour of simulation training, Color achieves impressive arrival rates of 84% and 90% on 32 simulated and 42 real-world LPP scenarios, respectively. The code and video of this paper are accessible on our website. 1}, archive = {J_EAAI}, author = {Jinghao Xin and Jinwoo Kim and Zhi Li and Ning Li}, doi = {10.1016/j.engappai.2024.109726}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109726}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Train a real-world local path planner in one hour via partially decoupled reinforcement learning and vectorized diversity}, volume = {141}, year = {2025}, }
• (2025). Ceramic tableware surface defect detection based on deep learning. EAAI, 141, 109723. (www)

  @article{ , abstract = {Detecting defects in ceramic tableware is a critical step in ensuring product quality, food safety, durability, and aesthetics, and it is important to both manufacturers and consumers. However, manual labour is still the main way to detect defects in ceramic tableware. This is because there are many difficulties in detecting defects in ceramic tableware: multi-scale, small-sized defects, and high demands on the accuracy and inference speed of the detection system. To address the challenges, we developed a detector based on You Only Look Once version 8 (YOLOv8) to achieve accurate and efficient detection of surface defects on ceramic tableware. First, a dataset containing five types of defects was produced, named the Ceramic Tableware Defect Dataset (CE5-DET). This was achieved by the image acquisition system we built. Second, a new efficient and lightweight convolution using a non-strided convolution and space-to-depth layer has been developed to solve the problems of insufficient feature extraction and loss of fine-grained information in traditional convolution. Finally, experiments were conducted on our CE5-DET and a public dataset (NEU-DET). Our model achieved a mean average precision (mAP) of 68.3% on CE5-DET and 78.7% on NEU-DET, which are 4.7% and 2.2% higher than that of baseline and significantly higher than the state-of-the-art (SOTA) detection methods. The experimental results demonstrated that the model exhibited a balance between detection accuracy and inference speed, and is expected to achieve automated detection of surface defects on ceramic tableware.}, archive = {J_EAAI}, author = {Pu Sun and Changchun Hua and Weili Ding and Changsheng Hua and Ping Liu and Ziqi Lei}, doi = {10.1016/j.engappai.2024.109723}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109723}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ceramic tableware surface defect detection based on deep learning}, volume = {141}, year = {2025}, }
• (2025). Fixed-time neural network composite learning control for uncertain nonlinear systems. EAAI, 141, 109722. (www)

  @article{ , abstract = {A novel fixed-time neural network composite learning control (FNNCLC) scheme is proposed for nonlinear strict-feedback systems with unknown dynamics. The neural network (NN) is employed to handle system uncertainty. By utilizing tracking errors and prediction errors to update NN weights, accurate network learning is achieved under a weaker excitation condition termed interval excitation (IE) condition, instead of the typically required strict persistent excitation (PE) condition. Moreover, for the first time, a high order term and first order term of the prediction error are introduced to design composite learning adaptive laws, achieving the convergence of NN weights within fixed time. Additionally, a smooth fixed-time (FXT) dynamic surface control scheme is constructed without potential singularity problems, which mitigates complexity explosion by avoiding fractional power terms and complex switching strategies when formulating the control law. The stability of the proposed control scheme is analyzed by using Lyapunov technique. Simulation results demonstrate the effectiveness of the proposed controller.}, archive = {J_EAAI}, author = {Zhonghua Wu and Zhikuan Zou and Xiangwei Bu and Jianjun Zhang and Kuncheng Ma}, doi = {10.1016/j.engappai.2024.109722}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109722}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fixed-time neural network composite learning control for uncertain nonlinear systems}, volume = {141}, year = {2025}, }
• (2025). The multi-module joint modeling approach: Predicting urban crowd flow by integrating spatial–temporal patterns and dynamic periodic relationship. EAAI, 141, 109721. (www)

  @article{ , abstract = {Accurately predicting traffic flow is crucial for forecasting the movement of people in specific regions, a core task for smart cities. Traffic flow data exhibit cyclic periodicity, with recurring patterns at regular intervals, particularly weekly. This study addresses deficiencies in periodicity modeling within crowd-flow data by proposing the Multi-Module Spatial–Temporal based Nested Encoder (MMSTNE). Unlike current methodologies, MMSTNE employs parallel periodic learning to predict traffic flow, simultaneously modeling cyclic variations between historical and future time frames. Direct prediction of dynamic traffic flow is challenging; thus, our approach focuses on learning stationary deviations to enhance model robustness. Using a highway approach within a nested framework, we establish parallel connections, incorporating Nested Long–Short-term Memory (NLSTM). These connections, linked with corresponding weekly observations, improve multi-step ahead predictions’ accuracy. Comprehensive experiments on two real-world datasets demonstrate our model’s superior accuracy compared to state-of-the-art approaches. On average, MMSTNE improved prediction accuracy by 20% over the best baseline model. This significant improvement highlights MMSTNE’s potential for dynamic urban traffic flow prediction, offering a robust solution for addressing the inherent challenges in periodicity modeling and forecasting in smart city applications.}, archive = {J_EAAI}, author = {Zain Ul Abideen and Xiaodong Sun and Chao Sun}, doi = {10.1016/j.engappai.2024.109721}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109721}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The multi-module joint modeling approach: Predicting urban crowd flow by integrating spatial–temporal patterns and dynamic periodic relationship}, volume = {141}, year = {2025}, }
• (2025). Handling data heterogeneity for wind turbine fault diagnosis via dynamic ensemble multilevel interactive learning. EAAI, 141, 109716. (www)

  @article{ , abstract = {The artificial intelligence methods used in wind turbine fault diagnosis have not adequately considered the inherent heterogeneity of SCADA data. This heterogeneity arises from the stochastic nature of wind and the operational characteristics of turbines. Neglecting the heterogeneity may lead to models struggling to capture fault characteristic patterns across varying wind speed ranges, diminishing diagnostic performance. To handle this, a novel dynamic ensemble multilevel interactive (DEMI) learning method is proposed. Firstly, a new index of wind speed jump value is presented and applied to the fine data division of diagnostic units. This process assists diagnostic units in focusing more on learning fault distribution patterns within each wind speed range. Simultaneously, to capture fault feature information from complex and variable data, deep small-world networks with short-path propagation and high clustering properties are designed within each unit for multilevel feature interaction learning. Subsequently, a multi-wind-speed unit model library is constructed, and a dynamic selection algorithm is employed to find high-performance classifiers to reduce the impact of these networks random topology. Finally, in the diagnostic phase, the final diagnostic results are obtained using online dynamic retrieval ensemble. The experimental results indicate that the DEMI enhances diagnostic performance accuracy compared to the currently advanced methods that overlook heterogeneity, while reducing false alarm rates.}, archive = {J_EAAI}, author = {Shuangxin Wang and Hongrui Li and Jiading Jiang and Meng Li and Junmei Ou and Dingli Yu}, doi = {10.1016/j.engappai.2024.109716}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109716}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Handling data heterogeneity for wind turbine fault diagnosis via dynamic ensemble multilevel interactive learning}, volume = {141}, year = {2025}, }
• (2025). Enhancing camouflaged object detection through contrastive learning and data augmentation techniques. EAAI, 141, 109703. (www)

  @article{ , abstract = {Camouflaged object detection (COD) aims to locate and segment objects that blend into their surroundings, presenting significant challenges due to the high similarity between the objects and their background. This work introduces a novel approach, Contrastive Learning with Augmented Data (CLAD), which enhances COD performance by leveraging contrastive learning and data augmentation. Our method formulates a simplified task by placing camouflaged objects in new environments, creating positive and negative samples for contrast learning. This process strengthens the model’s ability to differentiate camouflaged objects from complex backgrounds. Furthermore, we introduce a concatenated feature enhancement module to integrate and enrich multi-scale features, improving the overall expressive power of the model. Extensive experiments on four benchmark datasets demonstrate that CLAD outperforms state-of-the-art COD methods, and its effectiveness extends to salient object detection tasks, achieving competitive results across multiple metrics.}, archive = {J_EAAI}, author = {Cunhan Guo and Heyan Huang}, doi = {10.1016/j.engappai.2024.109703}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109703}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing camouflaged object detection through contrastive learning and data augmentation techniques}, volume = {141}, year = {2025}, }
• (2025). Complex deng entropy for uncertainty measure in complex evidence theory. EAAI, 141, 109696. (www)

  @article{ , abstract = {Dempster–Shafer (DS) evidence theory, an extension of probability theory, is widely utilized across various domains due to its adeptness in managing uncertain and imprecise information. Building on DS evidence theory, the complex evidence theory addresses uncertainty in decision-making in a complex plane framework. Uncertainty measurement holds a pivotal role in both evidence theory and probability theory. In this paper, capitalizing on the unique attributes of complex basic belief assignment (CBBA), we propose a novel measure of complex belief entropy, designed to evaluate total uncertainty in complex evidence theory. Notably, the proposed complex belief entropy encompasses not only discord and non-specificity but also interference, shedding light on the interactions among focal elements. Additionally, we conduct a thorough analysis of the properties associated with this newly proposed entropy. Finally, based on the proposed entropy model, a decision-making algorithm is proposed, demonstrating the superiority of the entropy model. Our findings reveal that the proposed complex belief entropy effectively measures the total uncertainty of CBBA in the framework of complex evidence theory.}, archive = {J_EAAI}, author = {Chen Tang and Fuyuan Xiao}, doi = {10.1016/j.engappai.2024.109696}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109696}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Complex deng entropy for uncertainty measure in complex evidence theory}, volume = {141}, year = {2025}, }
• (2025). Gait pattern recognition based on electroencephalogram signals with common spatial pattern and graph attention networks. EAAI, 141, 109680. (www)

  @article{ , abstract = {Assisting human locomotion in various gait patterns is one of the challenges in the interaction between human and lower limb exoskeleton. In this paper, we propose the graph attention network with electroencephalogram (EEG) signal graph structure data constructed by common spatial pattern (CSP) model to extract the spatial–temporal information from multi-channel EEG signals to recognize gait patterns including walk, run, stair descent, stair ascent, stand-to-sit, sit-to-stand, and jump. The CSP spatial filters are analyzed for EEG signals with multiple gait patterns. The EEG signal graph structure data is constructed based on the spatial–temporal features and spatial domain features of the CSP filtered data. The graph structure based on the correlation between channels and the graph structure based on the spatial relative position of EEG electrodes are constructed for comparison. The gait pattern recognition performance of the proposed model is significantly higher than that of comparison models, and the average recognition accuracy is 86.747%. The accuracy of gait pattern recognition along gait cycle is analyzed. Based on the analysis of the EEG signal graph structure reconstructed by multi-head attention coefficients of the proposed model, the learning characteristics of the model can be obtained. The relationship between EEG channels is analyzed based on the graph structures the models focusing on. Finally, the proposed method is validated on the open access brain-computer interface (BCI) competition IV Datasets 2a of EEG signals.}, archive = {J_EAAI}, author = {Yanzheng Lu and Hong Wang and Zhiguo Lu and Jianye Niu and Chong Liu}, doi = {10.1016/j.engappai.2024.109680}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109680}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Gait pattern recognition based on electroencephalogram signals with common spatial pattern and graph attention networks}, volume = {141}, year = {2025}, }
• (2025). Quantifying the trustworthiness of explainable artificial intelligence outputs in uncertain decision-making scenarios. EAAI, 141, 109678. (www)

  @article{ , abstract = {In the rapidly evolving risk management domain, Explainable AI (XAI) has transitioned from a theoretical concept to an indispensable tool, empowering decision-makers with deeper insights to make informed decisions. However, the reliability of these decisions hinges on the trustworthiness of XAI outputs. Though traditional XAI models shed light on decision-influencing factors, they should also guarantee the accuracy of their explanations. This is particularly critical when future decisions are influenced by various uncertain factors, necessitating a trustworthy determination of factors influencing those decisions. This paper introduces the Instance-based XAI for Uncertain and Dynamic Decisions (IXAI-UDD) framework to achieve this aim. IXAI-UDD combines instance-based learning with transparent, supervised, and unsupervised algorithms to enable decision-makers to adjust the feature values in light of uncertainties. This is used to first accurately identify the decision output for a given time slot and subsequently determine the key features affecting that output. In our case scenario, the framework achieved an accuracy rate of 96% for high, low, and safe risk classifications while identifying contributing features with over 85% precision across multiple time slots. This enables effective management of uncertainties or anomalies impacting the input data before determining the output and the features leading to that output. To validate the accuracy and reliability of IXAI-UDD outputs, we employ Mosaic for precision assessment and Heatmaps for multi-class evaluation. This comprehensive methodology enhances the decision-maker’s understanding of risk factors, improving risk management strategies in dynamic, uncertain environments.}, archive = {J_EAAI}, author = {Sonia Farhana Nimmy and Omar K. Hussain and Ripon K. Chakrabortty and Abderrahmane Leshob}, doi = {10.1016/j.engappai.2024.109678}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109678}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Quantifying the trustworthiness of explainable artificial intelligence outputs in uncertain decision-making scenarios}, volume = {141}, year = {2025}, }
• (2025). Upper limb musculoskeletal model as path generator for control a virtual orthosis: A dynamic neural network approach. EAAI, 141, 109670. (www)

  @article{ , abstract = {This work presents the design and implementation of a reference path generator for a virtual version of a robotic orthosis based on a semi-parametric model of a neuromusculoskeletal system. The proposed generator is used to regulate the movements of the mentioned virtual orthosis (VO) as a preliminary stage in designing rehabilitation strategies. The generator considers a differential neural network (DNN) identifier, which predicts the angular positions and velocities of specific articulations in the upper limb using the raw electromyographic (EMG) signals as input. The DNN-based model is validated using experimental data from the elbow joint and the Biceps Brachii muscle collected from ten healthy participants. To regulate the movements of the VO, a controller based on the sliding mode considers the motion restrictions of the articulation of the extremity intervened by the orthosis. The proposed controller guarantees that the VO reaches a set point following a reference path related to the rEMG while the motion constraints are satisfied. The functionality of the proposed path generator was tested with a VO device. The results showed that the VO followed the angular movements of the elbow. All these results confirm the applicability of the proposed semi-parametric path generator based on a DNN.}, archive = {J_EAAI}, author = {Alejandro Lozano and David Cruz-Ortiz and Mariana Ballesteros and Isaac Chairez}, doi = {10.1016/j.engappai.2024.109670}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109670}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Upper limb musculoskeletal model as path generator for control a virtual orthosis: A dynamic neural network approach}, volume = {141}, year = {2025}, }
• (2025). Using the attention layer mechanism in construction of a novel ratio control chart: An application to ethereum price prediction and automated trading strategy. EAAI, 141, 109652. (www)

  @article{ , abstract = {In the area of multivariate process quality control, it is sometimes important to monitor the ratio of two normal random variables denoted by RZ over time. The concept of control charts has often been harnessed in this field, leading to the application of various types of statistical models, including Shewhart, Exponentially Weighted Moving Average (EWMA), and so forth. However, there is little attention to implementation of machine learning-based control charts. To bridge this gap, a novel machine learning based model incorporating the attention mechanism approach, as an implemented Artificial Intelligence (AI) model, is proposed to monitor the RZ in Phase II applications. The proposed RZ method not only provides quicker Out-of-Control (OC) shift detection than conventional RZ control charts but also does not require the quality controller to have any prior information about the upward or downward shift patterns, which is a major assumption in most of the previous RZ models. We provide extensive performance comparison results to discuss the statistical performance of our proposed method through Monte Carlo simulations. Moreover, a comprehensive real example about surveillance of the cryptocurrency market is provided to illustrate the practical application of our proposed method. Through simulation and back-testing results, it is shown how the proposed method can lead to an automated trading strategy.}, archive = {J_EAAI}, author = {Ali Yeganeh and XueLong Hu and Sandile Charles Shongwe and Frans F. Koning}, doi = {10.1016/j.engappai.2024.109652}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109652}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Using the attention layer mechanism in construction of a novel ratio control chart: An application to ethereum price prediction and automated trading strategy}, volume = {141}, year = {2025}, }
• (2025). Opportunities and pitfalls of regression algorithms for predicting the residual value of heavy equipment — A comparative analysis. EAAI, 141, 109599. (www)

  @article{ , abstract = {The residual value of heavy equipment is essential for financial and economic considerations in the construction industry. In practice, empirical methods are frequently used to determine the residual value of a given piece of equipment. Here, various regression methods are compared based on a real-world dataset of used heavy equipment sales from a construction company. The results show that the prediction performance of traditional methods is clearly worse when compared to machine learning models not yet employed for this purpose. For the latter, preprocessing and parameter tuning are essential, and the article guides through these steps. Further, the article demonstrates how a variable importance value comparable across all methods can be obtained. These findings may also be useful in other applications.}, archive = {J_EAAI}, author = {Marco Huymajer and Peter Filzmoser and Alexandra Mazak-Huemer and Leopold Winkler and Hans Kraxner}, doi = {10.1016/j.engappai.2024.109599}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109599}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Opportunities and pitfalls of regression algorithms for predicting the residual value of heavy equipment — A comparative analysis}, volume = {141}, year = {2025}, }
• (2025). Hybrid grid search and bayesian optimization-based random forest regression for predicting material compression pressure in manufacturing processes. EAAI, 141, 109580. (www)

  @article{ , abstract = {In the realm of high-performance material development and industrial manufacturing, accurately detecting key data during the material compression process and predicting pressure are crucial. Nevertheless, extant methodologies have not attained accurate and real-time monitoring of data or the prediction of pressure in high-stress material compression processes for high-stress materials. To address this issue, this study introduces a pressure prediction methodology predicated on an optimized random forest regression model. This model undergoes efficient optimization via a univariate grid search combined with the integration of Bayesian filtering techniques, and engages in an analysis of variable importance through SHapley Additive exPlanations values. The study collects real-time compression deformation data and pressure values during the compression process of high-stress materials using an innovative material compression apparatus. It processes the experimental data and implements a noise-induced data augmentation strategy to address the issue of small sample sizes to construct a material compression process dataset. The findings reveal that this method reduces the training time to 3.5 s and diminishes the mean absolute percentage error to 2.65%, surpassing the performance of other regression algorithms. Moreover, simulating the predictive performance of this method in different noise environments indicates that this method has a lower root mean square error in most scenarios compared to the original algorithm, demonstrating excellent robustness. Conclusively, this methodology showcases exceptional accuracy and robustness in material compression experiments using 6061 aluminum alloy as an example and can be extended to pressure prediction of other similar materials, offering a broad application prospect.}, archive = {J_EAAI}, author = {Youcheng Zong and Yi Nian and Chaojie Zhang and Xinyu Tang and Lin Wang and LiQiang Zhang}, doi = {10.1016/j.engappai.2024.109580}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109580}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid grid search and bayesian optimization-based random forest regression for predicting material compression pressure in manufacturing processes}, volume = {141}, year = {2025}, }
• (2025). Machine learning-driven predictive modeling of magnetohydrodynamic double diffusion of non-newtonian hybrid ferrofluids with variable thermophysical properties within corrugated cylinders. EAAI, 141, 109455. (www)

  @article{ , abstract = {This study addresses the complex problem of magnetohydrodynamic (MHD) double diffusion in non-Newtonian hybrid ferrofluids within a concentric corrugated cylinder. Understanding this phenomenon is crucial due to its applications in advanced cooling systems, biomedical drug delivery, and industrial processes where precise control of fluid dynamics and heat transfer is essential. Traditional simulations face challenges in capturing these complexities accurately, making it imperative to explore more efficient modeling approaches. To overcome these challenges, we employed a computational fluid dynamics (CFD) framework combined with machine learning predictive modeling. Four machine learning algorithms – decision tree regression (DT), random forest regression (RF), feed-forward neural networks (FFNN), and 1-dimensional convolutional neural networks (1D-CNN) – were used to analyze the behavior of the fluid under various conditions. The CFD simulations revealed that increasing the power law index ( n ) results in a decrease in flow rates, with a 20.53% reduction for n from 0.8 to 1 and a 49.04% reduction for n from 1 to 1.4. Significant variations in volume fraction ( ϕ ) and Hartmann number ( H a ) were observed to affect average Nusselt number ( N u ¯ ) and Sherwood number ( S h ¯ ). The machine learning models demonstrated high predictive accuracy, with 1D-CNN achieving R 2 values of 0.9996, 0.9994, and 0.9995 for N u ¯ , S h ¯ , and | Ψ | , respectively. The FFNN achieved R 2 values of 0.9995, 0.9984, and 0.9995 for the same metrics. These results confirm that the studied parameters-Rayleigh number ( R a ), Hartmann number ( H a ), and power law index ( n )-significantly influence flow properties. The novelty of this work lies in the integration of machine learning techniques with CFD simulations to model MHD double diffusion in non-Newtonian hybrid ferrofluids. This approach provides a more precise and resource-efficient method for analyzing complex fluid behaviors compared to traditional simulation methods. The findings have significant implications for various applications, including energy systems, biomedical engineering, industrial cooling, and chemical processing. The study not only advances the understanding of MHD phenomena in non-Newtonian fluids but also offers practical insights for designing improved cooling systems and other applications where precise fluid control is essential.}, archive = {J_EAAI}, author = {Jawad Ibn Ahad and Md. Mamun Molla and Sadia Siddiqa and Sahrish Batool Naqvi}, doi = {10.1016/j.engappai.2024.109455}, journal = {Engineering Applications of Artificial Intelligence}, month = {2}, pages = {109455}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning-driven predictive modeling of magnetohydrodynamic double diffusion of non-newtonian hybrid ferrofluids with variable thermophysical properties within corrugated cylinders}, volume = {141}, year = {2025}, }
• (2025). Solving dynamic multi-objective optimization problem of immersed tunnel elements via multi-source evolutionary information clustering method. EAAI, 140, 109741. (www)

  @article{ , abstract = {Dynamic multi-objective optimization problems (DMOPs) are time- and space-varying, thus maintaining/improving the uncertainty degree of evolutionary information (i.e., information entropy) in the population and providing useful knowledge are two important tasks to make dynamic multi-objective evolutionary algorithms (DMOEAs) adapt to changing environments. To achieve the above objectives, a multi-source population clustering (MPC) method is proposed to assist DMOEAs in improving their tracking performance during the full-cycle optimization in the current study. In the MPC, three different information sources are used to provide diverse spatiotemporal evolutionary information, aiding DMOEAs in adapting to various changing environments. Subsequently, an enhanced spectral clustering approach is employed to group all evolutionary individuals from different information sources into many clusters/subspaces. Finally, the selected DMOEA is employed to search all subspaces in parallel via the high-performing computing method. The MPC is incorporated into a regularity model-based multi-objective estimation of distribution algorithm (called as MPC-RM-MEDA) and is compared with six famous DMOEAs on 14 10- and 30-dimensional DMOPs, which are proposed in IEEE Congress on Evolutionary computation 2018. Experimental results demonstrate that the overall tracking performance of the proposed MPC-RM-MEDA is significantly superior to that of other selected competitors in various dynamic environments. Additionally, the MPC-RM-MEDA is utilized to address a real-world DMOP involving an immersed tunnel element. The obtained results and comparison with the knee point-based transfer learning method verify that the MPC is an efficient and dependable approach for enhancing the tracking performance of other DMOEAs in solving actual DMOPs.}, archive = {J_EAAI}, author = {Qinqin Fan and Wentao Huang and Moduo Yu and Qirong Tang and Qingchao Jiang}, doi = {10.1016/j.engappai.2024.109741}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109741}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Solving dynamic multi-objective optimization problem of immersed tunnel elements via multi-source evolutionary information clustering method}, volume = {140}, year = {2025}, }
• (2025). A heterogeneous transfer learning method for fault prediction of railway track circuit. EAAI, 140, 109740. (www)

  @article{ , abstract = {Prediction and identification of faults in track circuits are crucial for improving the safety and efficiency of railway transportation. However, due to the absence of real data, the task of track circuit fault prediction through deep learning methods facing significant challenges. This paper proposed a novel heterogeneous transfer learning network structure for track circuit deep learning fault prediction. The proposed transfer learning network can reduce the reliance on track circuit data in the process of deep learning models training by utilizing public datasets from other similar tasks. In this paper, an index describing the data distribution is used to demonstrate the transferability between heterogeneous data firstly. Then a heterogeneous transfer learning network structure is proposed to help the deep learning model training on the track circuit fault prediction task. Finally, the effect of transfer learning is comprehensively examined. The simulation experimental results show that the proposed heterogeneous transfer learning network structure can transfer useful knowledge in other similar fields for tasks in track circuit fault prediction, and the resulting model can distinguish between nine different classes with a high accuracy level over 99% on the test dataset while reducing the amount of required training data to 10% of the traditional training methods.}, archive = {J_EAAI}, author = {Lan Na and Baigen Cai and Chongzhen Zhang and Jiang Liu and Zhengjiao Li}, doi = {10.1016/j.engappai.2024.109740}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109740}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A heterogeneous transfer learning method for fault prediction of railway track circuit}, volume = {140}, year = {2025}, }
• (2025). Physics-informed deep homogenization approach for random nanoporous composites with energetic interfaces. EAAI, 140, 109735. (www)

  @article{ , abstract = {This contribution presents a new physics-informed deep homogenization neural network model for identifying local displacement and stress fields, as well as homogenized moduli, of nanocomposites with periodic arrays of porosities under general loading conditions. Notably, it accounts for the surface elasticity effect, utilizing the Gurtin-Murdoch interface theory. First of all, a fully connected neural network model is established that maps the spatial coordinates, passing first through several sinusoidal functions, to the microscopic displacements. The loss function is formulated as the weighted sum of residuals of Navier-Cauchy equations in the bulk domains and the Young-Laplace equations on the energetic surfaces, evaluated on separate sets of collocation points. To more effectively predict stress concentrations inside the microstructures, we introduce fully trainable weights to each collocation point. The capacity and effectiveness of the new homogenization technique for capturing the size-dependent local and global response of nanocomposites with distinct pore sizes and shapes are verified upon extensive comparisons with the finite-element benchmark results, under various loading conditions. New results showcase the proposed theory's ability to model random distributions of nano-porosities with a high degree of accuracy, a task not easily achievable with alternative techniques except for the specialized finite-element method.}, archive = {J_EAAI}, author = {Qiang Chen and George Chatzigeorgiou and Fodil Meraghni and Xuefeng Chen and Zhibo Yang}, doi = {10.1016/j.engappai.2024.109735}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109735}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Physics-informed deep homogenization approach for random nanoporous composites with energetic interfaces}, volume = {140}, year = {2025}, }
• (2025). Predefined time convergence guaranteed performance control for uncertain systems based on reinforcement learning. EAAI, 140, 109734. (www)

  @article{ , abstract = {The prescribed performance control method (PPCM) is commonly employed to ensure the guaranteed performance control of non-linear systems. However, traditional approaches suffer from certain drawbacks, such as the dependence of parameter settings for the performance constraint function on the initial tracking error value and the inability to specify the convergence time of tracking error according to engineering requirements. This paper focuses on designing a fault tolerant control strategy with prescribed convergence time and prescribed transient performance for uncertain systems, considering parameter perturbance, actuator faults, and unknown initial states. Firstly, we introduce an error conversion function that transforms the tracking error with any initial value into a new error variable starting from zero. This resolves the issue of depending on the initial value of tracking error in setting parameters for the performance constraint function in prescribed performance control methods. Subsequently, we derive a novel Lyapunov stability criterion for predefined time (PDT) convergence and design a fault-tolerant control strategy using backstepping control method while ensuring prescribed convergence time and prescribed performance. In this approach, we propose a new online reinforcement learning intelligent algorithm to estimate compound interference caused by actuator faults, control saturation constraint increment, system parameter perturbation, and external interference. The theoretical proof establishes predefined time convergence of the closed-loop system. Finally, numerical simulations are conducted on industrial robots with actuator faults to validate the effectiveness of our designed control strategy.}, archive = {J_EAAI}, author = {Chun-Wu Yin}, doi = {10.1016/j.engappai.2024.109734}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109734}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predefined time convergence guaranteed performance control for uncertain systems based on reinforcement learning}, volume = {140}, year = {2025}, }
• (2025). Multimodal sentiment analysis based on multiple attention. EAAI, 140, 109731. (www)

  @article{ , abstract = {The development of the Internet makes various types of data widely appear on various social platforms, multimodal data provides a new perspective for sentiment analysis. Although the data types are different, there are information expressing the same sentiment. The existing researches on extracting those information are static, and this means that there is a problem of extracting common information in a fixed amount. Therefore, to address this problem, we proposes a method named multimodal sentiment analysis based on multiple attention(MAMSA). Firstly, this method utilized the adaptive attention interaction module to dynamically determine the amount of information contributed by text and image features in multimodal fusion, and multimodal common representations are extracted through cross modal attention to improve the performance of each modal feature representation. Secondly, using sentiment information as a guide to extract text and image features related to sentiment. Finally, using hierarchical manner to fully learning the internal correlations between sentiment-text association representation, sentiment-image association representation, and multimodal common information to improve the performance of the model. We conducted extensive experiments using two public multimodal datasets, and the experimental results validated the availability of the proposed method.}, archive = {J_EAAI}, author = {Hongbin Wang and Chun Ren and Zhengtao Yu}, doi = {10.1016/j.engappai.2024.109731}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109731}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodal sentiment analysis based on multiple attention}, volume = {140}, year = {2025}, }
• (2025). A return-to-home unmanned aerial vehicle navigation solution in global positioning system denied environments via bidirectional long short-term memory reverse flightpath prediction. EAAI, 140, 109729. (www)

  @article{ , abstract = {In this paper, bidirectional long short-term memory (B-LSTM) deep learning modeling is proposed as an approach to facilitate autonomous return-to-home (RTH) aerial navigation in environments with compromised global positioning system (GPS) reception. Logged samples of ten radiometric features are extracted from onboard sensors (i.e., accelerometer, barometer, GPS, gyroscope, magnetometer) in two outdoor experimental scenarios of different altitudes and velocities. These samples are used for training and validating B-LSTM models with single and parallel architectures. The former architecture consists of a single B-LSTM model that processes all input features across the x -, y -, and z -axes to predict a three-dimensional local position, whereas the latter comprises three parallel B-LSTM models, each for processing only the features of a specific dimension (i.e., x , y , or z ) and predicting local position in the respective axis. Evaluations demonstrate the validity of the proposed approach, with a 4-m average mean square error (MSE). This is achieved without imposing resource-consuming computational overhead, modifications to existing hardware, or changes to physical infrastructure and communication protocols. Due to using existing onboard sensors and accommodating varied scenarios, the proposed approach finds applications in autonomous navigation, including unmanned aerial vehicles (UAVs) and ground vehicles.}, archive = {J_EAAI}, author = {Mustafa Alkhatib and Mohammad Nayfeh and Khair Al Shamaileh and Naima Kaabouch and Vijay Devabhaktuni}, doi = {10.1016/j.engappai.2024.109729}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109729}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A return-to-home unmanned aerial vehicle navigation solution in global positioning system denied environments via bidirectional long short-term memory reverse flightpath prediction}, volume = {140}, year = {2025}, }
• (2025). A deep neural network-based multi-layer classifier ensembles for intrusion detection in fog-based internet of things environments. EAAI, 140, 109727. (www)

  @article{ , abstract = {Intrusion Detection Systems (IDS) are regarded as an efficient security mechanism in the Internet of Things (IoT). IDSs in the IoT face critical challenges, such as a vast amount of data with many features, limited resources, detection of unknown attacks, and many false alarms. To overcome these issues, this paper proposes a novel framework, namely a Multi-Layer Multi-Classification System (MLMCS). The proposed framework consists of four steps: In the first step, the data collected from information sources are pre-processed. The second step divides the feature space into several subspaces to reduce the model's complexity. The third step proposes a multilayer base classifier called a Multi-Layer Classifier System (MLCS) for each feature subspace. Each MLCS is layered to transform a multi-class classification into several binary classifications so that a specific category of attacks has been detected in each layer based on the Group Method of Data Handling (GMDH) neural network. In the fourth step, the results obtained from MLCSs based on the Dempster-Shafer theory are combined to reduce the problems related to overfitting. The simulation results were gained using three datasets, i.e., NSL-KDD, UNSW-NB15, and TON-IoT. They indicated that the proposed method, compared with other methods, improved F1-score by 5.64%, 0.65%, and 0.11%, respectively. Furthermore, the false alarm rate was reduced by 2.66%, 2.52%, and 0.02%. Moreover, the impact of four well-known adversarial attacks on the dataset was investigated. It was observed that the MLMCS method, with an average 5.21% reduction, could perform well against these attacks.}, archive = {J_EAAI}, author = {Hossein khosravifar and Mohammad Ali Jabraeil Jamali and Kambiz Majidzadeh and Mohammad Masdari}, doi = {10.1016/j.engappai.2024.109727}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109727}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep neural network-based multi-layer classifier ensembles for intrusion detection in fog-based internet of things environments}, volume = {140}, year = {2025}, }
• (2025). Deep reinforcement learning optimizer based novel caputo fractional order sliding mode data driven controller. EAAI, 140, 109725. (www)

  @article{ , abstract = {The design of controllers in engineering applications typically requires a model that accurately captures the dynamics of the real system. However, finding a precise model for controller design can be challenging in real engineering applications. Consequently, data-driven methods have gained widespread use in engineering systems. This paper presents a novel robust data-driven fractional-order controller optimized through deep reinforcement learning. Additionally, a new robust fractional-order observer has been introduced to improve both the robustness and speed of the system. To establish the stability of the proposed control system, a new Lyapunov stability theorem based on the Caputo fractional-order definition is provided. The proposed controller offers significant advantages, including enhanced robustness against external disturbances, increased resilience to parameter uncertainties and unmodeled nonlinear dynamics, improved accuracy, greater speed, and guaranteed optimal control coefficients. Furthermore, assured adaptability is demonstrated due to the optimization provided by deep reinforcement learning including enhanced robustness against external disturbances, uncertainties of parameters, and unmodeled nonlinear dynamics; improved accuracy; greater speed; and guaranteed optimal control coefficients. Furthermore, assured adaptability is demonstrated due to the optimization provided by deep reinforcement learning. The performance of the proposed method has been compared with that of conventional integer-order sliding mode control, highlighting the superiority of this approach. The proposed method has been evaluated under normal conditions, external disturbances, and system uncertainties. Notably, performance improvements of 15%, 30%, and 68% have been achieved under normal conditions, external disturbances, and internal uncertainties, respectively, compared to the conventional integer-order sliding mode controller.}, archive = {J_EAAI}, author = {Amir Veisi and Hadi Delavari}, doi = {10.1016/j.engappai.2024.109725}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109725}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Deep reinforcement learning optimizer based novel caputo fractional order sliding mode data driven controller}, volume = {140}, year = {2025}, }
• (2025). Towards robust ferrous scrap material classification with deep learning and conformal prediction. EAAI, 140, 109724. (www)

  @article{ , abstract = {The classification of ferrous scrap materials is a well-explored problem in the literature, recognized for its significance in the steel production industry. While deep learning models are effective for this task, their deployment in industrial settings requires addressing model uncertainties and ensuring proper calibration. This study proposes adapting split conformal prediction to quantify uncertainties and facilitate model calibration. The results indicate that the Hierarchical Vision Transformer using Shifted Windows (Swin) models, particularly Swin V2, serves as the most reliable backbone for this task. Although the performance of Swin models is comparable to other evaluated models, Swin V2 demonstrates superior confidence, achieving 95.51% accuracy and the lowest conformal prediction threshold. The method is rigorously evaluated on a real-world dataset comprising 8,147 images across nine classes of ferrous scrap widely used in the Brazilian steel industry. Explainability methods corroborate the results of conformal prediction, enhancing transparency and trust in model predictions, and thereby facilitating industrial adoption. This approach bridges the gap between advanced deep learning and practical application in ferrous scrap classification, underscoring the importance of model calibration in industrial deployment.}, archive = {J_EAAI}, author = {Paulo Henrique dos Santos and Valéria de Carvalho Santos and Eduardo José da Silva Luz}, doi = {10.1016/j.engappai.2024.109724}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109724}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards robust ferrous scrap material classification with deep learning and conformal prediction}, volume = {140}, year = {2025}, }
• (2025). Lightweight advanced deep-learning models for stress detection on social media. EAAI, 140, 109720. (www)

  @article{ , abstract = {Nowadays, stress reveals itself as a ubiquitous presence, manifesting in novel forms in our modern daily life. Indeed, digital platforms and social media collect various impressions, reactions, and feelings that could provide valuable real-time sentiment data. Nevertheless, understanding stress and mental states among people is difficult because it relies on self-reporting and detecting related expressions, statements, and articulations. In this paper, we consider extracting nuanced insights and stress expressions from Reddit and Twitter posts using lightweight advanced deep-learning methods and Bidirectional Encoder Representations from Transformers (BERT) embeddings. Our findings highlight the potency of transformer BERT models, whether utilized as embedding feature extractors or as text sentiment classifiers. Moreover, the proposed lightweight deep architectural models promoted the field of stress detection in social media, achieving high classification performance. Practically, the BERT Electra model reached 85.67% accuracy on the small Reddit dataset, while our Convolutional Neural Network (CNN) model obtained 97.62% on the large Twitter dataset. Our contributions are not only restricted to the scientific understanding of stress but also extend to the well-being of individuals and global mental health.}, archive = {J_EAAI}, author = {Mohammed Qorich and Rajae El Ouazzani}, doi = {10.1016/j.engappai.2024.109720}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109720}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight advanced deep-learning models for stress detection on social media}, volume = {140}, year = {2025}, }
• (2025). Reliable federated learning based on delayed gradient aggregation for intelligent connected vehicles. EAAI, 140, 109719. (www)

  @article{ , abstract = {As an organic combination of the Internet of Vehicles and intelligent vehicles, Intelligent Connected Vehicles (ICVs) have very high research and application value. Traditional data application methods require the local aggregation of sensitive user data, which poses a threat to user data privacy. Federated learning (FL) is a promising machine learning method that leverages distributed, personalized datasets to enhance performance while preserving user privacy. However, in mobile environments, unreliable client data can degrade the global model, reducing accuracy. Additionally, the mobility of ICVs can destabilize the training process, prolonging model updates and diminishing aggregation accuracy. To address these challenges, this paper proposes a dynamic asynchronous aggregation method that improves both reliability and training efficiency in FL for mobile networks. Therefore, it becomes crucial to find reliable aggregation of mobile device participation in FL tasks. To this end, we propose a reliable FL scheme, which only selects reliable mobile devices to participate in model aggregation to improve the generalization ability of the model. In addition, we design a dynamic asynchronous aggregation method based on reputation scores without affecting the model. Reduce model training time without compromising performance. Through experimental analysis, it is proved that this method can improve the reliability and effectiveness of FL tasks in mobile networks.}, archive = {J_EAAI}, author = {Zhigang Yang and Cheng Cheng and Zixuan Li and Ruyan Wang and Xuhua Zhang}, doi = {10.1016/j.engappai.2024.109719}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109719}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reliable federated learning based on delayed gradient aggregation for intelligent connected vehicles}, volume = {140}, year = {2025}, }
• (2025). Trajectory prediction based on the dynamic characteristics and coupling relationships among vehicles in highway scenarios. EAAI, 140, 109718. (www)

  @article{ , abstract = {In dynamic and highly interactive traffic scenarios, vehicle trajectory prediction has become a critical and complex problem in vehicle assistance systems due to the uncertainty of traffic participant behaviour. This study proposes a novel vehicle trajectory prediction method called the Dynamic Coupling-based Long and Short-Term Memory Network (DSKM-LSTM). The method innovatively combines a vehicle dynamics model with the dynamic coupling between vehicles by integrating the intrinsic motion laws of the vehicles with the correlation of the motion parameters (position, velocity, and acceleration) between the vehicles via Kalman filtering. Subsequently, this information is embedded into a Long Short-Term Memory (LSTM)-based encoder-decoder framework to encode and decode historical trajectories with the fused information in order to predict future driving trajectories. Additionally, this paper introduces a dynamic loss function that optimizes the generalization capability and iteration time of DSKM-LSTM by adjusting the threshold value in real time and dynamically updating the loss value. This approach effectively addresses the problem of insufficient objectivity and accuracy in existing vehicle trajectory prediction models and provides a theoretical foundation for generating objective and interpretable trajectories in the field of trajectory prediction. Validation results on three publicly available datasets show that DSKM-LSTM achieves efficient and accurate long-term (5-s) trajectory prediction in complex traffic scenarios (e.g., motorways).}, archive = {J_EAAI}, author = {Peichao Cong and Murong Deng and Yixuan Xiao and Yangang Zhu and Xin Zhang}, doi = {10.1016/j.engappai.2024.109718}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109718}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Trajectory prediction based on the dynamic characteristics and coupling relationships among vehicles in highway scenarios}, volume = {140}, year = {2025}, }
• (2025). Performance evaluation of uranium enrichment cascades using fuzzy based harmony search algorithm. EAAI, 140, 109710. (www)

  @article{ , abstract = {The production of energy in nuclear reactors needs enrichment of fuels. There is some interest in taking the fuel enrichment level to 3–5% by cascades. Optimization of the isotopic cascades is essential to make this process economic. This study presents a Fuzzy-based Harmony Search (FHS) algorithm aimed at dynamic parameter adaptation as well as establishing a balance between exploration and exploitation, which significantly increases the convergence speed of the algorithm. Accelerating the convergence of the algorithm is demonstrated in the Sphere, Schwefel, Ackley, and Drop-Waves benchmarks at first. This approach also enhances performance in several test cases of optimum cascade problems, with results validated through comparisons with conventional methods. According to the results, the total number of centrifuges using FHS reached 6306 in test case 1, which was reduced 44 pieces compared to the method used by Palkin, and 55 pieces compared to the real coded genetic algorithm. The total number of centrifuges using FHS reached 2808 in test case 4 with a different type of gas centrifuge, which decreased 27 pieces compared to the direct search method. Similar results were obtained in other test cases, indicating the effectiveness of the FHS algorithm in minimizing the total number of centrifuges and total flow rates.}, archive = {J_EAAI}, author = {S. Dadashzadeh and M. Aghaie}, doi = {10.1016/j.engappai.2024.109710}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109710}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Performance evaluation of uranium enrichment cascades using fuzzy based harmony search algorithm}, volume = {140}, year = {2025}, }
• (2025). An improved reinforcement learning-based differential evolution algorithm for combined economic and emission dispatch problems. EAAI, 140, 109709. (www)

  @article{ , abstract = {To overcome challenges posed by escalating environmental pollution and climate change, the combined economic and emission dispatch problem is proposed to balance economic efficiency with emission cost. The primary objective of the problem is to ensure that emissions are minimized while optimal economic costs are achieved simultaneously. However, due to the nonlinear and nonconvex characteristics of the model, the optimization is confronted with many difficulties. Hence, an innovative improved reinforcement learning-based differential evolution algorithm is proposed in this article, with reinforcement learning seamlessly integrated into the differential evolution algorithm. Q-learning from reinforcement learning technique is utilized to dynamically adjust parameter settings and select appropriate mutation strategies, thereby boosting the algorithm's adaptability and overall performance. The effectiveness of the proposed algorithm is tested on thirty testing functions and combined economic and emission dispatch problems in comparison with the other five algorithms. According to the experimental results of testing functions, superior performance is consistently achieved by the proposed algorithm, with the highest adaptability exhibited and an average ranking of 1.4167. Its superiority is further demonstrated through Wilcoxon tests on results of testing functions and combined economic and emission dispatch problems with the proportion of 100%, and the proposed algorithm is significantly better than other algorithms at a 0.05 significance level. The superiority of the proposed algorithm in optimizing combined economic and emission dispatch problems demonstrates that the proposed algorithm is shown to be adaptable to complex optimization environments, which proves useful for industrial applications and artificial intelligence.}, archive = {J_EAAI}, author = {Yuan Wang and Xiaobing Yu and Wen Zhang}, doi = {10.1016/j.engappai.2024.109709}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109709}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An improved reinforcement learning-based differential evolution algorithm for combined economic and emission dispatch problems}, volume = {140}, year = {2025}, }
• (2025). LCRTR-net: A low-cost real-time recognition network for rail corrugation in railway transportation. EAAI, 140, 109708. (www)

  @article{ , abstract = {Rail corrugation has a significant impact on the safety of high-speed railway operations, making its identification particularly important. Traditional manual inspection methods are infeasible for large-scale identification within limited time frames, while existing methods based on machine vision or axle box acceleration face challenges such as high costs, complex equipment installation and maintenance, as well as difficulties in achieving real-time performance. To address these challenges, this study proposes an innovative low-cost real-time recognition network (LCRTR-Net), which utilizes accelerometers installed on the underside of the train body and combines wavelet packet decomposition with dilated causal convolution in a residual neural network. Specifically, the approach first extracts the latent features of train body acceleration caused by rail corrugation through wavelet packet decomposition and reconstruction. Next, dilated causal convolution is employed to capture the temporal causal relationships and long-term dependencies of these latent features. Finally, the integration of residual connections further enhances the feature extraction performance and computational efficiency of LCRTR-Net. Experimental results demonstrate that LCRTR-Net exhibits significant generalization ability and real-time performance, achieving an average recognition accuracy exceeding 97.0%, with a recognition time of only 0.17 ms per rail corrugation sample, significantly outperforming existing rail corrugation recognition methods. This indicates that LCRTR-Net has broad application potential in practical railway operations. Future research directions will focus on unsupervised or few-shot learning algorithms and multi-sensor integration to further improve recognition accuracy and real-time performance, promoting the practical application of this technology.}, archive = {J_EAAI}, author = {Xueyang Tang and Xiaopei Cai and Yuqi Wang and Yue Hou}, doi = {10.1016/j.engappai.2024.109708}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109708}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {LCRTR-net: A low-cost real-time recognition network for rail corrugation in railway transportation}, volume = {140}, year = {2025}, }
• (2025). Semantic segmentation model based on edge information for rock structural surface traces detection. EAAI, 140, 109706. (www)

  @article{ , abstract = {Fast and accurate detection of rock structural surface traces is crucial for geology and engineering fields. In recent years, deep learning techniques like U-Net (UNet) have been applied to rock structural surface traces detection by virtue of its high accuracy and strong robustness. However, the loss of important information during the downsampling process may hinder the model performance for rock structural surface traces detection. To alleviate this problem, this paper proposes a semantic segmentation model based on edge information (Edge-UNet) for rock structural surface traces detection. In Edge-UNet, an edge pooling method is designed, which can retain more trace features rich in edge information in the downsampling process, so as to enhance the learning of the model for traces. Then, an edge semantic enhancement structure based on edge pooling is designed to strengthen the edge information in Edge-UNet's encoder. In addition, a channel space attention gate based on edge information is incorporated in Edge-UNet's decoder, which facilitates the model to capture fine trace features. These designs clarify the retention and utilization of edge information in principle which enhances the interpretability of the model. Finally, Convolutional neural network -based and Transformer-based semantic segmentation models were selected for comparison experiments with Edge-UNet, respectively. From the experimental results, Edge-UNet outperforms the other models in three performance metrics, which verifies the superior performance of Edge-UNet in rock structural surface trace detection task.}, archive = {J_EAAI}, author = {Xiaofeng Yuan and Dun Wu and Yalin Wang and Chunhua Yang and Weihua Gui and Shuqiao Cheng and Lingjian Ye and Feifan Shen}, doi = {10.1016/j.engappai.2024.109706}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109706}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semantic segmentation model based on edge information for rock structural surface traces detection}, volume = {140}, year = {2025}, }
• (2025). State of health estimation of lithium-ion battery cell based on optical thermometry with physics-informed machine learning. EAAI, 140, 109704. (www)

  @article{ , abstract = {Effective thermal management and accurate state of health (SOH) estimation of lithium-ion batteries is crucial for ensuring their safety, reliability, and longevity. This study presents three innovative physics-informed machine learning-based SOH estimation techniques trained and demonstrated using experimental temperature data. Temperature distribution measurements were obtained using optical frequency domain reflectometry with optical fibers embedded in a cylindrical lithium-ion battery cell under various SOH. One of the trained model accurately predicted the SOH of a cell within 2% with only a 10-minute measurement. This technique also enables the estimation of SOH for individual cells connected in series or parallel within a battery module or pack simultaneously, thereby reducing the overall SOH estimation uncertainty without the need for disassembly. Furthermore, this not only highlights the necessity of precise thermal management in maintaining battery health but also offers a practical and efficient solution for real-time SOH monitoring in battery systems.}, archive = {J_EAAI}, author = {Jeongwoo Jang and Junhyoung Jo and Jinsu Kim and Seungmin Lee and Tonghun Lee and Jihyung Yoo}, doi = {10.1016/j.engappai.2024.109704}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109704}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {State of health estimation of lithium-ion battery cell based on optical thermometry with physics-informed machine learning}, volume = {140}, year = {2025}, }
• (2025). Supporting multi-criteria decision-making processes with unknown criteria weights. EAAI, 140, 109699. (www)

  @article{ , abstract = {Decision support systems are crucial in today’s tech-driven world, assisting decision-makers with complex choices. Determining criteria weights is a paramount aspect, significantly influencing outcomes. Traditionally, criteria weights are derived from objective measures, subjective expert knowledge, or a combination of both, each with its own strengths and limitations. This paper presents a novel approach for addressing unknown criteria relevance by systematically generating weight vectors, thus exploring a broader decision problem space. The proposed methodology is adaptable to various multi-criteria methods, enhancing its applicability across different scenarios. Its effectiveness is empirically validated through two practical examples: Glomerular Filtration Rate (GFR) evaluation and bridge construction method selection, demonstrating its broad applicability. Comparative analysis with existing objective weighting techniques reveals the limitations of current approaches and highlights the improved decision-making capabilities enabled by the proposed method. This research addresses a critical gap in the reliability and robustness of existing methods, particularly in situations with unknown criteria weights. Key contributions include a new decision-making methodology and an innovative ranking formulation using fuzzy sets, with empirical verification strengthening the utility of the approach. This paper offers a promising solution for advancing multi-criteria decision analysis, especially in complex scenarios with uncertain criteria relevance.}, archive = {J_EAAI}, author = {Jakub Więckowski and Wojciech Sałabun}, doi = {10.1016/j.engappai.2024.109699}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109699}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Supporting multi-criteria decision-making processes with unknown criteria weights}, volume = {140}, year = {2025}, }
• (2025). High-order complementary cloud application programming interface recommendation with logical reasoning for incremental development. EAAI, 140, 109698. (www)

  @article{ , abstract = {Cloud application programming interface, as the best carrier for service delivery, data exchange, and capability replication, has been an indispensable element of innovation in today’s app-driven world. However, it is difficult for developers to select the suitable one when facing the sea of cloud application programming interfaces. Existing researches focus on generating single-function and high-quality recommendation lists, while ignoring developers’ needs for high-order complementary cloud application programming interfaces in incremental development. In this paper, we present a high-order complementary cloud application programming interface recommendation with logical reasoning. Firstly, we conduct data analysis to demonstrate the necessity of recommending high-order complementary cloud application programming interfaces and the existence of substitute noise. Secondly, a logical reasoning network is designed using projection, intersection, and negation three logic operators, wherein high-order complementary relations are mined and substitute noises are eliminated. Then, the cloud application programming interface base vector that is complementary but not substitute to the query set is generated, and Kullback–Leibler divergence is subsequently introduced to generate complementary recommendation results. Finally, experimental results demonstrate the superiority of our approach in low-, high-, and hybrid-order complementary recommendation scenarios, and there is a significant increase in hit rate, normalize discounted cumulative gain, mean reciprocal rank, and substitute degree by 11.43%/4.86%, 10.08%/4.28%, 7.50%/2.67%, and 36.33%/32.35% on ProgrammableWeb and Huawei AppGallery datasets respectively. The proposed approach is not only more likely to produce diversified results that meet developers’ needs but also help providers better formulate pricing strategies to achieve combined sales and improve revenue.}, archive = {J_EAAI}, author = {Zhen Chen and Denghui Xie and Xiaolong Wang and Dianlong You and Limin Shen}, doi = {10.1016/j.engappai.2024.109698}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109698}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {High-order complementary cloud application programming interface recommendation with logical reasoning for incremental development}, volume = {140}, year = {2025}, }
• (2025). A large-scale group decision making model with a clustering algorithm based on a locality sensitive hash function. EAAI, 140, 109697. (www)

  @article{ , abstract = {With the development of science and technology, an expanding array of decision-makers across various fields, including engineering and medicine, have been participating in collaborative decision-making for complex scenarios, such as earthquake relief and disease containment. The rapidly changing dynamics of real-world decision-making and the high complexity of consensus reaching among decision-makers require the development of more sophisticated models to handle these challenges. Considering the diversity and stability of group categories, this study proposes a large-scale group decision-making model based on a locality sensitive hash function. First, the volatility of attributes in real scenarios is considered, and a time-series decision matrix is constructed based on the average growth rates to make the results closer to reality. Then, hash functions are used to map the decision opinions to different dimensions and express the similarity through the Hamming distance, yielding clustering results with high stability and cohesion. To determine whether the decision-making group can reach a consensus, this study conducts hypothesis testing, adopting the idea of small probability counterfactuals to provide objective and fair standards for threshold judgment. Finally, through the case study and comparative analysis, it is proved that the proposed method improved 26.4% and 4.2% under the criteria of integrated cohesion and global consensus degree, respectively, with better clustering effect.}, archive = {J_EAAI}, author = {Zhangqian Mu and Yuanyuan Liu and Youlong Yang}, doi = {10.1016/j.engappai.2024.109697}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109697}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A large-scale group decision making model with a clustering algorithm based on a locality sensitive hash function}, volume = {140}, year = {2025}, }
• (2025). A modified multi-agent proximal policy optimization algorithm for multi-objective dynamic partial-re-entrant hybrid flow shop scheduling problem. EAAI, 140, 109688. (www)

  @article{ , abstract = {This paper extends a novel model for modern flexible manufacturing systems: the multi-objective dynamic partial-re-entrant hybrid flow shop scheduling problem (MDPR-HFSP). The model considers partial-re-entrant processing, dynamic disturbance events, green manufacturing demand, and machine workload. Despite advancements in applying deep reinforcement learning to dynamic workshop scheduling, current methods face challenges in training scheduling policies for partial-re-entrant processing constraints and multiple manufacturing objectives. To solve the MDPR-HFSP, we propose a modified multi-agent proximal policy optimization (MMAPPO) algorithm, which employs a routing agent (RA) for machine assignment and a sequencing agent (SA) for job selection. Four machine assignment rules and four job selection rules are integrated to choose optimum actions for RA and SA at rescheduling points. In addition, reward signals are created by combining objective weight vectors with reward vectors, and training parameters under each weight vector are saved to flexibly optimize three objectives. Furthermore, we design an adaptive trust region clipping method to improve the constraint of the proximal policy optimization algorithm on the differences between new and old policies by introducing the Wasserstein distance. Moreover, we conduct comprehensive numerical experiments to compare the proposed MMAPPO algorithm with nine composite scheduling rules and the basic multi-agent proximal policy optimization algorithm. The results demonstrate that the proposed MMAPPO algorithm is more effective in solving the MDPR-HFSP and achieves superior convergence and diversity in solutions. Finally, a semiconductor wafer manufacturing case is resolved by the MMAPPO, and the scheduling solution meets the responsive requirement.}, archive = {J_EAAI}, author = {Jiawei Wu and Yong Liu}, doi = {10.1016/j.engappai.2024.109688}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109688}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A modified multi-agent proximal policy optimization algorithm for multi-objective dynamic partial-re-entrant hybrid flow shop scheduling problem}, volume = {140}, year = {2025}, }
• (2025). A multi-scale patch mixer network for time series anomaly detection. EAAI, 140, 109687. (www)

  @article{ , abstract = {With the development of Internet of Things (IoT) technology, a large amount of data with temporal characteristics is collected and stored. How to efficiently and accurately identify anomalies from these data is a major challenge. At present, there are many problems in the application of anomaly detection, including non-stationary data, complex and difficult-to-collect anomalies, the need for real-time detection and the limitation of computing resources. But few methods can comprehensively consider these issues. To overcome these challenges, we propose a lightweight neural network, Multi-scale Patch Mixer Network (MP-MixerNet). It is mainly composed of a Mixer Block based on fully connected layer design, which contains a Temporal-Mixer and a Spatial-Mixer, and can simultaneously model the intra- and inter-series dependencies of multivariate time series. We also perform multi-scale patch segmentation based on frequency analysis, which helps the model extract robust features from multiple period views. In addition, we design an Input Stabilization module to help the model deal with data distribution shift. Experimental results on a public time series anomaly detection dataset show that we are able to achieve higher comprehensive performance with fewer parameters and inference time.}, archive = {J_EAAI}, author = {Qiushi Wang and Yueming Zhu and Zhicheng Sun and Dong Li and Yunbin Ma}, doi = {10.1016/j.engappai.2024.109687}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109687}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-scale patch mixer network for time series anomaly detection}, volume = {140}, year = {2025}, }
• (2025). Benchmarking neural radiance fields for autonomous robots: An overview. EAAI, 140, 109685. (www)

  @article{ , abstract = {Neural Radiance Field (NeRF) has emerged as a powerful paradigm for scene representation, offering high-fidelity renderings and reconstructions from a set of sparse and unstructured sensor data. In the context of autonomous robotics, where perception and understanding of the environment are pivotal, NeRF holds immense promise for improving performance. However, few survey has discussed such a potential. To fill this gap, we have collected over 200 papers since the publication of original NeRF in 2020 and present a thorough analysis of how NeRF can be used to enhance the capabilities of autonomous robots. We especially focus on the perception, localization and navigation, and decision-making modules of autonomous robots and delve into tasks crucial for autonomous operation, including 3-dimensional reconstruction, segmentation, pose estimation, simultaneous localization and mapping, navigation and planning, and interaction. Our survey meticulously benchmarks existing NeRF-based methods, comparing their reported performance, and providing insights into their strengths and limitations. Moreover, we target the existing challenges of applying NeRF in autonomous robots, including real-time processing, sparse input views, and explore promising avenues for future research and development in this domain. We especially discuss potential of integrating advanced deep learning techniques like 3-dimensional Gaussian splatting, large language models, and generative artificial intelligence. This survey serves as a roadmap for researchers seeking to leverage NeRF to empower autonomous robots, paving the way for innovative solutions that can navigate and interact seamlessly in complex environments.}, archive = {J_EAAI}, author = {Yuhang Ming and Xingrui Yang and Weihan Wang and Zheng Chen and Jinglun Feng and Yifan Xing and Guofeng Zhang}, doi = {10.1016/j.engappai.2024.109685}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109685}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Benchmarking neural radiance fields for autonomous robots: An overview}, volume = {140}, year = {2025}, }
• (2025). Edge artificial intelligence and super-resolution for enhanced weapon detection in video surveillance. EAAI, 140, 109684. (www)

  @article{ , abstract = {The prevalence of crimes involving handguns and knives underscores the importance of early weapon detection. This, along with the spread of video surveillance systems, boosted the development of automatic approaches for weapon detection from surveillance cameras. Despite the advancements from classical computer vision to Deep Learning (DL) techniques, accurately detecting weapons in real-time remains challenging due to their small size. Current DL methods, which attempt to mitigate this issue using complex detection architectures, are resource-intensive, resulting in high costs and energy usage, and hindering their deployment on efficient edge devices. This creates challenges in resource-limited environments, making these methods impractical for edge and real-time applications. To address these shortcomings, our work proposes YOLOSR, which integrates a You Only Look Once (YOLO) v8-small model with an Enhanced Deep Super Resolution (EDSR)-based network using a shared backbone. During training, the auxiliary Super Resolution (SR) helps in learning better features, which could benefit the weapon detection task. During inference, the SR branch is removed, keeping the detector’s computational complexity unchanged. The YOLOSR’s accuracy and efficiency were validated on our WeaponSense dataset and on a NVIDIA Jetson Nano, against other weapon detectors. The results exhibited that YOLOSR, compared to the state-of-the-art YOLOv8-small model, maintained the same computational complexity with 28.8 billion floating point operations and on-device latency of 101 ms per image, while increasing the Average Precision by 10.2 percentage points. Thus, the YOLOSR emerges as an effective solution for real-time weapon detection in resource-constrained environments, achieving an optimal trade-off between efficiency and accuracy.}, archive = {J_EAAI}, author = {Daniele Berardini and Lucia Migliorelli and Alessandro Galdelli and Manuel J. Marín-Jiménez}, doi = {10.1016/j.engappai.2024.109684}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109684}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Edge artificial intelligence and super-resolution for enhanced weapon detection in video surveillance}, volume = {140}, year = {2025}, }
• (2025). Advancing explainability of adversarial trained convolutional neural networks for robust engineering applications. EAAI, 140, 109681. (www)

  @article{ , abstract = {The great success of Convolutional Neural Networks has driven their widespread use in Industry 5.0, especially in safety-critical applications where model robustness is essential. While adversarial training is commonly used to make models more robust, it often changes how the model fuses features, leading to complex trade-offs and making its behavior harder to understand. This lack of transparency contradicts the demand for interpretable and trustworthy artificial intelligence systems in modern industry, where applications like autonomous driving and intelligent medical systems require model interpretability to assign post-accident responsibility and support legal and liability decisions. Our work introduces an innovative approach to enhance the interpretability of adversarially trained Convolutional Neural Networks, providing detailed and quantifiable insights through a six-dimensional semantic framework. This framework, which examines objects, parts, scenes, materials, textures, and colors, offers fine-grained explanations that closely mirror human cognitive processes. We conduct extensive experiments across multiple classical models, and the results reveal that adversarial training induces structure-dependent changes in feature preferences, offering insights into the trade-off between standard and robust accuracy in adversarially trained Convolutional Neural Networks. These findings provide rational support for the use of artificial intelligence in real-world applications and promote the development and implementation of explainable artificial intelligence in modern industries.}, archive = {J_EAAI}, author = {Dehua Zhou and Ziyu Song and Zicong Chen and Xianting Huang and Congming Ji and Saru Kumari and Chien-Ming Chen and Sachin Kumar}, doi = {10.1016/j.engappai.2024.109681}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109681}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advancing explainability of adversarial trained convolutional neural networks for robust engineering applications}, volume = {140}, year = {2025}, }
• (2025). Analysis of interrelationships of human errors using linguistic decision-making trial and evaluation laboratory with consensus reaching process. EAAI, 140, 109676. (www)

  @article{ , abstract = {Analyzing human errors' interrelationships is one of the most important assignments for human reliability improvement in sociotechnical systems. Human factor analysis and classification system (HFACS) is effective in human error analysis due to its taxonomy and systematical perspective. It reveals interrelationships among human errors emerging in a multi-hierarchy of systems. However, the conventional HFACS method is incapable of quantifying their interrelationship. Especially, due to the nature of human errors, their objective data is limited. Experts' opinions are important resources to facilitate human error analysis. However, limited improved HFACS considers experts' consensus on interrelationships analysis results, especially in linguistic environments. Accordingly, this paper aims to address HFACS-based interrelationships analysis problems utilizing linguistic decision-making trial and evaluation laboratory (DEMATEL) with consensus reaching process (CRP). First, probabilistic linguistic terms are utilized to represent experts' opinions on human errors' interrelationships. Second, CRP is introduced to derive consensual opinions on human errors' interrelationships, shifting the focus to identifying human errors with low consensus levels rather than experts. Then, a hybrid weighting method is introduced to determine the weight of experts' opinions in the information fusion phase, which reflects inherent uncertainty and inter-recognition of experts’ opinions. Furthermore, DEMATEL is introduced to model direct and indirect interrelationships among human errors. Finally, a case study of a drug administration process is conducted to validate the efficiency of the proposed method. The case study indicates that neglect of safety culture development and limited financial and human resources are the top two human errors, with importance degree 0.148 and 0.107.}, archive = {J_EAAI}, author = {Qiaohong Zheng and Xinwang Liu}, doi = {10.1016/j.engappai.2024.109676}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109676}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Analysis of interrelationships of human errors using linguistic decision-making trial and evaluation laboratory with consensus reaching process}, volume = {140}, year = {2025}, }
• (2025). Self-supervised anomaly detection and localization for X-ray cargo images: Generalization to novel anomalies. EAAI, 140, 109675. (www)

  @article{ , abstract = {Robust detection of illicit items using X-ray inspection methods has gained increasing importance in recent years due to the large volume of cargo crossing international borders. In addition to detecting the presence of such items, determining their location, size, and shape is challenging due to the unpredictable nature of anomalies, but essential for expediting security inspections. Viewing the illicit items as anomalies relative to expected cargo, we propose a self-supervised learning framework consisting of an encoder–decoder–classifier–segmenter model, a multi-component loss function, coupled with a training strategy to extract discriminative features tailored for detection of the presence of anomalies, as well as localization of such items in X-ray cargo images. Our framework addresses the challenges posed by limited labeled data and offers a model capable of both detecting and localizing anomalies effectively. Moreover, we present a diverse dataset encompassing various cargo scenarios with and without anomalies, providing a robust evaluation environment for this class of problems. Unlike existing approaches, which are trained to detect specific types of objects with a fixed set of illicit items, our framework is adaptable to real-world scenarios where a wide range of illicit items may be present in the cargo. This versatility enhances the practical applicability of our model. We evaluate the performance of our framework on our dataset as well as two other publicly available datasets, demonstrating our method’s strong detection and localization performance even when faced with complex novel anomalies significantly different from those encountered during training.}, archive = {J_EAAI}, author = {Bipin Gaikwad and Abani Patra and Carl R. Crawford and Eric L. Miller}, doi = {10.1016/j.engappai.2024.109675}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109675}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-supervised anomaly detection and localization for X-ray cargo images: Generalization to novel anomalies}, volume = {140}, year = {2025}, }
• (2025). WA-net: Wavelet integrated attention network for silk and bamboo character recognition. EAAI, 140, 109674. (www)

  @article{ , abstract = {Chu Bamboo and Silk ancient Chinese character (CBSC) was originated in the Chu state over 2000 years ago, representing an intermediate script between oracle bone script and seal script. Existing text images have degraded and suffered damage due to their ancient historical origins and insufficient preservation. Due to distinct structural and stroke texture characteristics, significant differences exist between CBSC and contemporary characters, posing challenges for intelligent recognition. Targeting these aforementioned characteristics, we propose a method called Wavelet Integrated Attention Network (WA-Net). This method integrates discrete wavelet transform and attention mechanisms to extract more discriminative features from severe noise interference and degraded text images. Additionally, a dataset named Chu Bamboo and Silk 730 (Chu730) for CBSC recognition has been created due to the lack of publicly available datasets. WA-Net introduces the discrete wavelet attention among layer (L-DWT) to broaden the feature learning space of convolutional neural networks into the wavelet domain, capturing latent information across various frequencies. Subsequently, a wavelet convolution (C-DWT) module is proposed to mitigate the partial information loss of conventional convolution operations. In the W-bneck module, the SE (Squeeze-and-Excitation) attention module and average pooling downsampling are introduced to enhance the extraction of valuable feature maps. Extensive experiments were conducted, including a baseline method that achieved top-1 recognition accuracy of 87.42%. The proposed method achieved an accuracy of 89.27%, and other top-n results also significantly surpassed the baseline accuracy. Other experiment results demonstrate the superiority of the proposed modules and theirvaluable applications in ancient text intelligent recognition and cultural heritage digital preservation. Furthermore, this approach holds significant promise in facilitating the study of other handwritten or ancient characters recognition. Dataset and code are available at: https://github.com/Nancy45-ui/WA-Net .}, archive = {J_EAAI}, author = {Shengnan Li and Chi Zhou and Kaili Wang}, doi = {10.1016/j.engappai.2024.109674}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109674}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {WA-net: Wavelet integrated attention network for silk and bamboo character recognition}, volume = {140}, year = {2025}, }
• (2025). Motion planning for 7-degree-of-freedom bionic arm: Deep deterministic policy gradient algorithm based on imitation of human action. EAAI, 140, 109673. (www)

  @article{ , abstract = {Smart bionic arms have played a great role in returning amputees to society. However, most of the current bionic arms are radial configuration type with few degrees of freedom and humeral form configuration type, which are only applicable to patients with proximal amputation. Patients with shoulder amputation urgently need a 7-degree-of-freedom bionic arm that can fully mimic human upper limb movements. Meanwhile, bionic arms often require specific programming to be implemented for the subject to initially meet the control requirements, which makes it difficult to match the motion of the bionic arm with the wearer's movement habits and reduces the adaptability and reliability of human-computer interaction. To address this problem, this paper proposes a motion imitation based on human upper limb joint point guidance and a motion planning algorithm based on reinforcement learning method to achieve the purpose of making the shoulder disconnected bionic arm accomplish humanoid motion by learning the dynamic motion imitation of the human upper limb. The algorithm analyzes and learns 3D poses of human arm movement features from unlabeled videos, then designs a reward function based on human motion patterns, and uses a reinforcement learning algorithm based on deep deterministic policy gradient (DDPG) to train the humanoid motion of the bionic arm. We evaluated the effectiveness of shoulder detached bionic arms through several tasks in a simulation environment, and the results showed that this method has good performance in planning the humanoid motion of bionic arms and can be widely applied in bionic machine control.}, archive = {J_EAAI}, author = {Baojiang Li and Shengjie Qiu and Haiyan Ye and Yuting Guo and Haiyan Wang and Jibo Bai}, doi = {10.1016/j.engappai.2024.109673}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109673}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Motion planning for 7-degree-of-freedom bionic arm: Deep deterministic policy gradient algorithm based on imitation of human action}, volume = {140}, year = {2025}, }
• (2025). Dual-branch feature reinforcement transformer for preoperative parathyroid gland segmentation. EAAI, 140, 109672. (www)

  @article{ , abstract = {The parathyroid glands are easily injured or accidentally removed during thyroid surgery, causing temporary or even permanent hypocalcemia. Thus, accurate preoperative identification and localization of the parathyroid glands by ultrasound is crucial in protecting the parathyroid glands and preventing parathyroid injury during thyroid surgery. However, there are only a few methods used for highlighting the parathyroid gland in ultrasound images before thyroid surgery. In this study, we propose a Dual-branch feature Reinforcement Transformer Network (DRT-Net) for preoperative parathyroid gland segmentation. DRT-Net incorporates a dual-branch structure, consisting of a devised convolution network (CNN) backbone called Feature Reinforcement subnet (FR-subnet) and a Transformer branch capturing detailed features and context information from the confused ultrasound image. Furthermore, we design a Margin Tracking Attention (MTA) that optimizes the ability of FR-subnet to process margin information by tracking margin pixels of feature map. Finally, we employ a Cross-channel Feature Reinforcement Module (CFRM) to fuse the extracted detailed features from the CNN branch with the global context information from the Transformer branch. We trained and evaluated the DRT-Net on the self-built parathyroid gland segmentation dataset and an open-access Kvasir-SEG dataset. Extensive experiments have been carried out to validate the efficiency of our method.}, archive = {J_EAAI}, author = {Lei Lyu and Chen Pang and Qinghan Yang and Kailin Liu and Chong Geng}, doi = {10.1016/j.engappai.2024.109672}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109672}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-branch feature reinforcement transformer for preoperative parathyroid gland segmentation}, volume = {140}, year = {2025}, }
• (2025). Adaptive neural boundary control for multi-agent manipulators system with uncertainties through cooperative disturbance observers network. EAAI, 140, 109669. (www)

  @article{ , abstract = {This paper addresses vibration control problem of multi-agent flexible manipulators systems in the presence of simultaneous uncertainty and unknown external disturbance. Particularly, the goal is to suppress vibration of both flexible link and joint angular. In this paper, the dynamic model of the considered flexible manipulator is described by the fourth order partial differential equation. Without control, the system is unstable and vibrate constantly due to initial states, the external unknown disturbances and system uncertainties. To compensate the uncertainty in each agent, the neural networks are employed and novel adaptation laws are developed to update weighting parameters in the neural networks. While for the compensation of the external disturbance a cooperative network of disturbance observers is proposed to enhance the observation reliability. With the resulting estimations of uncertainties and the unknown disturbance, adaptive distributed boundary controllers are derived to suppress vibration in-domain and keep joint angular position to zero. The closed-loop system is proven to be uniform ultimately bounded through Lyapunov stability theory. Numerical simulations result shows that compared with the proportional–derivative control, the proposed method almost reduces all overshoot and steady-state error.}, archive = {J_EAAI}, author = {Zhibo Zhao and Yuan Yuan and Xiaodong Xu and Biao Luo and Tingwen Huang}, doi = {10.1016/j.engappai.2024.109669}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109669}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive neural boundary control for multi-agent manipulators system with uncertainties through cooperative disturbance observers network}, volume = {140}, year = {2025}, }
• (2025). Quantum-inspired semantic matching based on neural networks with the duality of density matrices. EAAI, 140, 109667. (www)

  @article{ , abstract = {Social media text can be semantically matched in different ways, viz paraphrase identification, answer selection, community question answering, and so on. The performance of the above semantic matching tasks depends largely on the ability of language modeling. Neural network based language models and probabilistic language models are two main streams of language modeling approaches. However, few prior work has managed to unify them in a single framework on the premise of preserving probabilistic features during the neural network learning process. Motivated by recent advances of quantum-inspired neural networks for text representation learning, we fill the gap by resorting to density matrices, a key concept describing a quantum state as well as a quantum probability distribution. The state and probability views of density matrices are mapped respectively to the neural and probabilistic aspects of language models. Concretizing this state-probability duality to the semantic matching task, we build a unified neural-probabilistic language model through a quantum-inspired neural network. Specifically, we take the state view to construct a density matrix representation of sentence, and exploit its probabilistic nature by extracting its main semantics, which form the basis of a legitimate quantum measurement. When matching two sentences, each sentence is measured against the main semantics of the other. Such a process is implemented in a neural structure, facilitating an end-to-end learning of parameters. The learned density matrix representation reflects an authentic probability distribution over the semantic space throughout the training process. Experiments show that our model significantly outperforms a wide range of prominent classical and quantum-inspired baselines.}, archive = {J_EAAI}, author = {Chenchen Zhang and Qiuchi Li and Dawei Song and Prayag Tiwari}, doi = {10.1016/j.engappai.2024.109667}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109667}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Quantum-inspired semantic matching based on neural networks with the duality of density matrices}, volume = {140}, year = {2025}, }
• (2025). Neural arithmetic logic units with two transition matrix and independent gates. EAAI, 140, 109663. (www)

  @article{ , abstract = {Neural Networks have traditionally been used to handle numerical information based on their training. However, they often struggle with systematic generalization, particularly when the numerical range during testing differs from that used in training. To tackle this issue, we propose an enhanced version of an existing architecture known as Neural Arithmetic Logic Units (NALU), incorporating Independent Gates. We refer to this new architecture as Neural Arithmetic Logic Units with Independent Gates (NALUIG), which can represent numerical values through linear activations. It employs primitive arithmetic operators, managed by learned gates that operate independently of the input, to differentiate weight matrices for both the adder and multiplier. Additionally, we introduce two new architectures: Neural Arithmetic Logic Unit with two Transition Matrices (NALU2M) and Neural Arithmetic Logic Unit with two Transition Matrices and Independent Gates (NALU2MIG). Our experiments demonstrate that the enhanced neural networks can effectively learn to perform arithmetic and numeric image classification from the Modified National Institute of Standards and Technology database (MNIST), achieving significantly lower error rates compared to other existing neural networks. This approach utilizes independent gates to represent numerical values as distinct neurons without introducing non-linearity. In this paper, we present improved results regarding numerical range generalization compared to the current state-of-the-art.}, archive = {J_EAAI}, author = {Sthefanie Jofer Gomes Passo and Vishal H. Kothavade and Wei-Ming Lin and Clair Walton}, doi = {10.1016/j.engappai.2024.109663}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109663}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Neural arithmetic logic units with two transition matrix and independent gates}, volume = {140}, year = {2025}, }
• (2025). Semantic graph neural network with multi-measure learning for semi-supervised classification. EAAI, 140, 109647. (www)

  @article{ , abstract = {Graph Neural Networks (GNNs) have attracted increasing attention in recent years and have achieved excellent performance in semi-supervised node classification tasks. The success of most GNNs is attributed to the availability of the original graph structure. However, recent studies have shown that GNNs are vulnerable to the complex underlying structure of the graph, making it necessary to learn comprehensive and robust graph structures for downstream tasks, rather than relying only on the raw graph structure. In light of this, we seek to learn optimal graph structures for downstream tasks and propose a novel framework for semi-supervised classification. Specifically, based on the structural context information of graph and node representations, we encode the complex interactions in semantics and generate semantic graphs to preserve the global structure. Moreover, we develop a novel multi-measure attention layer to optimize the similarity rather than prescribing it a priori, so that the similarity can be adaptively evaluated by integrating measures. These graphs are fused and optimized together with GNN towards semi-supervised classification objective. Extensive experiments and ablation studies on six real-world datasets clearly demonstrate the effectiveness of our proposed model and the contribution of each component. The proposed model not only addresses the inherent vulnerabilities of GNNs to complex graph structures, but also introduces a pioneering approach to learning comprehensive and robust graph representations for semi-supervised classification tasks.}, archive = {J_EAAI}, author = {Junchao Lin and Yuan Wan and Jingwen Xu and Xingchen Qi}, doi = {10.1016/j.engappai.2024.109647}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109647}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semantic graph neural network with multi-measure learning for semi-supervised classification}, volume = {140}, year = {2025}, }
• (2025). Exploring structural components in autoencoder-based data clustering. EAAI, 140, 109562. (www)

  @article{ , abstract = {Clustering is a fundamental machine-learning task that has received extensive popularity in the literature. The foundational tenet of traditional clustering approaches is that data are learned to be vectorized features through various representational learning techniques. The conventional clustering methods can no longer manage the high-dimensional data as the data gets more intricate. Numerous representational learning strategies using deep architectures have been presented over the years, particularly deep unsupervised learning due to its superiority over conventional approaches. In most existing research, especially in the autoencoder-based approaches, only the distance information of pair-of-points in the original data space is retained in the latent space. However, combining this with additional preserving factors like the variance and independent component in the original data and latent space, respectively, is important. In addition, the model’s stability under noisy conditions is crucial. This paper provides a unique method for clustering data that combines autoencoder (AE), principal component analysis (PCA), and independent component analysis (ICA) to capture a relevant latent space representation. A further aid in lowering the dimensionality to improve clustering effectiveness is employing two-dimensional reduction algorithms, i.e., PCA and t − distributed stochastic neighbor embedding ( t − SNE). The proposed technique produces more precise and reliable clustering by utilizing the advantages of both approaches. To compare the efficiency of the proposed methods to conventional clustering methods and stand-alone autoencoders, we conduct comprehensive experiments on 13 real-life datasets. The outcomes demonstrate the approach’s intriguing potential for addressing complicated clustering problems, and importantly, effectiveness is demonstrated even under noisy conditions.}, archive = {J_EAAI}, author = {Sujoy Chatterjee and Suvra Jyoti Choudhury}, doi = {10.1016/j.engappai.2024.109562}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109562}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Exploring structural components in autoencoder-based data clustering}, volume = {140}, year = {2025}, }
• (2025). Low-cost language models: Survey and performance evaluation on python code generation. EAAI, 140, 109490. (www)

  @article{ , abstract = {Large Language Models (LLMs) have become a popular choice for many Natural Language Processing (NLP) tasks due to their versatility and ability to produce high-quality results. Specifically, they are increasingly used for automatic code generation to help developers tackle repetitive coding tasks. However, LLMs’ substantial computational and memory requirements often make them inaccessible to users with limited resources. This paper focuses on very low-cost models which offer a more accessible alternative to resource-intensive LLMs. We notably: (1) propose a thorough semi-manual evaluation of their performance in generating Python code, (2) introduce a Chain-of-Thought (CoT) prompting strategy to improve model reasoning and code quality, and (3) propose a new dataset of 60 programming problems, with varied difficulty levels, designed to extend existing benchmarks like HumanEval and EvalPlus. Our findings show that some low-cost compatible models achieve competitive results compared to larger models like ChatGPT despite using significantly fewer resources. We will make our dataset and prompts publicly available to support further research.}, archive = {J_EAAI}, author = {Jessica López Espejel and Mahaman Sanoussi Yahaya Alassan and Merieme Bouhandi and Walid Dahhane and El Hassane Ettifouri}, doi = {10.1016/j.engappai.2024.109490}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109490}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Low-cost language models: Survey and performance evaluation on python code generation}, volume = {140}, year = {2025}, }
• (2025). FTPComplEx: A flexible time perspective approach to temporal knowledge graph completion. EAAI, 139, 109717. (www)

  @article{ , abstract = {The dynamic nature of interconnected data evolving over time poses significant challenges for graph representation and reasoning, particularly as temporal knowledge graphs scale in size and complexity. Existing models like TPComplEx (Time Perspective Complex Embedding) leverage tensor decomposition techniques to capture temporal dynamics, but their static weighting approach often lacks the flexibility needed to adapt to the nuanced evolution of relationships and entities. This rigidity can lead to missed temporal dependencies and loss of valuable insights, especially in large-scale graphs comprising millions or even billions of factual entries. To overcome these limitations, we propose FTPComplEx (Flexible Time Perspective Complex Embedding), a novel embedding model that introduces adjustable weights to dynamically modulate the influence of temporal information. This flexibility enables FTPComplEx to more accurately capture the intricate interactions between entities, relations, and time, providing a more robust understanding of temporal dynamics within knowledge graphs. Our extensive evaluations on benchmark datasets, including YAGO15k, ICEWS, and GDELT, demonstrate that FTPComplEx achieves state-of-the-art results, outperforming TPComplEx and other existing models. Notably, on the YAGO15k dataset, FTPComplEx achieves a 9.04% improvement in Mean Reciprocal Rank (MRR) and an 11.35% increase in Hits@1, demonstrating its effectiveness in managing complex temporal relationships. Further analysis shows that FTPComplEx maintains strong performance even with lower-rank embeddings, significantly reducing computational costs while maintaining accuracy.}, archive = {J_EAAI}, author = {Ngoc-Trung Nguyen and Thuc Ngo and Nguyen Hoang and Thanh Le}, doi = {10.1016/j.engappai.2024.109717}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109717}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {FTPComplEx: A flexible time perspective approach to temporal knowledge graph completion}, volume = {139}, year = {2025}, }
• (2025). A knowledge-refined hybrid graph model for quality prediction of industrial processes. EAAI, 139, 109711. (www)

  @article{ , abstract = {The complexity of industrial processes has spurred the application of soft sensor techniques for predicting key quality variables based on easy-measurable process variables. Currently, data-driven soft sensors based on Artificial Intelligence techniques have become the mainstream. However, these soft sensing models deeply rely on the quality of training data, where the domain knowledge is often ignored. Meanwhile, a significant amount of labeled data is not fully utilized. To address these issues, this paper proposes a supervised framework based on a knowledge-refined hybrid graph network, which contributes to the artificial intelligence application of nonlinear dynamic soft sensors. The problems of applying traditional artificial intelligence models in soft sensor have been addressed by reconstructing the input module of graph neural networks with knowledge-guided approaches. Both spatial and temporal correlations of process data are captured and the hybrid network significantly improves the reliability and interpretability of the soft sensing model. By incorporating labeled data into the model, the representation of quality information is also enhanced. Finally, the proposed framework was applied to an industrial debutanizer column, and the experimental results fully demonstrate the effectiveness and superiority of the method.}, archive = {J_EAAI}, author = {Yang Wang and Feifan Shen and Lingjian Ye}, doi = {10.1016/j.engappai.2024.109711}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109711}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A knowledge-refined hybrid graph model for quality prediction of industrial processes}, volume = {139}, year = {2025}, }
• (2025). A depthwise convolutional neural network model based on active contour for multi-defect wafer map pattern classification. EAAI, 139, 109707. (www)

  @article{ , abstract = {As semiconductor manufacturing processes continue to witness increased integration density and design complexity, semiconductor wafers are experiencing a growing diversity and complexity of defects. While previous research in wafer map classification using deep learning has made significant advancements in dealing with single defect patterns, the classification of mixed-type defects has received less attention due to their considerably higher difficulty level compared to single defects. This research addresses this critical gap, emphasizing the need for improved methods to classify mixed-type defects, which are more complex and challenging. To tackle this challenge, this paper introduces the active contour-based lightweight depthwise network (AC-LDN) model for the classification of multi-defect wafer map patterns. Initially, multi-defect features are extracted using an active contour-based segmentation model. Subsequently, the learning model employs a depthwise convolutional neural network (CNN) architecture that combines separable CNN and dilated CNN techniques. This unique approach optimizes the model in the separable segment while effectively addressing defect complexity in the depthwise segments. Consequently, AC-LDN outperforms other state-of-the-art models, offering a balance between lightweight characteristics and high accuracy. The proposed method demonstrates its superiority over previous models when evaluated on the extsdsensive multi-wafer map dataset, achieving an average classification accuracy exceeding 98% and a confusion matrix coefficient surpassing 0.97.}, archive = {J_EAAI}, author = {Jeonghoon Choi and Dongjun Suh}, doi = {10.1016/j.engappai.2024.109707}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109707}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A depthwise convolutional neural network model based on active contour for multi-defect wafer map pattern classification}, volume = {139}, year = {2025}, }
• (2025). The nexus of intelligent transportation: A lightweight bi-input fusion detection model for autonomous-rail rapid transit. EAAI, 139, 109705. (www)

  @article{ , abstract = {Autonomous-rail Rapid Transit (ART) possesses various advantages in intelligent transportation, but it does not effectively recognize road conditions caused solely by deploying single-modal cameras. In this paper, a lightweight fusion-based object detection neural network was designed with multi-modal sensors for the ART. Firstly, the Light Detection and Ranging (LiDAR) applied additional encoding and preprocessing to the point cloud. Secondly, a backbone and a detection head of the network structure were proposed through re-parameterization and pruning techniques. Furthermore, a fusion module was designed with a selective soft attention mechanism to fuse the extracted features. The proposed model was tested on the open autonomous driving dataset; it achieved a 7.38% improvement in mean average precision (mAP) compared to the original you only look once (YOLO) as well as other state-of-the-art (SOTA) models. Finally, practical experiments were conducted in the maintenance center of ART to simulate the operational scenarios and validate the feasibility of the proposed method in this study. By fully utilizing the information in different modalities and addressing the limitations of single-modal recognition, efforts were made to improve the robustness of road object detection for ART under different road conditions. Consequently, our method provides effective solutions which benefit intelligent transportation with advanced algorithms and strategies.}, archive = {J_EAAI}, author = {Hongjie Tang and Jirui Wang and Jiaoyi Wu and Yanni Zhao and Jiangfan Chen and Fujian Liang and Zutao Zhang}, doi = {10.1016/j.engappai.2024.109705}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109705}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The nexus of intelligent transportation: A lightweight bi-input fusion detection model for autonomous-rail rapid transit}, volume = {139}, year = {2025}, }
• (2025). Informer learning framework based on secondary decomposition for multi-step forecast of ultra-short term wind speed. EAAI, 139, 109702. (www)

  @article{ , abstract = {Accurate and dependable wind speed prediction holds paramount importance in facilitating the dispatch and safe operation of power systems. Nonetheless, the inherent instability of wind speed makes wind speed prediction challenging. Consequently, a short-term wind speed prediction framework, amalgamating secondary decomposition (SD)-Informer, has been proposed in this paper. Initially, the variational mode decomposition (VMD) is applied to decompose the primary wind speed sequence. Through the VMD feature decomposition module, it effectively filters and eliminates superfluous noise from wind speed data. Subsequently, the complete ensemble empirical mode decomposition with adaptive noise technique is introduced for a secondary decomposition targeting the high-frequency components derived from the initial decomposition. To address the limitation of neural network models in capturing essential information from lengthy sequential data concurrently, a predictive model based on Informer is proposed as wind speed prediction module, thereby enhancing prediction accuracy. The validation of this hybrid model encompasses four distinct time ranges. Multiple models are scrutinized through comparative analysis to ascertain the superior performance of the proposed hybrid model. The root mean square error of the proposed method is reduced by 33.02%、25.46%、24.26%, and 23.12% compared to gate recurrent unit (GRU), vision Transformer (ViT), attention (AT)-ViT, and CNN-atteneion (CA)-Bi-directional long short-term memory (BiLSTM) respectively. The mean absolute error of the proposed method in the first quarter is 0.432, with model comparison values reduction of 36.19%、22.99%、20.44%, and17.71% respectively. The experimental results indicate that the proposed model exhibits a strong capability in capturing the long-term dependencies between the input and output sequences of wind speed. It can perform multi-step predictions while ensuring high prediction accuracy.}, archive = {J_EAAI}, author = {Zihao Jin and Xiaomengting Fu and Ling Xiang and Guopeng Zhu and Aijun Hu}, doi = {10.1016/j.engappai.2024.109702}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109702}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Informer learning framework based on secondary decomposition for multi-step forecast of ultra-short term wind speed}, volume = {139}, year = {2025}, }
• (2025). Machine learning for predicting maximum displacement in soil-pile-superstructure systems in laterally spreading ground. EAAI, 139, 109701. (www)

  @article{ , abstract = {Extensive damage to pile-supported structures, often caused by earthquake-induced lateral spreading, has been reported frequently in numerous major earthquakes. To mitigate such damage, accurate prediction of the seismic behavior of the soil-pile-superstructure system (SPSS) has been extensively studied through experimental and numerical simulations. However, these methods typically require substantial time and high cost, making them challenging to adapt in practical engineering scenarios. This study successfully applied machine learning (ML) techniques to predict the maximum seismic response of the SPSS, offering a more efficient and flexible solution for engineers. Six ML algorithms were used: decision tree (DT), k-nearest neighbor (KNN), extreme gradient boosting (XGB), random forest (RF), artificial neural network (ANN), and Gaussian process regression (GPR). A detailed evaluation of these algorithms has shown that ML models can effectively predict the maximum displacement of both pile and soil. Notably, XGB outperformed other methods in terms of accuracy, stability, and efficiency. Furthermore, the study indicates that the velocity-dependent ground motion parameter, root mean square velocity ( v RMS ), effectively represents the ground motion parameters for accurately predicting maximum pile-soil displacement. This study demonstrates the potential of ML in geotechnical earthquake engineering, establishing a basis for further applications and contributing to enhanced seismic design of pile-supported structures in liquefiable soils.}, archive = {J_EAAI}, author = {Xiaoyu Zhang and Desheng He and Junjie Wang and Shengkun Wang and Meixiang Gu}, doi = {10.1016/j.engappai.2024.109701}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109701}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning for predicting maximum displacement in soil-pile-superstructure systems in laterally spreading ground}, volume = {139}, year = {2025}, }
• (2025). The multi-depot pickup and delivery vehicle routing problem with time windows and dynamic demands. EAAI, 139, 109700. (www)

  @article{ , abstract = {The rapid development of the urban logistics recycling industry, combined with the complexity of the pickup and delivery networks, has created a surge in dynamic customer demands and exacerbated the difficulty of logistics resource sharing. Accordingly, this work focuses on a multi-depot pickup and delivery vehicle routing problem with time windows and dynamic demands, which incorporates resource sharing. A bi-objective mathematical model is formulated to minimize the total operating cost and number of vehicles. A three-dimensional affinity propagation clustering and an adaptive nondominated sorting genetic algorithm-II are combined to find Pareto optimal solutions. A dynamic demand insertion strategy is proposed to determine the vehicle service sequences for dynamic situations. Combined with an elite iteration mechanism to prevent the proposed algorithm from falling into search stagnation and improve the convergence performance. The superiority of the proposed algorithm is verified by comparing with CPLEX solver (i.e., ILOG CPLEX Optimization Studio 12.10), multi-objective ant colony optimization, multi-objective particle swarm optimization, multi-objective evolutionary algorithm, multi-objective genetic algorithm, and decomposition-based multi-objective evolutionary algorithm with tabu search. Besides, the proposed model and algorithm are tested by a real-world case study in Chongqing city, China, and the further analysis indicates that significant improvement can be achieved. Furthermore, by incorporating the recognition and prediction techniques of artificial intelligence on dynamic demand data, the proposed approach can realize the self-optimization of multi-depot vehicle routes and the precise allocation of logistics resources in dynamic environments. This study is conducive to the construction of a digitally-intelligent urban logistics system.}, archive = {J_EAAI}, author = {Yong Wang and Mengyuan Gou and Siyu Luo and Jianxin Fan and Haizhong Wang}, doi = {10.1016/j.engappai.2024.109700}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109700}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The multi-depot pickup and delivery vehicle routing problem with time windows and dynamic demands}, volume = {139}, year = {2025}, }
• (2025). Chimney detection and size estimation from high-resolution optical satellite imagery using deep learning models. EAAI, 139, 109686. (www)

  @article{ , abstract = {This study shows an efficient method to estimate the location and size of chimneys from high-resolution satellite optical images using deep learning models. Korean multi-purpose satellite (KOMPSAT) −3 and -3A satellite images with spatial resolutions of 0.7 m and 0.55 m were used for model performance estimation, and the You Only Look Once version 8 (YOLOv8) and Residual Network (ResNet) regression models were integrated for the detection and size estimation of the chimneys. In the chimney detection and size estimation, we compared the model performances between 1) imbalanced and balanced data, 2) South Korea and Thailand data, and 3) KOMPSAT-3 and -3A data. We also analyzed the model performance according to the ResNet convolutional layers in chimney size estimation. In chimney detection, the model performances between the imbalanced and balanced data, South Korea and Thailand data, and KOMPSAT-3 and -3A data were about 0.723 and 0.739, 0.674 and 0.805, and 0.702 and 0.786 in the average precision (AP) 50–95 measure, respectively. The model performance between the South Korea and Thailand data showed a significant difference, likely because the chimneys in South Korea are very diverse, making it harder to generalize the YOLOv8 model. Furthermore, the model root mean square errors (RMSE) between the imbalanced and balanced data, South Korea and Thailand data, and KOMPSAT-3 and -3A data were about 2.917 and 2.788, 2.690 and 2.951, and 2.913 and 2.580 in chimney height, respectively, and about 1.285 and 1.190, 1.228 and 1.120, and 1.291 and 1.013 in chimney diameter, respectively. Keywords: Chimneys; deep learning; You Only Look Once version 8; Residual Network; Korean Multi-purpose Satellite-3/3A; object detection; regression model.}, archive = {J_EAAI}, author = {Che-Won Park and Hyung-Sup Jung and Won-Jin Lee and Kwang-Jae Lee and Kwan-Young Oh and Joong-Sun Won}, doi = {10.1016/j.engappai.2024.109686}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109686}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Chimney detection and size estimation from high-resolution optical satellite imagery using deep learning models}, volume = {139}, year = {2025}, }
• (2025). Semi-supervised segmentation model for crack detection based on mutual consistency constraint and boundary loss. EAAI, 139, 109683. (www)

  @article{ , abstract = {Detecting and measuring cracks are crucial for ensuring the safety of civil infrastructures. Traditional fully supervised methods require an amount of high-precision labeled data, making them time-consuming to deploy. In this paper, a semi-supervised learning network model designed for crack segmentation is proposed to address this issue, which incorporates a mutual consistency constraint and a boundary loss function. The mutual consistency constraint enables the model to utilize information from unlabeled data, thereby improving its performance and efficacy. Meanwhile, the boundary loss function enhances the model's ability to predict images when the background pixels outnumber the crack pixels. To comprehensively evaluate the model performance, a new dataset featuring various environmental interferences is constructed. The model optimal hyperparameters and architecture are determined through the experiments and an ablation study. To highlight the advantages of the proposed network model, its prediction results are compared with other segmentation models. It is demonstrated that the proposed model delivers high-precision and robust results while requiring less labeled data.}, archive = {J_EAAI}, author = {Tianxiang Shi and Yangyang Wang and Yu Fang and Yongqiang Zhang}, doi = {10.1016/j.engappai.2024.109683}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109683}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Semi-supervised segmentation model for crack detection based on mutual consistency constraint and boundary loss}, volume = {139}, year = {2025}, }
• (2025). The predictability of the effectiveness of chains of classifiers in the out-of-domain detection. EAAI, 139, 109682. (www)

  @article{ , abstract = {Models in machine learning (ML) can produce erroneous results when exposed to the out-of-domain (OOD) data. Thus the problem of correctly incorporating the detection of OOD samples should be considered for all production systems that use the ML component. The production system can incorporate the filtering of OOD in three ways: pre-filtering data, filtering it in-model, and posteriori filtering, with the option to chain them into larger ensembles. Considering the number of possible types of filters, as well the length of the ensemble, it is important to be able to predict the upper and the lower bound of the performance of such ensembles on the basis of the performance of individual filters. The paper analyses challenges associated with the construction of the model, and then proposes the model where the performance of any chain of filters can be predicted, without actually assembling and testing it. An experiment is used to demonstrate the feasibility of the model, using the simple digit classification task, and F1 as the metric, resulting in the prediction error well below 0.1.}, archive = {J_EAAI}, author = {Piotr Cofta}, doi = {10.1016/j.engappai.2024.109682}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109682}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The predictability of the effectiveness of chains of classifiers in the out-of-domain detection}, volume = {139}, year = {2025}, }
• (2025). A collaborative surface target detection and localization method for an unmanned surface vehicle swarm. EAAI, 139, 109679. (www)

  @article{ , abstract = {A single unmanned surface vehicle (USV) designed for marine missions suffers from limited payload, low efficiency and weak intelligence, while a swarm of USVs shows significant advantages in mission flexibility, diverse payload and task efficiency. One of the key issues for an USV swarm is how to achieve highly efficient collaborative perception. To address this issue, a method framework of collaborative surface target detection and localization based on multiple sensors for a swarm including 4 USVs is designed. First, perception systems are constructed, a joint calibration method for different sensors is proposed, and a lightweight target detection method improved with attention mechanism and lightweight adaptive spatial feature fusion is designed. Second, a specialized fusion method using sensor principles based on an extended Kalman filter (EKF) is proposed for a single USV to obtain a target state model. Third, the obtained target models from different USVs are registered with fuzzy matching and integrated into the complete model in a geographic coordinate system. The proposed method is applied to the collaborative perception system on our developed 4 USV swarm and verified in real marine environment and simulation. Experimental results show that our proposed method framework significantly improves the accuracy, efficiency, and reliability of the target detection and localization. The proposed LAF-YOLOv8-s reduces the model size by 5.1M, while the mean average precision (mAP) reaches 68.7%, which is significantly superior to other methods. The average collaborative localization error is reduced by 2.9m. The dataset is available at https://github.com/maochenyu1/WSLight .}, archive = {J_EAAI}, author = {Bo Wang and Chenyu Mao and Kaixin Wei and Xueyi Wu and Ye Li}, doi = {10.1016/j.engappai.2024.109679}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109679}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A collaborative surface target detection and localization method for an unmanned surface vehicle swarm}, volume = {139}, year = {2025}, }
• (2025). Domain adaptation based automatic identification method of vortex induced vibration of long-span bridges without prior information. EAAI, 139, 109677. (www)

  @article{ , abstract = {Machine learning algorithms can sensitively capture the characteristics of vortex induced vibration (VIV) of the girder in long span bridge from the extensive historical data accumulated by structural health monitoring (SHM) system over several years. These algorithms have gradually become a promising method of VIV identification. However, the algorithms proposed by previous researchers require historical VIV data to select the threshold or parameters to identify VIV. Most long-span bridges have not recorded a significant amount of VIV data since VIV is rare, or the bridge were not equipped with SHM system before. This study proposes an adaptive VIV identification method based on domain adaptation methods, which can identify VIV in real-time or in historical monitoring datasets of the target bridge without prior VIV information or parameter settings. The strong generalization ability of the proposed method is verified on the SHM dataset of two long-span suspension bridges in China. It is found that the VIV recognition accuracy of the balanced distribution adaptation (BDA) based VIV identification method is higher than that of other algorithms. In this study, the BDA based algorithm is also applied to the 8 months monitoring datasets of a long span bridge and successfully identifies more than 20 VIV events of the main girder, which has shown the stability and accuracy of the proposed algorithm.}, archive = {J_EAAI}, author = {Chunfeng Wan and Jiale Hou and Guangcai Zhang and Shuai Gao and Youliang Ding and Sugong Cao and Hao Hu and Songtao Xue}, doi = {10.1016/j.engappai.2024.109677}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109677}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Domain adaptation based automatic identification method of vortex induced vibration of long-span bridges without prior information}, volume = {139}, year = {2025}, }
• (2025). Dynamic localization based-utility decision approach under type-2 pythagorean fuzzy set for developing internet of modular self-reconfiguration robot things. EAAI, 139, 109671. (www)

  @article{ , abstract = {The Internet of Modular Robot Things (IoMRT) has emerged through the integration of robotic systems into the Internet of Things (IoT), offering a wide range of solutions to meet continuously growing demands. Six self-reconfiguration functionalities/criteria have been proposed for developing IoMRT. However, no study has fully developed an IoMRT that satisfies all the necessary functionalities. Additionally, there is a lack of scholarly research proposing a decision-based approach for evaluating and ranking IoMRT, which highlights a significant research gap. A complex multiple-criteria decision-making (MCDM) problem has arisen in evaluating and ranking IoMRT due to the diversity of functionalities, the need to prioritize these functionalities based on their importance, and data variability. To address this issue, the study proposes a novel decision-based approach for evaluating and ranking IoMRT, which consists of three phases: (i) Developing a novel weighting method called T2PFS-FWZICbIP (Type-2 Pythagorean Fuzzy Set–Fuzzy Weighted Zero Inconsistency based on Interrelationship Process) to measure the importance of the identified functionalities; (ii) Formulating a decision matrix by cross-referencing potential IoMRT developments with the six self-reconfiguration functionalities resulted in the selection of a random sample of 50 IoMRTs as proof of concept. Following this, the DLbU (Dynamic Localization-based Utility) method was proposed, integrating dynamic localization and utility procedures to manage binary data within the decision matrix; (iii) Developing a novel ranking method, T2PFS-DNMA (Type-2 Pythagorean Fuzzy Set–Double Normalization-based Multiple Aggregation), to address the diversity of functionalities and concerns regarding data variance. The results revealed that the Distributed functionality (C1) received the highest weight value of 0.4060 according to T2PFS-FWZICbIP, indicating its high importance in the ranking of IoMRT. In contrast, the High-Fidelity functionality (C5) received a weight value of 0.0733, indicating its very low importance in the ranking. IoMRT2 and IoMRT35 were identified as the most and least favored, respectively, according to T2PFS-DNMA. The robustness of the proposed approach was assessed through sensitivity analysis and comparative studies.}, archive = {J_EAAI}, author = {Nahia Mourad and A.A. Zaidan and Hassan A. Alsattar and Sarah Qahtan and B.B. Zaidan and Muhammet Deveci and Dragan Pamucar and Witold Pedrycz}, doi = {10.1016/j.engappai.2024.109671}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109671}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic localization based-utility decision approach under type-2 pythagorean fuzzy set for developing internet of modular self-reconfiguration robot things}, volume = {139}, year = {2025}, }
• (2025). Enhanced cross-domain lithology classification in imbalanced datasets using an unsupervised domain adversarial network. EAAI, 139, 109668. (www)

  @article{ , abstract = {Recent advancements in Artificial Intelligence (AI), particularly deep learning, have significantly improved lithology identification in reservoir exploration by leveraging micrographic rock imagery. Deep neural networks excel in feature extraction, enhancing classification accuracy. However, these models are prone to domain shifts, which often degrade their performance in real-world applications. This paper proposes an unsupervised domain adaptation framework that integrates Fisher linear discriminant analysis and Online Hard Example Mining (OHEM) to mitigate domain shifts and improve classification, particularly in datasets with imbalanced classes. The model employs a ω -balanced global–local domain discriminator to align feature distributions between different domains and introduces focal loss with class-wise weighted factors for better handling of imbalanced data. Additionally, an adapted version of OHEM identifies difficult samples during training, allowing the model to concentrate on challenging cases. The proposed method is validated on micrographic rock imagery from the Tibet, Qinghai, and Xinjiang regions, achieving an average accuracy of 83.2%, which is 13.8% higher than ResNet50 and at least 1% superior to other domain adaptation models. This research highlights the potential of AI-driven solutions in geoscientific applications and provides a robust framework for unsupervised lithology classification.}, archive = {J_EAAI}, author = {Yunxin Xie and Liangyu Jin and Chenyang Zhu and Weibin Luo and Qian Wang}, doi = {10.1016/j.engappai.2024.109668}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109668}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced cross-domain lithology classification in imbalanced datasets using an unsupervised domain adversarial network}, volume = {139}, year = {2025}, }
• (2025). Fine-tuning pre-trained networks with emphasis on image segmentation: A multi-network approach for enhanced breast cancer detection. EAAI, 139, 109666. (www)

  @article{ , abstract = {Accurate classification of mammography images into normal and cancerous categories is critical for the early detection of breast cancer. This study utilizes transfer learning and deep learning models to extract and diversify features from a combined dataset consisting of the Mini Digital Database for Screening Mammography (DDSM, containing 7808 images) and the Mammographic Image Analysis Society (MIAS) dataset (containing 322 images). The preprocessing steps involve image cropping, removal of artifacts, and enhancement of contrast using Contrast-Limited Adaptive Histogram Equalization (CLAHE). Data augmentation techniques, including the application of median and Gaussian blur, were used to improve the robustness of the models. Three pre-trained networks—Residual Networks with 50 layers (ResNet-50), Visual Geometry Group Network with 19 layers (VGG-19), and Residual Networks with 152 layers Version 2 (ResNet-152V2)—were fine-tuned specifically for mammography data. Image segmentation and the removal of the pectoral muscle significantly improved classification accuracy. The VGG-19 model achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.80 for segmented images and 0.86 for non-segmented images. A stacked generalization model, which combined features from all three networks, further optimized performance. Artificial Neural Network (ANN) and Extreme Gradient Boosting (XGBoost) models achieved AUCs of 0.897 and 0.890, respectively, for segmented images. Data augmentation improved performance by 2.7%–4%.}, archive = {J_EAAI}, author = {Parviz Ghafariasl and Masoomeh Zeinalnezhad and Shing Chang}, doi = {10.1016/j.engappai.2024.109666}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109666}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fine-tuning pre-trained networks with emphasis on image segmentation: A multi-network approach for enhanced breast cancer detection}, volume = {139}, year = {2025}, }
• (2025). Attention based network for fusion of polarimetric and contextual features for polarimetric synthetic aperture radar image classification. EAAI, 139, 109665. (www)

  @article{ , abstract = {Polarimetric synthetic aperture radar (PolSAR) images containing polarimetric, scattering and contextual features are useful radar data for ground surface classification. Appropriate feature extraction and fusion by using a small set of available labeled samples is an important and challenging task. Several transformers with self-attention mechanism have recently achieved great success for PolSAR image classification. While almost all methods just exploit the self-attention features from the PolSAR cube, the feature fusion method proposed in this work, which is called attention based scattering and contextual (ASC) network, utilizes the polarimetric self-attention beside two cross-attention blocks. The cross-attention blocks extract the polarimetric-scattering dependencies and polarimetric-contextual interactions, individually. The proposed ASC network uses three inputs: the PolSAR cube, the scattering feature maps obtained by clustering of the entropy-alpha features, and the segmentation maps obtained by a super-pixel generation algorithm. The features extracted by self- and cross-attention blocks are fused together, and the residual learning improves the feature learning. While transformers and attention-based networks usually need large training sets, the proposed ASC network shows high efficiency with relatively low number of training samples in various real and synthetic PolSAR images. For example, in the Flevoland PolSAR image containing 15 classes acquired by AIRSAR in L-band, with using 100 training samples per class (less than 1% of labeled samples), the ASC network achieves the overall accuracy of 99.51, which is statistically preferred than the self-attention-based network according to the McNemars test.}, archive = {J_EAAI}, author = {Maryam Imani}, doi = {10.1016/j.engappai.2024.109665}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109665}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Attention based network for fusion of polarimetric and contextual features for polarimetric synthetic aperture radar image classification}, volume = {139}, year = {2025}, }
• (2025). Predicting rapid impact compaction of soil using a parallel transformer and long short-term memory architecture for sequential soil profile encoding. EAAI, 139, 109664. (www)

  @article{ , abstract = {This study presents an advanced deep learning approach for predicting the effectiveness of Rapid Impact Compaction (RIC). The model integrates the focused attention mechanisms of transformer architectures with the sequential data processing capabilities of Long Short-Term Memory (LSTM) networks. Input parameters include the initial soil profile and feature vectors representing the soil's initial state, applied compaction effort, and compaction hammer energy. Utilizing an encoder-decoder framework, the model encodes soil profile information at various depths into tokens, which are subsequently decoded to predict the resulting ground improvement. An ablation study was conducted to assess the significance of each model component. The model's predictive accuracy was validated using field test data, demonstrating a strong correlation with observed outcomes (mean absolute error of 0.42 for test data). Shapley value analysis of the trained model revealed that compaction effort exerted the highest influence on predictions, followed by fine content and fill thickness. The model architecture also demonstrated successful application to alternative RIC case studies, indicating potential generalizability. Furthermore, the model's capability to simulate hypothetical scenarios with varying compaction efforts provides valuable insights for strategic planning and optimization of RIC project designs.}, archive = {J_EAAI}, author = {Sompote Youwai and Sirasak Detcheewa}, doi = {10.1016/j.engappai.2024.109664}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109664}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Predicting rapid impact compaction of soil using a parallel transformer and long short-term memory architecture for sequential soil profile encoding}, volume = {139}, year = {2025}, }
• (2025). Workload-based adaptive decision-making for edge server layout with deep reinforcement learning. EAAI, 139, 109662. (www)

  @article{ , abstract = {Mobile edge computing (MEC) is crucial in applications such as intelligent transportation, innovative healthcare, and smart cities. By deploying servers with computing and storage capabilities at the network edge, MEC enables low-latency services close to end users. However, the configuration of edge servers needs to meet the low-latency requirements and effectively balance the servers’ workloads. This paper proposes an adaptive layout and dynamic optimization method, modeling the edge server layout problem as a Markov decision process. It introduces a workload-based server placement rule that adjusts the locations of edge servers according to the load of base stations, enabling the learning of low-latency and load-balanced server layout strategies. Experimental validation on a real dataset from Shanghai Telecom shows that the proposed algorithm improves average latency performance by about 40% compared to existing technologies, and enhances workload balancing performance by about 17%.}, archive = {J_EAAI}, author = {Shihua Li and Yanjie Zhou and Bing Zhou and Zongmin Wang}, doi = {10.1016/j.engappai.2024.109662}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109662}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Workload-based adaptive decision-making for edge server layout with deep reinforcement learning}, volume = {139}, year = {2025}, }
• (2025). Prediction and evaluation of key parameters in coalbed methane pre-extraction based on transformer and inversion model. EAAI, 139, 109661. (www)

  @article{ , abstract = {Accurate parameter prediction in the coalbed methane (CBM) pre-extraction process is crucial for formulating effective control measures and preventing CBM-related accidents. Traditional prediction methods rely on feature extraction or complex physical model parameter calculations, which require extensive manual intervention and have limited practical applicability. Additionally, simple neural network methods are prone to overfitting and gradient vanishing when handling parameters, and they lack the capability to dynamically monitor gas pressure during extraction, leading to inefficient and blind extraction operations. This study proposes a CBM pre-extraction parameter and completion time prediction method based on the Transformer model. By integrating autoregressive models and wavelet denoising techniques, the approach effectively captures temporal features and long-term dependencies in CBM data. Experimental results demonstrate that the proposed model outperforms traditional methods in short-, medium-, and long-term predictions, with a median R 2 value of 0.99072, and 76% of the training results exceeding 0.9. Furthermore, a CBM pressure inversion model was developed, combining dimensional analysis and physical similarity principles with the Transformer model, enabling the dynamic detection of high- and low-pressure regions in coal seams. In single borehole compliance time predictions, the median compliance time for the first stage is 4 days, with an average of 49 days and a maximum of 277 days, providing adjustment guidance for boreholes with extended compliance times. The proposed model significantly improves prediction accuracy and stability, offering critical support for developing scientifically sustainable pre-extraction plans and advancing intelligent CBM management.}, archive = {J_EAAI}, author = {Li Yan and Hu Wen and Zhenping Wang and Yongfei Jin and Jun Guo and Yin Liu and Shixing Fan}, doi = {10.1016/j.engappai.2024.109661}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109661}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Prediction and evaluation of key parameters in coalbed methane pre-extraction based on transformer and inversion model}, volume = {139}, year = {2025}, }
• (2025). Multilingual entity alignment by abductive knowledge reasoning on multiple knowledge graphs. EAAI, 139, 109660. (www)

  @article{ , abstract = {Objectives: Entity alignment (EA) seeks to identify similar real-world objects in different multilingual knowledge graphs (KGs), also known as ontology alignment. EA assists in handling a wide range of language semantics and in building integrated knowledge bases. However, most mainstream studies have focused on structural information, paying little attention to insufficient contextual information and limited handling of complex relationships. This paper aims to address these limitations and improve EA performance and efficiency. Methods: This paper investigates multilingual EA techniques and proposes a novel Abductive Knowledge Reasoning (AKR) model to address these issues. AKR can compute complex relationship semantics context by reasoning and enrich counterpart entity contextual information through centrality calculation, which helps connect distant entities in multilingual KGs. Novelty: The proposed AKR model introduces a new approach to EA by integrating centrality calculation and relational semantics reasoning. This method overcomes the limitations of existing EA techniques by effectively handling insufficient contextual information and complex relationships in multilingual KGs. Findings: AKR outperforms all state-of-the-art EA models across five datasets. AKR achieves H i t @ 1 score of 79.4%, for entity alignment between Chinese-to-English knowledge graphs representing 19.9% improvement over the best-performing translation-based model, Neighborhood-Aware Attentional Representation Entity Alignment, and a 5.0% improvement over the best-performing graph neural network-based model, Relational Semantics Augmentation.}, archive = {J_EAAI}, author = {Muhammad Usman Akhtar and Jin Liu and Zhiwen Xie and Xiaohui Cui and Xiao Liu and Bo Huang}, doi = {10.1016/j.engappai.2024.109660}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109660}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multilingual entity alignment by abductive knowledge reasoning on multiple knowledge graphs}, volume = {139}, year = {2025}, }
• (2025). Dual graph-regularized low-rank representation for hyperspectral image denoising. EAAI, 139, 109659. (www)

  @article{ , abstract = {Hyperspectral images have a wide range of applications in many fields. However, when hyperspectral images are captured by spectrometers, there is inevitably considerable noise, which affects subsequent research. In recent years, many hyperspectral image denoising methods based on low-rank representations have been proposed. Artificial intelligence denoising methods are also popular. However, the research on multi noise denoising is rarely mentioned, and most literatures only focus on one noise in hyperspectral images. Thus, we propose a denoising model for hyperspectral image based on dual graph-regularized low-rank representation, which can not only reduce multiple types of noise simultaneously, but also preserves details of the original image. In particular, this is the first time that the dual low-rank representation and dual graph regularizations are used on hyperspectral images. We solve this method using the linearized alternating direction method with adaptive penalty. Finally, we conduct experiments on simulated and real data sets to verify the effectiveness of our method. The experimental results show that our method can not only effectively remove a variety of mixed noises, but also well retain the details of the image.}, archive = {J_EAAI}, author = {Chengcai Leng and Mingpei Tang and Zhao Pei and Jinye Peng and Anup Basu}, doi = {10.1016/j.engappai.2024.109659}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109659}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual graph-regularized low-rank representation for hyperspectral image denoising}, volume = {139}, year = {2025}, }
• (2025). Classification of similar electronic components by transfer learning methods. EAAI, 139, 109658. (www)

  @article{ , abstract = {Proper selection of electronic components and automated component identification is critical for fast production processes in industry. In addition, for Internet of Things (IoT) systems, accurate and fast selection of similar electronic components is an important problem. In this study, a transfer learning-based method is proposed to classify electronic components that are difficult to select due to their similarity. Eight different convolutional neural network (CNN) models and a novel model developed only in this study were used to classify electronic components. In addition to the transfer learning methods, the parallel CNN method, in which hyperparameter determination is done by trial and error, was developed and used to solve the classification problem. In addition to the transfer learning method, the parameters were tried to be determined by the trial-and-error method for hyperparameter selection. The effect of batch size and learning rate hyperparameter variations on the prediction success of parallel CNN models is analyzed. The effect of two different batch sizes and learning rate values for transfer learning models is also analyzed. Metrics such as confusion matrix, accuracy, and loss were used for evaluation methods. The number of parameters and runtime metrics of the models were also evaluated. All experiments were averaged to obtain a general intuition of success. The success of the proposed method is given by the evaluation metrics. According to the accuracy metric, the Densely Connected Convolutional Networks (DenseNet-121) model was the most successful model with a value of 98.2925%.}, archive = {J_EAAI}, author = {Göksu Taş}, doi = {10.1016/j.engappai.2024.109658}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109658}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Classification of similar electronic components by transfer learning methods}, volume = {139}, year = {2025}, }
• (2025). Building image reconstruction and dimensioning of the envelope from two-dimensional perspective drawings. EAAI, 139, 109657. (www)

  @article{ , abstract = {In the construction industry, a project typically begins with the creation of two-dimensional (2D) building plans, defining the client’s specifications. Using these plans, a digital three-dimensional (3D) model is developed to visualize the anticipated outcome and to verify the model’s alignment with the client’s expectations. The process of converting from 2D to 3D can become time-intensive if there is a need for modifications or if the project’s overall complexity is high. To enhance efficiency and accuracy, this research introduces an end-to-end framework referred to as BIRD which stands for B uilding I mage R econstruction and D imensioning. BIRD is capable of accepting five 2D perspective drawings of a building as inputs and generating a proportionate 3D model of the building envelope as an output. This is accomplished through the integration of multiple techniques that use convolutional neural networks to extract a refined set of line segments, identify measurements, and align each perspective with the floor plan drawing. The key contributions of this study includes: (1) a novel deep learning model designed for the identification of line segments in building plans; (2) novel algorithms that facilitate the generation of information required for 3D modeling; (3) an end-to-end framework for building reconstruction; and (4) novel performance metrics specifically tailored for the 2D to 3D conversion challenge. The practical application of this research was validated through the use of complete building plans provided by an industry partner. In summary, it was observed that BIRD demonstrated high accuracy in the creation of 3D visualizations, highlighting its real-world efficacy.}, archive = {J_EAAI}, author = {Andrew Fisher and Lucas Moreira and Muntasir Billah and Pawan Lingras and Vijay Mago}, doi = {10.1016/j.engappai.2024.109657}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109657}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Building image reconstruction and dimensioning of the envelope from two-dimensional perspective drawings}, volume = {139}, year = {2025}, }
• (2025). Online prediction of hydro-pneumatic tensioner system of floating platform under internal waves. EAAI, 139, 109656. (www)

  @article{ , abstract = {To address the issue of low accuracy in the current motion response prediction model of the floating platform tensioner system, this paper proposes an online prediction method that integrates Empirical Mode Decomposition (EMD), Kernel Principal Component Analysis (KPCA), and Long Short-Term Memory (LSTM). The EMD technique is employed to decompose the sequence of environmental factors, reducing their non-stationarity. Subsequently, KPCA is used to extract key influencing factors and reduce input dimensionality. Finally, LSTM neural networks are applied to capture long-term dependencies in features and make accurate predictions. By validating the model using motion response data from the tensioner platform device under two scenarios with and without internal waves, it is compared against other models. The results show that the EMD-KPCA-LSTM model has high prediction accuracy in both scenarios. In particular, compared with the Convolutional Neural Network (CNN) model, the mean Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE) of the displacement and tension of the system decreased by 52.97%, 55.14%, 56.31%, 68.97%, 71.02% 57.60%, respectively, and R-square (R 2 ) increased by 7.14% and 12.37%. In summary, the model has a good ability for data fitting and high prediction accuracy and has important practical value.}, archive = {J_EAAI}, author = {Jianwei Wang and Xiaofan Jin and Xuchu Liu and Ze He and Jiachen Chai and Pengfa Liu and Yuqing Wang and Wei Cai and Rui Guo}, doi = {10.1016/j.engappai.2024.109656}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109656}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Online prediction of hydro-pneumatic tensioner system of floating platform under internal waves}, volume = {139}, year = {2025}, }
• (2025). A velocity adaptive steering control strategy of autonomous vehicle based on double deep Q-learning network with varied agents. EAAI, 139, 109655. (www)

  @article{ , abstract = {Autonomous vehicle steering control is sensitive to the vehicle driving speed and traditional model-based approaches are limited by the accuracy of the control model in various driving speed scenarios. To address these challenges, this study proposes a model-free control strategy based on deep reinforcement learning (DRL). In this strategy, the improved double deep Q-learning network (DDQN) with varied agents is employed for steering control to minimize the tracking errors across varying speeds. According to the kinematic characteristics of the vehicle, a dynamic action space is applied to enhance the tracking capability at high speeds. Furthermore, to ensure the output of the agent is more stable, a velocity adaptive reward function is designed by incorporating an action penalty factor. The performance of the proposed strategy is evaluated through simulation and experimental comparisons with other existing algorithms at a double-lane change maneuver. The results demonstrate that the DDQN-based strategy can effectively adapt to various vehicle speeds and perform the tracking task more accurately and stably. Finally, the feasibility of this strategy is verified using an actual prototype vehicle.}, archive = {J_EAAI}, author = {Xinyou Lin and Jiawang Huang and Biao Zhang and Binhao Zhou and Zhiyong Chen}, doi = {10.1016/j.engappai.2024.109655}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109655}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A velocity adaptive steering control strategy of autonomous vehicle based on double deep Q-learning network with varied agents}, volume = {139}, year = {2025}, }
• (2025). Laparoscopic stereo matching using 3-dimensional fourier transform with full multi-scale features. EAAI, 139, 109654. (www)

  @article{ , abstract = {3 - D imensional (3D) reconstruction of laparoscopic surgical scenes is a key task for future surgical navigation and automated robotic minimally invasive surgery. Binocular laparoscopy with stereo matching enables 3D reconstruction. Stereo matching models used for natural images such as autopilot tend to be less suitable for laparoscopic environments due to the constraints of small samples of laparoscopic images, complex textures, and uneven illumination. In addition, current stereo matching modules use 3D convolutions and transformers in the spatial domain as the base module, which is limited by the ability to learn in the spatial domain. In this paper, we propose a model for laparoscopic stereo matching using 3D F ourier T ransform combined with F ull M ulti-scale F eatures (FT-FMF Net). Specifically, the proposed F ull M ulti-scale F usion M odule (FMFM) is able to fuse the full multi-scale feature information from the feature extractor into the stereo matching block, which densely learns the feature information with parallax and FMFM fusion information in the frequency domain using the proposed D ense F ourier T ransform M odule (DFTM). We validated the proposed method in both the laparoscopic dataset (SCARED) and the endoscopic dataset (SERV-CT). In comparison with other popular and advanced deep learning models available at present, FT-FMF Net achieves the most advanced stereo matching performance available. In the SCARED and SERV-CT public datasets, the End-Point-Error (EPE) was 0.7265 and 2.3119, and the R oot M ean S quare E rror Depth (RMSE Depth) was 4.00 mm and 3.69 mm, respectively. In addition, the inference time is only 0.17s. Our project code is available on https://github.com/wurenkai/FT-FMF .}, archive = {J_EAAI}, author = {Renkai Wu and Pengchen Liang and Yinghao Liu and Yiqi Huang and Wangyan Li and Qing Chang}, doi = {10.1016/j.engappai.2024.109654}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109654}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Laparoscopic stereo matching using 3-dimensional fourier transform with full multi-scale features}, volume = {139}, year = {2025}, }
• (2025). Automatic collaborative learning for drug repositioning. EAAI, 139, 109653. (www)

  @article{ , abstract = {Drug repositioning seeks to identify new therapeutic uses for existing drugs, accelerating development and reducing costs. While traditional wet lab experiments are costly, computational methods offer a low-cost, efficient alternative. Despite their potential, most research in this field has uncritically employed the standard message-passing mechanism of Graph Neural Network (GNN), limiting the assessment of collaborative effects on prediction accuracy. In this paper, we introduce a novel model, an automatic collaborative learning framework for drug repositioning. Initially, we propose a metric to measure the interaction levels among neighbors and integrate it with the intrinsic message-passing mechanism of GNN, thereby enhancing the impact of various collaborative effects on prediction accuracy. Furthermore, we introduce an advanced contrastive learning technique to align feature consistency between the disease–drug association space and the customized neighbor space. This approach leverages the inherent regularities across different feature dimensions to minimize feature redundancy. Extensive experiments conducted on three benchmark datasets demonstrate substantial improvements of this novel model over various state-of-the-art methods. Case studies further highlight the practical utility of this model.}, archive = {J_EAAI}, author = {Yi Wang and Yajie Meng and Chang Zhou and Xianfang Tang and Pan Zeng and Chu Pan and Qiang Zhu and Bengong Zhang and Junlin Xu}, doi = {10.1016/j.engappai.2024.109653}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109653}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Automatic collaborative learning for drug repositioning}, volume = {139}, year = {2025}, }
• (2025). A framework for robust glaucoma detection: A confidence-aware deep uncertainty quantification approach with a comprehensive assessment for enhanced clinical decision-making. EAAI, 139, 109651. (www)

  @article{ , abstract = {Glaucoma poses a significant threat to public health worldwide, as it can result in irreversible vision loss. Timely identification is vital for halting the progression of visual field deterioration. In recent years, deep neural networks (DNNs) have become increasingly popular in medical imaging due to their ability to identify patterns. As a result, this study introduces a new computer-aided diagnosis (CAD) system based on deep learning (DL) algorithms for glaucoma detection that extracts meaningful features from retinal fundus images (RFIs) and employs uncertainty quantification (UQ) models, including Monte Carlo dropout (MCD), ensemble Bayesian, and ensemble Monte Carlo dropout (EMCD), to generate both point estimates and confidence values for the outputs, thereby capturing the uncertainty associated with the classifications. The proposed framework is validated using well-known clinical datasets, and the reliability of the outputs is evaluated using comprehensive performance metrics such as expected calibration error (ECE), entropy analysis, and a multi-criteria UQ assessment. Experimental results demonstrate the superiority of the ensemble model, with uncertainty accuracies registering at 97.64%, 97.26%, and 98.97% for the “ACRIMA” , “RIM-ONE-DL” , and “ORIGA” datasets, respectively. Moreover, the proposed algorithms can alert users to the majority of erroneous diagnoses by assigning uncertainty labels, providing valuable insights for clinicians in glaucoma detection. Such tools can assist healthcare professionals in reducing the probability of misdiagnosis and ensuring that patients receive timely and appropriate treatment.}, archive = {J_EAAI}, author = {Javad Zarean and AmirReza Tajally and Reza Tavakkoli-Moghaddam and Seyed Mojtaba Sajadi and Niaz Wassan}, doi = {10.1016/j.engappai.2024.109651}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109651}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A framework for robust glaucoma detection: A confidence-aware deep uncertainty quantification approach with a comprehensive assessment for enhanced clinical decision-making}, volume = {139}, year = {2025}, }
• (2025). Analysis of computer communication networks based on evaluation of domination and double domination for interval-valued T-spherical fuzzy graphs and their applications in decision-making problems. EAAI, 139, 109650. (www)

  @article{ , abstract = {This research introduces the Interval-Valued T-Spherical Fuzzy Graph (IVTSFG), a novel extension of fuzzy graph theory designed to address imprecision in decision-making processes, network analysis, and Computer Communication Networks (CCNs). Integrating four types of membership degrees-membership, non-membership, abstinence, and hesitancy-the IVTSFG framework significantly enhances the ability to model and analyze complex systems with uncertain data. The study explores the theories of domination and double domination within the context of IVTSFGs, presenting new methods for evaluating network resilience and optimization. Key findings include the development of innovative techniques for applying domination and double domination in IVTSFGs, demonstrating improved performance in managing CCNs. Comparative analysis with existing fuzzy graph models highlights the advantages of IVTSFGs, particularly in capturing nuanced relationships within network structures. The research provides practical examples and empirical comparisons, showcasing the framework's effectiveness in various decision-making scenarios.}, archive = {J_EAAI}, author = {Sami Ullah Khan and Fiaz Hussain and Tapan Senapati and Shoukat Hussain and Zeeshan Ali and Domokos Esztergár-Kiss and Sarbast Moslem}, doi = {10.1016/j.engappai.2024.109650}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109650}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Analysis of computer communication networks based on evaluation of domination and double domination for interval-valued T-spherical fuzzy graphs and their applications in decision-making problems}, volume = {139}, year = {2025}, }
• (2025). Detecting critical diseases associated with higher mortality in electronic health records using a hybrid attention-based transformer. EAAI, 139, 109649. (www)

  @article{ , abstract = {Electronic health records (EHRs) are crucial for modern medical practices, providing digital storage of patient health information. However, accurately identifying diseases that lead to mortality within EHRs remains challenging. This study introduces a novel hybrid intra-multi-head attention with a bidirectional gated recurrent unit (Hybrid-IMHA-BiGRU) model to detect critical diseases associated with higher patient mortalities. Our main objective is to capture critical contextual features in text sequences. To achieve this, we propose both a two-channeled attention mechanism (TC-AM) and a single-channeled AM (SC-AM). The TC-AM focuses on symptoms, diseases, and mortalities, while the SC-AM discerns intricate patterns in long text sequences. Our approach’s novelty lies in bidirectional processing, enhancing the model’s ability to comprehend entire sequences by capturing contextual information from both preceding and following tokens. Existing studies struggle with capturing relevant contexts in variable-length EHRs due to issues with long-term dependencies, dynamic sequences, and variations in patient data representation. To address these challenges, we recommend a medical context-based robustly optimized bidirectional encoder representations from transformers pretraining approach (Med-RoBERTa) for embedding computations, which efficiently captures nuanced relationships and semantic dependencies within variable-length EHRs. Additionally, our model incorporates a cross-AM to adaptively weigh parts of the input sequence, which is beneficial for handling sequences of varying lengths. Our findings indicate that the proposed model outperforms state-of-the-art models, achieving 98.10% and 97.90% accuracies and F1 scores of 94.72% and 94.45% on EHRs datasets. This underscores the effectiveness of our approach in precisely identifying diseases associated with patient mortalities in EHRs.}, archive = {J_EAAI}, author = {Dheeraj Kodati and Chandra Mohan Dasari}, doi = {10.1016/j.engappai.2024.109649}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109649}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Detecting critical diseases associated with higher mortality in electronic health records using a hybrid attention-based transformer}, volume = {139}, year = {2025}, }
• (2025). Designing on-board explainable passenger flow prediction. EAAI, 139, 109648. (www)

  @article{ , abstract = {Nowadays, predicting public transport passenger flow (PF) is essential to optimize service planning and provide information to commuters. However, while current research focuses on enhancing accuracy using advanced models, like recurrent and graph neural networks, other key aspects, such as interpretability and computational efficiency, are often neglected. To fill this gap, we propose a framework to design on-board explainable PF prediction based on eXtreme Gradient Boosting (XGBoost). This framework enhances model interpretability and reduces computational costs, making the resulting PF predictive model suitable for inference on low-end devices. The proposed framework is validated on a real-world dataset from a major Italian city, involving 25 buses. Our results show that the framework achieves performance comparable to Convolutional Neural Networks (CNNs), with only a 0.2% percentage increase in Mean Absolute Percentage Error (MAPE). Additionally, it significantly reduces both inference time and energy consumption, with percentage decrease of 63%. Finally, we present examples to illustrate the interpretability of the predictions using the SHapley Additive exPlanation (SHAP) method.}, archive = {J_EAAI}, author = {Mario Barbareschi and Antonio Emmanuele and Nicola Mazzocca and Franca Rocco di Torrepadula}, doi = {10.1016/j.engappai.2024.109648}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109648}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Designing on-board explainable passenger flow prediction}, volume = {139}, year = {2025}, }
• (2025). Positive discrimination of minority classes through data generation and distribution: A case study in olive disease classification. EAAI, 139, 109646. (www)

  @article{ , abstract = {Deep learning models have achieved remarkable success in various tasks, especially in classification. This success is particularly evident in the precise classification of plant diseases, which is crucial for effective agricultural management. However, accurate classification faces challenges, particularly in data collection, where certain classes are underrepresented, namely the minority classes. This issue can significantly impact model performance. To tackle this challenge, this paper introduces a novel methodology that differs from existing approaches. We focus on addressing the issue of minority classes in image-based classification tasks, particularly for olive diseases. We employ data generation methods, including basic transformations, to produce augmented data and utilize Deep Convolutional Generative Adversarial Networks (DCGAN) to produce generated data. Next, we apply the Frechet Inception Distance (FID) to the generated dataset to select the highest-quality images. We then distribute varying percentages (25%, 50%, 75%, 100%) of this new data into the minority classes of the original dataset. Our data distribution strategies involve incorporating specific amounts of (1) augmented data, (2) generated data, and (3) a combination of both augmented and generated data to achieve target percentages (T.P) in the resulting dataset. Our experiments focus on classifying olive diseases into seven distinct categories using a pre-trained Convolutional Neural Network (CNN) architecture. We observe significant improvements in the model’s performance, particularly in the accurate classification of minority classes. This approach enhances diagnostic accuracy and optimizes data distribution, which is crucial for effectively addressing the challenges posed by minority classes.}, archive = {J_EAAI}, author = {Hicham El Akhal and Aissa Ben Yahya and Abdelbaki El Belrhiti El Alaoui}, doi = {10.1016/j.engappai.2024.109646}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109646}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Positive discrimination of minority classes through data generation and distribution: A case study in olive disease classification}, volume = {139}, year = {2025}, }
• (2025). Health evaluation of shield tunnel lining using combination weighting and finite interval cloud model. EAAI, 139, 109645. (www)

  @article{ , abstract = {To solve the problem of inaccurate and unreasonable health evaluation of shield tunnel lining, a novel health evaluation model of shield tunnel lining based on the combination weighting method and finite interval cloud model is proposed. A health evaluation index system including 6 level-Ⅰ indexes and 15 level-II indexes and evaluation criteria are established for the shield tunnel lining. The weights of evaluation indexes are calculated by the game theory combination weighting method. The finite interval cloud model is used to evaluate the health of shield tunnel lining, which considers the uncertainty of various information within the interval. To verify the applicability of the proposed approach, it was applied to the shield construction section from Bei Chen Station to the Olympic Sports Center Station of Xi'an Metro Line 14. The results show that: (1) The health evaluation grade of shield tunnel lining in Samples 1–3 is level II. The result is in agreement with the actual situation which validates the practicality of the employed methodology. (2) The change in the evaluation index has little influence on the evaluation results, and the evaluation results are level II. The key risk factors were identified as U 32 , U 31 , and U 12 by sensitivity analysis. Corresponding measures should be taken to ensure the stability of these three indexes and to ensure the safety of shield tunnel operation. Therefore, the proposed approach maximizes the assurance of the rationality of the evaluation results, which can be feasibly used in various applications and can provide guidance for other similar projects.}, archive = {J_EAAI}, author = {Yu-Wei Zhang and De-Sai Guo and Zhan-Ping Song and Yi-Duo Zhang and Lei Ruan and Zhao-Bo Yan}, doi = {10.1016/j.engappai.2024.109645}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109645}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Health evaluation of shield tunnel lining using combination weighting and finite interval cloud model}, volume = {139}, year = {2025}, }
• (2025). Gearbox fault diagnosis based on temporal shrinkage interpretable deep reinforcement learning under strong noise. EAAI, 139, 109644. (www)

  @article{ , abstract = {Gearbox fault diagnosis is crucial for the safe operation of mechanical systems. While Deep Learning (DL) has demonstrated promising results in this area, most existing methods rely on static supervised learning, lacking the dynamic, interactive learning capabilities similar to human decision-making. To tackle this issue, this study presents a novel approach that combines the strengths of Deep Reinforcement Learning (DRL) with the interpretability of a temporal shrinkage interpretable network. DRL integrates the perception abilities of DL with the decision-making capabilities of Reinforcement Learning (RL), offering a more comprehensive solution for complex challenges. In this method, gearbox fault diagnosis is formulated as a sequential decision problem within a Classification Markov Decision Process (CMDP). A multi-scale temporal shrinkage module is utilized to construct an interpretable network, which enhances model interpretability and reduces the negative impact of noisy data in harsh working conditions. The diagnosis agent autonomously learns the optimal classification policy, reducing the need for manual intervention and human expertise. Experimental results show excellent generalization and stability, achieving over 98.5% accuracy even in noisy conditions. It outperforms existing methods and highlights its robustness in challenging operational environments.}, archive = {J_EAAI}, author = {Zeqi Wei and Hui Wang and Zhibin Zhao and Zheng Zhou and Ruqiang Yan}, doi = {10.1016/j.engappai.2024.109644}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109644}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Gearbox fault diagnosis based on temporal shrinkage interpretable deep reinforcement learning under strong noise}, volume = {139}, year = {2025}, }
• (2025). Multimodal transformer for early alarm prediction. EAAI, 139, 109643. (www)

  @article{ , abstract = {Alarms are an essential part of distributed control systems designed to help plant operators keep the processes stable and safe. In reality, however, alarms are often noisy and thus can be easily overlooked. Early alarm prediction can give the operator more time to assess the situation and introduce corrective actions to avoid downtime and negative impact on human safety and environment. Existing studies on alarm prediction typically rely on signals directly coupled with these alarms. However, using more sources of information could benefit early prediction by letting the model learn characteristic patterns in the interactions of signals and events. Meanwhile, multimodal deep learning has recently seen impressive developments. Combination (or fusion) of modalities has been shown to be a key success factor, yet choosing the best fusion method for a given task introduces a new degree of complexity, in addition to existing architectural choices and hyperparameter tuning. This is one of the reasons why real-world problems are still typically tackled with unimodal approaches. To bridge this gap, we introduce a multimodal Transformer model for early alarm prediction based on a combination of recent events and signal data. The model learns the optimal representation of data from multiple fusion strategies automatically. The model is validated on real-world industrial data. We show that our model is capable of predicting alarms with the given horizon and that the proposed multimodal fusion method yields state-of-the-art predictive performance while eliminating the need to choose among conventional fusion techniques, thus reducing tuning costs and training time.}, archive = {J_EAAI}, author = {Nika Strem and Devendra Singh Dhami and Benedikt Schmidt and Kristian Kersting}, doi = {10.1016/j.engappai.2024.109643}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109643}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multimodal transformer for early alarm prediction}, volume = {139}, year = {2025}, }
• (2025). A framework for interpretable deep learning in cross-subject detection of event-related potentials. EAAI, 139, 109642. (www)

  @article{ , abstract = {Event-related potential-based Brain-Computer Interfaces are becoming widely popular due to their ability to send commands quickly with high accuracy. However, the stationary characteristics of electroencephalographic signals, coupled with their low signal-to-noise ratio, lead to variations in amplitude, time period, and latency in the patterns of event-related potentials across different trials, sessions, days and subjects. Conventional feature extraction and machine learning algorithms are not designed to handle these differences, requiring the development of methods that can address these variations. Here, we propose a novel lightweight deep neural network for event-related potential classification, consisting of three modules. In this model, we have a spatio-temporal module that learns local features simultaneously across channels and time points. Following this, there's a component extractor module comprising depthwise convolutions, inspired by mixed depthwise convolutions, to capture the event-related potential characteristics with different temporal durations. Lastly, an advanced temporal layer addresses event-related potential shape and scale variations using deformable convolutions. We conducted experiments on event-related potential detection in a subject-independent scenario using one error-related negativity potential dataset and three perturbation-evoked potential datasets. Comparisons were made with established methods including two conventional machine learning algorithms and three well-known deep learning architectures, demonstrating that our model outperformed them in terms of classification accuracy and parameter efficiency. In our analysis, we aimed to understand the model's performance using gradient-weighted class activation mapping and t-distributed stochastic neighbor embedding. These methods facilitated the visualization and interpretation of our model's effectiveness, providing insights into its relationship with the neuroscientific characteristics of event-related potentials.}, archive = {J_EAAI}, author = {Shayan Jalilpour and Gernot Müller-Putz}, doi = {10.1016/j.engappai.2024.109642}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109642}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A framework for interpretable deep learning in cross-subject detection of event-related potentials}, volume = {139}, year = {2025}, }
• (2025). Assessing the factors of blockchain technology-enabled hospitals using an integrated interval-valued q-rung orthopair fuzzy decision-making model. EAAI, 139, 109641. (www)

  @article{ , abstract = {In the current era, blockchain technology (BT) has emerged as a novel technique to maintain the operations of healthcare management systems. Assessment of blockchain technology (BT)-enabled hospitals can be considered as a multi-criteria decision making (MCDM) problem because of the existence of several criteria. The aim of this study is to develop a hybrid MCDM method for evaluating the factors of multi-criteria BT-enabled hospital selection problem under interval-valued q-rung orthopair fuzzy sets (IVq-ROFSs). For this purpose, a weighted aggregated sum product assessment (WASPAS) model is presented with the combination of IVq-ROF interaction aggregation operators, the standard deviation (SD)-based model and pivot pairwise relative criteria importance assessment (PIPRECIA) tool called IV-q-ROF-SD-PIPRECIA-WASPAS model within the context of IVq-ROFSs. For this purpose, some new IVq-ROF interaction aggregation operators are developed with their desirable characteristics. Next, the standard deviation-based model and PIPRECIA model on IVq-ROFSs are proposed to obtain the final weight of criteria, whereas the rank-based formula is presented to determine the decision experts’ weights with IVq-ROF information. The presented IV-q-ROF-SD-PIPRECIA-WASPAS model is applied on a case study of BT-enabled hospitals assessment, which confirms its applicability and usefulness. Sensitivity analysis and comparative discussion have been performed to reveal the consistency, robustness and efficiency of the presented model. The BT-enabled hospital-II with highest UD (0.4453) has emerged as the best choice among a set of BT-enabled hospitals. The factor "flexibilty" with highest weight (0.0898) value followed that the scalability (0.0809), transaction speed and accountability with same weight (0.0779) value, and network availability with weight (0.0771) for BT-enabled hospitals assessment. The final results conclude that the developed methodology can provide more accurate decisions while considering multiple indicators and input uncertainties.}, archive = {J_EAAI}, author = {Rashmi Pathak and Badal Soni and Naresh Babu Muppalaneni and Muhammet Deveci}, doi = {10.1016/j.engappai.2024.109641}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109641}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Assessing the factors of blockchain technology-enabled hospitals using an integrated interval-valued q-rung orthopair fuzzy decision-making model}, volume = {139}, year = {2025}, }
• (2025). Learning discriminative representations by a canonical correlation analysis-based siamese network for offline signature verification. EAAI, 139, 109640. (www)

  @article{ , abstract = {In offline signature verification tasks, capturing different writing behaviors between genuine and forged signatures is a crucial and challenging step. In this paper, a novel writer independent Canonical Correlation Analysis-based Siamese Network (CCASigNet) is proposed to learn discriminative representations between different signature pairs. Specifically, we first construct signature pairs with three types: genuine-genuine, genuine-forged, and forged-forged. Then, different signature pairs are fed into CCASigNet for training with the Canonical Correlation Analysis (CCA) and classification-based losses. After network training, we extract the feature of signatures by CCASigNet and use writer-dependent classifiers to construct a comprehensive verification system. Extensive experiments on four benchmark signature datasets demonstrate that the proposed CCASigNet learns discriminative representations between different signature pairs and achieves state-of-the-art or competitive performance compared with advanced verification systems. In addition, the proposed CCASigNet has good generalization ability and is easy to transfer to different datasets with different language scripts within the realm of offline signature verification tasks.}, archive = {J_EAAI}, author = {Lidong Zheng and Xingbiao Zhao and Shengjie Xu and Yuanyuan Ren and Yuchen Zheng}, doi = {10.1016/j.engappai.2024.109640}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109640}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Learning discriminative representations by a canonical correlation analysis-based siamese network for offline signature verification}, volume = {139}, year = {2025}, }
• (2025). Rice leaf disease identification and classification using machine learning techniques: A comprehensive review. EAAI, 139, 109639. (www)

  @article{ , abstract = {In recent times, various researchers attempted to develop artificial intelligence (AI) assisted techniques in the field of agriculture for early detection, surveillance and treatment related to plant leaf, seed, root, and stem diseases. Rice leaf disease detection is one of such important areas, where the crop is frequently affected by various diseases. Farmer inspects usually at a later stage causing enormous damage. This manual inspection is subjective, time-consuming and error prone. Under such situation, AI-enabled tools and techniques play crucial role for early and more precise prediction of rice diseases. This paper demonstrates a comprehensive review on application of AI-assisted rice leaf disease detection in the last two decades. Research studies were searched using relevant keywords through the online databases [ PubMed: 246; Science Direct: 100; Scopus: 56; Web of Science: 8; Willey online library:16; Cochrane:0; Cross references:20 ]. A total of 446 titles and abstracts were identified as suitable for this study and finally, 48 most-appropriate state-of-art articles were considered. Furthermore, this study summarizes the visual characteristics of rice leaf diseases, imaging modalities and image acquisition techniques. Various image processing techniques for infected leaf area segmentation and feature extraction were also summarized. Finally, the reported machine learning (ML) algorithms were discussed and compared in respect to their advantages and limitations. In addition, AI-enabled mobile applications for rice disease detection have been discussed.}, archive = {J_EAAI}, author = {Rashmi Mukherjee and Anushri Ghosh and Chandan Chakraborty and Jayanta Narayan De and Debi Prasad Mishra}, doi = {10.1016/j.engappai.2024.109639}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109639}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Rice leaf disease identification and classification using machine learning techniques: A comprehensive review}, volume = {139}, year = {2025}, }
• (2025). Towards salient object detection via parallel dual-decoder network. EAAI, 139, 109638. (www)

  @article{ , abstract = {Salient object detection, an important preprocessing step in computer vision, segments the most prominent objects in an image. However, existing research in this field utilizes transformer-based methods to capture global context information, failing to effectively obtain local spatial features. To solve this issue, we propose a parallel dual-decoder network, which consists of a novel semantic decoder and a modified salient decoder. Specifically, the proposed semantic decoder is designed to learn the local spatial details, and the salient decoder utilizes the learnable queries to establish global saliency dependencies among objects. Moreover, the two decoders establish correlations between saliency and multi-scale semantic representations through cross-attention interaction, significantly enhancing the performance of salient object detection. In other words, we obtain global context information in the decoder to prevent discriminative features from being diluted during information propagation. Extensive experiments on 15 benchmark datasets demonstrate that our model significantly outperforms other comparison methods and shows promising potential for real-world applications such as challenging optical remote sensing, underwater, low-light, and other open scenarios. In addition, our method shows excellent performance in other downstream tasks such as camouflaged object detection, transparent object detection, shadow detection, and semantic segmentation.}, archive = {J_EAAI}, author = {Chaojun Cen and Fei Li and Zhenbo Li and Yun Wang}, doi = {10.1016/j.engappai.2024.109638}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109638}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Towards salient object detection via parallel dual-decoder network}, volume = {139}, year = {2025}, }
• (2025). A novel approach in constructing virtual real driving emission trips through genetic algorithm optimization. EAAI, 139, 109637. (www)

  @article{ , abstract = {The Real Driving Emission (RDE) test became a critical part of the process conducted by manufacturers to fulfill the approval procedure of every new vehicle model. This test measures the regulated emissions from a vehicle during a trip, which follows a specific set of operation requirements, aiming to assess the vehicle’s emission levels in real-world conditions. Additionally, In-Service Conformity (ISC) tests, which consist in performing an RDE trip, were also introduced to demonstrate vehicles emissions compliance over their lifespan. Considering that modern vehicles embed exhaust emission sensors and connectivity capabilities, it is believed that there is an opportunity for manufacturers to leverage the data generated by these vehicles to forecast the outcomes of an ISC test. However, as this study presents through the analysis of an extensive database of more than 600 trips from a mild-hybrid diesel vehicle, none of the real-world trips might comply with all the driving requirements of the RDE standard. Faced with this outcome, this work proposes the application of a Genetic Algorithm (GA) optimization to construct virtual RDE trips from real-driving data. In particular, the proposed methodology leverages such algorithm to combine real driving fragments from various trips in order to align with the main RDE trip requirements. The methodology focuses on vehicle, engine, and exhaust after-treatment variables, utilizing signal optimization connections to create a realistic analysis of vehicle pollutants. The research suggests that a combination of vehicle speed, coolant temperature, exhaust temperature, and Selective Catalytic Reduction (SCR) load leads to a significant number of RDE-compliant results under simplified legislative conditions, from which emissions profiles could be assessed. The proposed methodology details the development of an Adaptive Genetic Algorithm (AGA) and the data pipeline to create specific RDE trips, offering the capability to customize the desired Driving Cycles (DC).}, archive = {J_EAAI}, author = {Jose Ponce and Alvin Barbier and Carlos E. Palau and Carlos Guardiola}, doi = {10.1016/j.engappai.2024.109637}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109637}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel approach in constructing virtual real driving emission trips through genetic algorithm optimization}, volume = {139}, year = {2025}, }
• (2025). A novel multi-objective artificial bee colony algorithm for solving the two-echelon load-dependent location-routing problem with pick-up and delivery. EAAI, 139, 109636. (www)

  @article{ , abstract = {This study considers a two-echelon load-dependent location routing problem with pick-up and delivery (2E-LDLRPPD). As a variant of the two-echelon vehicle routing problem with pick-up and delivery (2E-VRPPD), the 2E-LDLRPPD includes additional variants such as two-echelon location-routing problem (2E-LRP) and load-dependent vehicle routing problem (LDVRP). However, much of the existing research work has traditionally focused on a single objective, predominantly aimed to minimize costs. In our case, we build a multi-objective model that concurrently minimizes costs, carbon emissions, and the number of vehicles used. Heuristic algorithms are commonly used to solve complex location-routing problems. Therefore, we propose a hybrid heuristic algorithm named the improved elite-guided multi-objective artificial bee colony algorithm with variable neighborhood search (IEMOABC-VNS). Base on elite-guided multi-objective artificial bee colony algorithm (EMOABC), a two-archive elite-guide strategy is deployed to strike a balance between diversity and convergence. The efficacy of the IEMOABC-VNS is compared experimentally with four other hybrid heuristic algorithms on test instances and a real-world case. Computational results demonstrate that the IEMOABC-VNS outperforms the competing algorithms in solving 2E-LDLRPPD, and obtains a high-quality Pareto front in a relatively short time. Especially, the algorithm exhibits significant performance enhancements when applied to large-scale instances.}, archive = {J_EAAI}, author = {Dekun Tan and Xuhui Liu and Ruchun Zhou and Xuefeng Fu and Zhenzhen Li}, doi = {10.1016/j.engappai.2024.109636}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109636}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel multi-objective artificial bee colony algorithm for solving the two-echelon load-dependent location-routing problem with pick-up and delivery}, volume = {139}, year = {2025}, }
• (2025). Quantum-inspired metaheuristic algorithms for industry 4.0: A scientometric analysis. EAAI, 139, 109635. (www)

  @article{ , abstract = {Quantum-inspired Metaheuristic algorithms have redefined non-deterministic polynomial time hard optimization challenges by leveraging quantum mechanics principles. These algorithms herald a broad range of application scenarios in Industry 4.0 and offer feasible time solutions for complex, large-scale industrial landscapes. The potential benefits provided by the quantum-inspired metaheuristic algorithms have accelerated the scientific advancements in this domain. Consequently, the present research contributes to the existing knowledge base by presenting the intellectual landscape through scientometric and systematic literature analysis. The study is conducted on the dataset derived from the Scopus and Web of Science databases, covering 2001 to 2023. The study employs co-citation and co-occurrence analyses to discern prominent research topics, emerging research frontiers, significant authors, and the most collaborating countries. The research findings underscore that electric vehicles, energy efficiency, and combinatorial optimization are prominent research topics, while carbon emission, resource management, and path planning are burgeoning areas of exploration in this knowledge domain. The intricate and entangled network linkage determines that the research community in this domain fosters a dynamic and synergistic relationship. Overall, the pivotal insights and the research challenges articulated in this article offer valuable insights to researchers and the academic community, aiding in discerning the intellectual terrain and emerging research patterns in quantum-inspired metaheuristic algorithms. This, in turn, fosters the advancement of innovation and facilitates well-informed decision-making within this evolving research paradigm.}, archive = {J_EAAI}, author = {Pooja and Sandeep Kumar Sood}, doi = {10.1016/j.engappai.2024.109635}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109635}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Quantum-inspired metaheuristic algorithms for industry 4.0: A scientometric analysis}, volume = {139}, year = {2025}, }
• (2025). Diameter-adjustable mandrel for thin-wall tube bending and its domain knowledge-integrated optimization design framework. EAAI, 139, 109634. (www)

  @article{ , abstract = {In response to the growing demand for small-batch bending tube production, traditional bending dies require separate customization for each tube size, resulting in extended design cycles and high costs. To meet bending requirements for tubes of different diameters using a single mandrel, a novel adjustable diameter mechanism (DAM) and its optimization design method are proposed. Initially, the DAM based on a planetary bevel gear-screw transmission set is developed for bending tubes of varying diameters. Subsequently, a domain knowledge-integrated optimization design framework is introduced. To reduce the cost of acquiring training samples for training surrogate models, a monotonicity-constrained neural network based on cascade boosting architecture (CB-MCNN) is introduced that enhances prediction accuracy while maintaining monotonicity. To improve the optimization speed and quality of Evolutionary Algorithms (EAs), a domain knowledge-guided EA (DK-EA) method is proposed, incorporating domain knowledge into the population initialization phase. The results indicate that: (1) CB-MCNN outperforms traditional methods and shows excellent performance on small-sample datasets. (2) DK-EA accelerates optimization processes and produces better outcomes. As a result, the domain knowledge-integrated optimization design framework enables the DAM to achieve a wider diameter variation range and enhanced reliability. The optimized DAM demonstrates the capability to bend tubes with diameters of 46–60 mm.}, archive = {J_EAAI}, author = {Zili Wang and Jie Li and Xiaojian Liu and Shuyou Zhang and Yaochen Lin and Jianrong Tan}, doi = {10.1016/j.engappai.2024.109634}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109634}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Diameter-adjustable mandrel for thin-wall tube bending and its domain knowledge-integrated optimization design framework}, volume = {139}, year = {2025}, }
• (2025). An integrated outranking technique with spherical fuzzy rough numbers for the treatment of cadmium-contaminated water problem in china. EAAI, 139, 109633. (www)

  @article{ , abstract = {The Chinese economy is one of the largest and most dynamic economies in the world. Over the past few decades, China has experienced rapid economic growth from agrarian to industrial powerhouse fueled by manufacturing, exports, and services. However, this rapid growth has also brought about challenges, including environmental issues like water contamination. The indulgence of cadmium metal in regular used water can cause serious health issues, including kidney damage and cancer. Many strategies have been implemented for treatment of water contamination. The main focus of this research is to introduce a novel methodology for treatment of cadmium contaminated water problem in China. This study seeks to demonstrate the multi-criteria group decision-making ability based on the outranking relations within the confines of a contemporary, well-organized and extremely flexible model of spherical fuzzy rough numbers. Spherical fuzzy rough numbers, amalgamation of rough numbers with traditional spherical fuzzy numbers, make the use of membership, non-membership and neutral membership degrees along with the manipulation of the subjectivity and reliance on objective uncertainties. The combination of spherical fuzzy rough numbers with an outranking multi-criteria group decision making technique, Elimination and Choice Expressing Reality, integrates spherical fuzzy logic to handle uncertainty and imprecision in multi-criteria decision-making. This approach captures degrees of uncertainty and hesitancy with spherical fuzzy numbers, improving the handling of imprecise information. The working mechanism involves generation of outranking relations among alternatives by comparing predominant and subdominant options, calculating score degrees, concordance and discordance sets, and incorporating subjective spherical fuzzy rough criteria weights. Unlike traditional methods that use crisp or conventional fuzzy numbers, this technique provides a more reliable and flexible evaluation by integrating rough set theory for better handling of imprecision and uncertainty. Finally, an outranking graph is drawn that points from the supreme option to inferior one. The legitimacy of the proposed technique is, then, testified by making its comparison with other existing techniques.}, archive = {J_EAAI}, author = {Muhammad Akram and Maheen Sultan and Cengiz Kahraman}, doi = {10.1016/j.engappai.2024.109633}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109633}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An integrated outranking technique with spherical fuzzy rough numbers for the treatment of cadmium-contaminated water problem in china}, volume = {139}, year = {2025}, }
• (2025). Time-aware cross-domain point-of-interest recommendation in social networks. EAAI, 139, 109630. (www)

  @article{ , abstract = {Point-of-Interest recommendation within the single domain is quite easy compared to the cross-domain recommendation, as there is an acute dearth of check-in records for the target regions, aggravating the cold start problem. We propose a self-ensembled contextual Thompson sampling for cross-domain Point-of-Interest recommendation to solve this. This approach utilizes user preference transfer and user preference drift in the target domain for enhanced recommendation by deploying enhanced contextual sampling. As the user-context pairs of the target domain are not labeled for the given user, domain adaptation is highly sought. The approach has four major steps: i) Mining Point-of-Interests based on the long-term preferences of the target user and the user with a similar trajectory in the source domain, ii) Computing rewards for the Point-of-Interests in the source domain using multi-layer perceptron, iii) Estimate the rewards for unlabeled Point-of-Interests in the target domain and iv) Form the ensemble of rewards that are used to decide the final arm pulls. The rewards obtained for Point-of-Interests in the source and target domain are combined to form an ensemble of rewards with the help of self ensembling domain adaptation technique. Each Point-of-Interest in the ensemble rewards is termed an arm of action. We use this ensemble of rewards to control the diversity measure and the switching probability of the various arms, potential Point-of-Interests, in the contextual Thompson Sampling. Contextual Thompson sampling is modified to incorporate exploitation-exploration tradeoffs using this reward ensemble. The implicit weight measure of the different arms decides the probability of exploitation or exploration. The final arm pulls results in the final Point-of-Interest recommendation. For experimentation, we have used two real-world datasets, namely, Gowalla and Foursquare, and extracted the data for seven domains. We have obtained an accuracy of approximately 65% for Point-of-Interest recommendations on cold-start users.}, archive = {J_EAAI}, author = {Malika Acharya and Krishna Kumar Mohbey}, doi = {10.1016/j.engappai.2024.109630}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109630}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Time-aware cross-domain point-of-interest recommendation in social networks}, volume = {139}, year = {2025}, }
• (2025). Graph regularized discriminative nonnegative matrix factorization. EAAI, 139, 109629. (www)

  @article{ , abstract = {It is well known that both the label information and the local geometry structure information are very important for image data clustering and classification. However, nonnegative matrix factorization (NMF) and its variants do not fully utilize the information or only use one of them. This paper presents a graph regularized discriminative nonnegative matrix factorization (GDNMF) for image data clustering, in which the local geometrical structure and label information of the observed samples are thoroughly considered. In the objective function of NMF, two constraint terms are added to preserve the above information. One is a sparse graph, which is adaptively constructed to obtain the local geometrical structure information. The other is data label information, which is used to capture discriminative information of the original data. By using local and label information, the proposed regularized discriminative nonnegative matrix factorization indeed improves the discrimination power of matrix decomposition. In addition, the F -norm formulation based cost function of regularized discriminative nonnegative matrix factorization is given, and the update rules for the optimization function of regularized discriminative nonnegative matrix factorization are proved. The experiment results on several public image datasets demonstrate the effectiveness of GDNMF algorithm. The innovation of this paper lies in extending unsupervised NMF to semi-supervised case and adaptively capturing the local structure of data based on sparse graph. However, the proposed method does not take into account the challenges of multiview data processing.}, archive = {J_EAAI}, author = {Zhonghua Liu and Fa Zhu and Hao Xiong and Xingchi Chen and Danilo Pelusi and Athanasios V. Vasilakos}, doi = {10.1016/j.engappai.2024.109629}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109629}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Graph regularized discriminative nonnegative matrix factorization}, volume = {139}, year = {2025}, }
• (2025). Model-agnostic local explanation: Multi-objective genetic algorithm explainer. EAAI, 139, 109628. (www)

  @article{ , abstract = {Late detection of plant diseases leads to irreparable losses for farmers, threatening global food security, economic stability, and environmental sustainability. This research introduces the Multi-Objective Genetic Algorithm Explainer (MOGAE), a novel model-agnostic local explainer for image data aimed at the early detection of citrus diseases. MOGAE enhances eXplainable Artificial Intelligence (XAI) by leveraging the Non-dominated Sorting Genetic Algorithm II (NSGA-II) with an adaptive Bit Flip Mutation (BFM) incorporating densify and sparsify operators to adjust superpixel granularity automatically. This innovative approach simplifies the explanation process by eliminating several critical hyperparameters required by traditional methods like Local Interpretable Model-Agnostic Explanations (LIME). To develop the citrus disease classification model, we preprocess the leaf dataset through stratified data splitting, oversampling, and augmentation techniques, then fine-tuning a pre-trained Residual Network 50 layers (ResNet50) model. MOGAE’s effectiveness is demonstrated through comparative analyses with the Ensemble-based Genetic Algorithm Explainer (EGAE) and LIME, showing superior accuracy and interpretability using criteria such as numeric accuracy of explanation and Number of Function Evaluations (NFE). We assess accuracy both intuitively and numerically by measuring the Euclidean distance between expert-provided explanations and those generated by the explainer. The appendix also includes an extensive evaluation of MOGAE on the melanoma dataset, highlighting its versatility and robustness in other domains. The related implementation code for the fine-tuned ResNet50 and MOGAE is available at https://github.com/KhaosResearch/Plant-disease-explanation .}, archive = {J_EAAI}, author = {Hossein Nematzadeh and José García-Nieto and Sandro Hurtado and José F. Aldana-Montes and Ismael Navas-Delgado}, doi = {10.1016/j.engappai.2024.109628}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109628}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Model-agnostic local explanation: Multi-objective genetic algorithm explainer}, volume = {139}, year = {2025}, }
• (2025). Artificial intelligent pancreas for type 1 diabetic patients using adaptive type 3 fuzzy fault tolerant predictive control. EAAI, 139, 109627. (www)

  @article{ , abstract = {In this paper, the design methodology of artificial intelligent pancreas is presented. Accurate regulation of blood glucose levels in type 1 diabetic patients is of great importance in the presence of possible faults caused by sensor measurements. Regulation of blood glucose levels using a type 3 fuzzy predictive controller in type 1 diabetic patients in the presence of sensor faults is considered. The proposed structure includes a main control structure and a virtual dynamic, in which the main structure includes a fuzzy identifier, predictive controller, and an adaptive compensator, and the virtual structure is used to identify the sensor faults. Glucose is unknown in the dynamics of type 1 diabetes and is estimated on-line using a type 3 fuzzy system. Also, Lyapunov stability analysis is used to design the adaptive compensator to ensure the stability of the closed-loop system. The proposed methodology is evaluated based on Bergman’s minimum model for different patients under various parametric uncertainties and disturbances.}, archive = {J_EAAI}, author = {Arman Khani and Peyman Bagheri and Mahdi Baradarannia and Ardashir Mohammadzadeh}, doi = {10.1016/j.engappai.2024.109627}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109627}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Artificial intelligent pancreas for type 1 diabetic patients using adaptive type 3 fuzzy fault tolerant predictive control}, volume = {139}, year = {2025}, }
• (2025). Adaptive patch contrast for weakly supervised semantic segmentation. EAAI, 139, 109626. (www)

  @article{ , abstract = {Weakly Supervised Semantic Segmentation (WSSS), using only image-level labels, has garnered significant attention due to its cost-effectiveness. Typically, the framework involves using image-level labels as training data to generate pixel-level pseudo-labels with refinements. Recently, methods based on Vision Transformers (ViT) have demonstrated superior capabilities in generating reliable pseudo-labels, particularly in recognizing complete object regions. However, current ViT-based approaches have some limitations in the use of patch embeddings, being prone to being dominated by certain abnormal patches, as well as many multi-stage methods being time-consuming and lengthy in training, thus lacking efficiency. Therefore, in this paper, we introduce a novel ViT-based WSSS method named Adaptive Patch Contrast (APC) that significantly enhances patch embedding learning for improved segmentation effectiveness. APC utilizes an Adaptive-K Pooling (AKP) layer to address the limitations of previous max pooling selection methods. Additionally, we propose a Patch Contrastive Learning (PCL) to enhance patch embeddings, thereby further improving the final results. We developed an end-to-end single-stage framework without CAM, which improved training efficiency. Experimental results demonstrate that our method performs exceptionally well on public datasets, outperforming other state-of-the-art WSSS methods with a shorter training time.}, archive = {J_EAAI}, author = {Wangyu Wu and Tianhong Dai and Zhenhong Chen and Xiaowei Huang and Jimin Xiao and Fei Ma and Renrong Ouyang}, doi = {10.1016/j.engappai.2024.109626}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109626}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive patch contrast for weakly supervised semantic segmentation}, volume = {139}, year = {2025}, }
• (2025). SenticNet and abstract meaning representation driven attention-gate semantic framework for aspect sentiment triplet extraction. EAAI, 139, 109625. (www)

  @article{ , abstract = {Aspect sentiment triplet extraction aims to analyze aspect-level sentiment in the form of triplets, including extracting aspect-opinion pairs and predicting the sentiment polarities of these pairs. Many recent works rely on syntactic information (e.g. part-of-speech and syntactic dependency relation) to handle this semantic task, which ignores uncommon part-of-speech items and matches semantically unrelated words. To overcome these drawbacks, we propose a SenticNet and Abstract Meaning Representation (AMR) driven Attention-Gate semantic framework (SAAG), which introduces semantic sentiment knowledge SenticNet and semantic structure AMR as semantic information to replace syntactic information. To highlight the affective meanings in words, an affective-driven attention mechanism is designed to emphasizes sentiment intent within word representations. To match semantically related words, the designed AMR-driven gate mechanism balances the word pair expressions under varying semantic contexts. Extensive experiments on two public datasets demonstrate the effectiveness of our approach.}, archive = {J_EAAI}, author = {Xiaowen Sun and Jiangtao Qi and Zhenfang Zhu and Meng Li and Hongli Pei and Jing Meng}, doi = {10.1016/j.engappai.2024.109625}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109625}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SenticNet and abstract meaning representation driven attention-gate semantic framework for aspect sentiment triplet extraction}, volume = {139}, year = {2025}, }
• (2025). A generative-adversarial-network-based temporal raw trace data augmentation framework for fault detection in semiconductor manufacturing. EAAI, 139, 109624. (www)

  @article{ , abstract = {In modern semiconductor manufacturing, where sophisticated process control mechanisms are standard, processing tools are equipped with sensors that generate vast amounts of raw trace data for process monitoring and fault detection. However, one of the major challenges data scientists face is the scarcity of sufficient raw trace data for defective wafers, creating an imbalance that complicates the training of machine learning models for effective fault detection. To address this issue, this paper proposes novel data augmentation structures and strategies utilizing Cycle Generative Adversarial Networks (CycleGANs) as an artificial intelligence application to synthesize temporal raw trace data for defective wafers. The effectiveness of these methods is demonstrated using a real-world dataset from the thin-film process in semiconductor fabrication. Several machine learning classification models—Gaussian Naive Bayes, Adaptive Boosting, eXtreme Gradient Boosting, and Light Gradient Boosting Machine—are employed to evaluate the performance of the augmented data. The paper identifies the optimal augmentation structure and strategy to enhance classification performance within the CycleGAN-based framework. For the thin-film processing dataset under study, the best classification performance achieves an accuracy rate of up to 99.30%, with a notably low false negative rate of 6.45%.}, archive = {J_EAAI}, author = {Shu-Kai S. Fan and Wei-Yu Chen}, doi = {10.1016/j.engappai.2024.109624}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109624}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A generative-adversarial-network-based temporal raw trace data augmentation framework for fault detection in semiconductor manufacturing}, volume = {139}, year = {2025}, }
• (2025). Data-driven estimation of the amount of under frequency load shedding in small power systems. EAAI, 139, 109617. (www)

  @article{ , abstract = {This paper presents a data-driven methodology for estimating under frequency load shedding (UFLS) in small power systems. UFLS plays a vital role in maintaining system stability by shedding load when the frequency drops below a specified threshold following loss of generation. Using a dynamic system frequency response (SFR) model we generate different values of UFLS (i.e., labels) predicated on a set of carefully selected operating conditions (i.e., features). Machine learning (ML) algorithms are then applied to learn the relationship between chosen features and the UFLS labels. A novel regression tree and the Tobit model are suggested for this purpose and we show how the resulting non-linear model can be directly incorporated into a mixed integer linear programming (MILP) problem. The trained model can be used to estimate UFLS in security-constrained operational planning problems, improving frequency response, optimizing reserve allocation, and reducing costs. The methodology is applied to the La Palma island power system, demonstrating its accuracy and effectiveness. The results confirm that the amount of UFLS can be estimated with the mean absolute error (MAE) as small as 0.179 MW for the whole process, with a model that is representable as a MILP for use in scheduling problems such as unit commitment among others.}, archive = {J_EAAI}, author = {Mohammad Rajabdorri and Matthias C.M. Troffaes and Behzad Kazemtabrizi and Miad Sarvarizadeh and Lukas Sigrist and Enrique Lobato}, doi = {10.1016/j.engappai.2024.109617}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109617}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven estimation of the amount of under frequency load shedding in small power systems}, volume = {139}, year = {2025}, }
• (2025). Decoding text from electroencephalography signals: A novel hierarchical gated recurrent unit with masked residual attention mechanism. EAAI, 139, 109615. (www)

  @article{ , abstract = {Progress in both neuroscience and natural language processing has opened doors for investigating brain to text techniques to reconstruct what individuals see, perceive, or focus on from human brain activity patterns. Non-invasive decoding, utilizing electroencephalography (EEG) signals, is preferred due to its comfort, cost-effectiveness, and portability. In brain-to-text applications, a pressing need has arisen to develop effective models that can accurately capture the intricate details of EEG signals, such as global and local contextual information and long-term dependencies. In response to this need, we propose the Hierarchical Gated Recurrent Unit with Masked Residual Attention Mechanism (HGRU-MRAM) model, which ingeniously combines the hierarchical structure and the masked residual attention mechanism to deliver a robust brain-to-text decoding system. Our experimental results on the ZuCo dataset demonstrate that this model significantly outperforms existing baselines, achieving state-of-the-art performance with Bilingual Evaluation Understudy Score (BLEU), Recall-Oriented Understudy for Gisting Evaluation (ROUGE), US National Institute of Standards and Technology Metric (NIST), Metric for Evaluation of Translation with Explicit Ordering (METEOR), Translation Edit Rate (TER), and BiLingual Evaluation Understudy with Representations from Transformers (BLEURT) scores of 48.29, 34.84, 4.07, 34.57, 21.98, and 40.45, respectively. The code is available at https://github.com/qpuchen/EEG-To-Sentence .}, archive = {J_EAAI}, author = {Qiupu Chen and Yimou Wang and Fenmei Wang and Duolin Sun and Qiankun Li}, doi = {10.1016/j.engappai.2024.109615}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109615}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Decoding text from electroencephalography signals: A novel hierarchical gated recurrent unit with masked residual attention mechanism}, volume = {139}, year = {2025}, }
• (2025). A novel hybrid data-driven domain generalization approach with dual-perspective feature fusion for intelligent fault diagnosis. EAAI, 139, 109614. (www)

  @article{ , abstract = {Domain generalization-based fault diagnosis (DGFD) approaches do not require access to the target domain during model training, but they usually rely on numerous labeled source domain data. However, only few labeled source domain data can be obtained in actual diagnosis scenarios. Therefore, a novel hybrid data-driven domain generalization (DG) approach with dual-perspective feature fusion for intelligent fault diagnosis (FD) is proposed. Firstly, to solve the problem of scarce training samples in the source domains, the rolling bearing (RB) and the gear simulated vibration models are established to generate numerous labeled simulated vibration data, and the improved auxiliary classifier generative adversarial network (ACGAN) is used to effectively balance the simulated and real data. Secondly, a simulated and real data-driven DG network that fuses intra-domain invariant features and mutually-invariant features between domains (SRDGN-IM) is proposed, where the intra-domain invariant features are learned through distillation idea and the mutually-invariant features are learned through adversarial training, which can make the diagnosis model better learn the key generalization features from source domains to obtain more accurate diagnosis results. Finally, a series of DG experiments are conducted on the gearbox and bearing datasets, and the average FD accuracies of the proposed approach reach 87.45% and 89.10% respectively under different DG tasks.}, archive = {J_EAAI}, author = {Lanjun Wan and Jian Zhou and Jiaen Ning and Yuanyuan Li and Changyun Li}, doi = {10.1016/j.engappai.2024.109614}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109614}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel hybrid data-driven domain generalization approach with dual-perspective feature fusion for intelligent fault diagnosis}, volume = {139}, year = {2025}, }
• (2025). Practical framework for generative on-branch soybean pod detection in occlusion and class imbalance scenes. EAAI, 139, 109613. (www)

  @article{ , abstract = {The number of pods per plant can serve as an effective indicator of soybean yield, and accurately determining this is essential for evaluating high-quality soybean varieties. However, traditional manual pod counting is time-consuming and laborious. Although deep learning-based pod detection methods have attracted much attention, there are still considerable challenges for the effective detection of pods in occlusion and class imbalance scenes. As a remedy, this study proposes a framework that leverages synthetic pod image generation and multi-stage transfer learning to generate detection model of on-branch soybean pods in complex scenes. This framework employs a novel pipeline: initially separating individual pods from non-occluded pod images in an off-branch pod training set, then using these to generate synthetic datasets with diverse pod features. Next, a multi-stage transfer learning method is employed to train an on-branch pod detection model, leveraging both real and synthetic datasets to enhance pod feature extraction in complex scenes. The detection model of proposed framework, YOLOv7-tiny (tiny version of You Only Look Once v7), integrates an angle prediction module based on Circular Smooth Label for rotated object detection, Coordinate Attention modules for enhanced feature extraction and Minimum Point Distance Intersection over Union Loss for precise bounding box perception. Experimental results show that proposed framework achieves an 81.1% mAP (mean Average Precision) for detecting on-branch pods in complex scenes, surpassing the best-performing model by 23.7%. This proposed method presents an effective solution for complex on-branch pod detection, having great potential of serving as robust pipeline for similar agricultural tasks.}, archive = {J_EAAI}, author = {Kanglei Wu and Tan Wang and Yuan Rao and Xiu Jin and Xiaobo Wang and Jiajia Li and Zhe Zhang and Zhaohui Jiang and Xing Shao and Wu Zhang}, doi = {10.1016/j.engappai.2024.109613}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109613}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Practical framework for generative on-branch soybean pod detection in occlusion and class imbalance scenes}, volume = {139}, year = {2025}, }
• (2025). Parallax-aware dual-view feature enhancement and adaptive detail compensation for dual-pixel defocus deblurring. EAAI, 139, 109612. (www)

  @article{ , abstract = {Defocus deblurring using dual-pixel sensors has gathered significant attention in recent years. However, current methodologies have not adequately addressed the challenge of defocus disparity between dual views, resulting in suboptimal performance in recovering details from severely defocused pixels. To counteract this limitation, we introduce in this paper a parallax-aware dual-view feature enhancement and adaptive detail compensation network (PA-Net), specifically tailored for dual-pixel defocus deblurring task. Our proposed PA-Net leverages an encoder–decoder architecture augmented with skip connections, designed to initially extract distinct features from the left and right views. A pivotal aspect of our model lies at the network’s bottleneck, where we introduce a parallax-aware dual-view feature enhancement based on Transformer blocks, which aims to align and enhance extracted dual-pixel features, aggregating them into a unified feature. Furthermore, taking into account the disparity and the rich details embedded in encoder features, we design an adaptive detail compensation module to adaptively incorporate dual-view encoder features into image reconstruction, aiding in restoring image details. Experimental results demonstrate that our proposed PA-Net exhibits superior performance and visual effects on the real-world dataset.}, archive = {J_EAAI}, author = {Yuzhen Niu and Yuqi He and Rui Xu and Yuezhou Li and Yuzhong Chen}, doi = {10.1016/j.engappai.2024.109612}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109612}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Parallax-aware dual-view feature enhancement and adaptive detail compensation for dual-pixel defocus deblurring}, volume = {139}, year = {2025}, }
• (2025). An intelligent model approach for leakage detection of modified atmosphere pillow bags. EAAI, 139, 109611. (www)

  @article{ , abstract = {Modified atmosphere pillow bags have been widely used to package various food products due to their advantages for preservation and shipment. Sealing defects are statistically inevitable, although modern packaging machinery and manual inspection utilized by manufacturers continue reducing the leakage probability. Hence the bag contents may spoil if the seal is broken. Instead of manual inspection and various destructive methods utilized by factories, this study introduces non-destructive leakage detection using deep learning methods. Firstly, a squeezing method is developed to aggravate the feature difference between positive samples and negative samples without destroying the bag content, thus 2160 images of three different pillow bags are acquired to establish dataset. Secondly, the deep learning model Vision Transformer (ViT) is deployed and studied so that feasibility of computer vision method is verified. Then the Semantic segmentation and Contour Extraction model combining ViT (SCE-ViT) is proposed and improved to the Multi-dimensional Fusion model (SCE-MdF). The accuracies of SCE-MdF reached 97.5%, 97.5%, and 97.5%, respectively. The F1-scores of SCE-MdF reached 97.6%, 97.6%, and 97.4%, respectively. Compared to averaged accuracies of SCE-ViT, accuracies introduced in the ultimate model SCE-MdF improved by 19.17%, 5.84%, and 11.67%, respectively. Therefore, combination of unique squeezing method and Semantic segmentation Contour Extraction with Multi-dimensional Fused ViT, is eventually validated viable on leakage detection of modified atmosphere pillow bags. Hence a cost-effective, efficient and non-destructive leakage detection method for modified atmosphere pillow bags in relevant industry is introduced, filling a gap between artificial intelligence and food packaging industry.}, archive = {J_EAAI}, author = {Xiangdong Guo and Jingfa Yao and Guoyu Yan and Guifa Teng}, doi = {10.1016/j.engappai.2024.109611}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109611}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An intelligent model approach for leakage detection of modified atmosphere pillow bags}, volume = {139}, year = {2025}, }
• (2025). Transformer-based partner dance motion generation. EAAI, 139, 109610. (www)

  @article{ , abstract = {With the rapid development of information technology, particularly the recent breakthroughs in the field of artificial intelligence, the field of dance education has experienced an unprecedented transformation. Today, an increasing number of dance enthusiasts can experience an efficient and personalized immersive dance learning process via advanced artificial intelligence applications. Compared with the individual practice of solo dance, the learning process of partner dancing is more complex. Partner dancing requires not only a coordinated dance partner but also the abilities of both individuals to accurately follow each other's movements in real time as well as adjust and present the correct dance postures. In this study, we developed an innovative real-time virtual reality dance training framework tailored for interactive entertainment and teaching. Our framework used an enhanced transformer model that could generate partner movement sequences based on user movements. Furthermore, we established a somatosensory virtual reality interaction environment by integrating somatosensory devices to capture user movements in real time. These movement sequences were then fed into a network to produce partner movement sequences by generating virtual characters and creating an interactive dance learning system that can offer real-time feedback from a virtual partner. Our experiments validated the robustness and effectiveness of the proposed sequence generation model. In addition, we incorporated Laban movement analysis to explore the semantic representations of partner dances and devised a comprehensive set of evaluation indicators to assess the quality of the generated partner dance movements. The rigorous testing of each component and actual user testing demonstrated the effectiveness of the system. The system offered users rich interactive feedback and a scientifically grounded approach for learning to dance with a partner. Therefore, this study offers significant prospects for both intelligent dance teaching and human-computer interaction in virtual reality among other applications.}, archive = {J_EAAI}, author = {Ying Wu and Zizhao Wu and Chengtao Ji}, doi = {10.1016/j.engappai.2024.109610}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109610}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer-based partner dance motion generation}, volume = {139}, year = {2025}, }
• (2025). Less is more: A semi-supervised fine-grained object detection for satellite video. EAAI, 139, 109609. (www)

  @article{ , abstract = {Satellite video fine-grained object detection aims to accurately locate and finely identify objects of interest in sequence frames with a wide range of application requirements. Powerful detection performance mainly relies on abundant and accurate labels, however, the high cost of labeling results in the limited availability of high-quality satellite video-labeled data. In this paper, a semi-supervised fine-grained object detection method for satellite video is proposed for low-cost object detection engineering applications in remote sensing. Firstly, we design a dual-branch feature extraction network that utilizes satellite video frames and corresponding dense optical flow as joint inputs. After model training with the initial labeled frame, the network extracts appearance and motion features concurrently. Secondly, an attentional fusion module is proposed to enrich the feature representation of the object by combining channel and spatial attention to appearance features and nonlinear interactions with motion features. Finally, supervised contrastive learning is added to the network’s classification head to improve the network’s ability to distinguish between fine-grained object categories through feature encoding and similarity metrics. Evaluation results on the large-scale satellite video dataset demonstrate that under the semi-supervised learning constraints of training using only the initial frame labels, the proposed method outperforms the other 15 advanced fully-supervised detection methods, with less labeling cost and more robust detection performance, and meets the efficiency requirements of real-time satellite video detection. The relevant code is released at https://github.com/dagezhuang/SFOD-SAT .}, archive = {J_EAAI}, author = {Zhuang Zhou and Shengyang Li}, doi = {10.1016/j.engappai.2024.109609}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109609}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Less is more: A semi-supervised fine-grained object detection for satellite video}, volume = {139}, year = {2025}, }
• (2025). Elite-based multi-objective improved iterative local search algorithm for time-dependent vehicle-drone collaborative routing problem with simultaneous pickup and delivery. EAAI, 139, 109608. (www)

  @article{ , abstract = {This paper focuses on solving a time-dependent multi-objective vehicle-drone collaborative routing problem with simultaneous pickup and delivery, in which multiple visits per drone trip, simultaneous pickup and delivery, soft time windows, and time-dependent road network are considered. With the maximum completion time and total violation time as the optimization objectives, we first formulate the mathematical model of the problem. Then, in order to effectively solve the problem, an Elite-based multi-objective improved iterative local search algorithm developed within a collaborative optimization framework is proposed. Specifically, the multi-objective problem is decomposed into two subproblems, each of which is solved by minimizing a single objective. Meanwhile, the algorithm uses an elite set to record non-dominated solutions, guide the search, and achieve information exchange between subproblems. In the proposed algorithm, an individual is encoded as a vector consisting of two parts, a customer sequence and a sequence recording the customers' visiting modes, and can be decoded into subroutes for the vehicle and drone. To guarantee the feasibility of the solution, an adjustment method is proposed to repair the individual. In addition, based on individual representation and problem characteristics, six neighborhood structures are designed, through which new individuals can be generated. Then, by using the neighborhood structures, a problem-specific local search strategy and an iterative local search strategy are proposed to improve the search capability of the algorithm. Experimental tests and analyses demonstrate the correctness of the established mathematical model and the effectiveness of the proposed algorithm in solving this complex vehicle-drone collaborative routing problem.}, archive = {J_EAAI}, author = {Haohao Duan and Xiaoling Li and Guanghui Zhang and Yanxiang Feng and Qingchang Lu}, doi = {10.1016/j.engappai.2024.109608}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109608}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Elite-based multi-objective improved iterative local search algorithm for time-dependent vehicle-drone collaborative routing problem with simultaneous pickup and delivery}, volume = {139}, year = {2025}, }
• (2025). Selection of internet of things-enabled sustainable real-time monitoring strategies for manufacturing processes using a disc spherical fuzzy Schweizer–Sklar aggregation model. EAAI, 139, 109607. (www)

  @article{ , abstract = {The emergence of the Internet of Things (IoT) for monitoring in real-time is geared towards sustainable energy consumption practices by taking control over energy loss. The promising potential of current IoT real-time monitoring systems paves the way for future developments in monitoring devices with eco-friendly sensing capabilities. As a result, the creation of effective IoT real-time monitoring devices targeted at decreasing energy loss becomes crucial. This modeling procedure falls under the realm of multiple-attribute group decision-making (MAGDM), aiming to integrate the Schweizer–Sklar (SS) τ -norm and τ -conorm within the disc spherical fuzzy (D-SF) framework. The objective is to enhance the flexibility of D-SF in dealing with intricate and uncertain data. The core focus of this research is on deriving SS τ -norm and τ -conorm for D-SF data, consequently introducing innovative aggregation operators. The article offers the fundamental D-SF operations using SS aggregation operators in a systematic manner, with thorough theorem justifications. A new MAGDM tool is presented, created simply to manage ambiguous and imprecise data utilizing the suggested operators. Our model is specifically designed to tackle the critical issue of reducing energy loss in IoT real-time monitoring systems. The research not only focuses on model accuracy but also emphasizes its effectiveness in solving this pressing problem, demonstrating significant advancements in sustainable energy practices. Moreover, the proposed aggregation operators are subjected to a comparative analysis. This comprehensive comparison not only enhances the operators’ efficacy but also underscores their relevance in real-world decision-making scenarios.}, archive = {J_EAAI}, author = {Shahzaib Ashraf and Muhammad Naeem and Wania Iqbal and Hafiz Muhammad Athar Farid and Hafiz Muhammad Shakeel and Vladimir Simic and Erfan Babaee Tirkolaee}, doi = {10.1016/j.engappai.2024.109607}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109607}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Selection of internet of things-enabled sustainable real-time monitoring strategies for manufacturing processes using a disc spherical fuzzy Schweizer–Sklar aggregation model}, volume = {139}, year = {2025}, }
• (2025). Hybrid multi-attention transformer for robust video object detection. EAAI, 139, 109606. (www)

  @article{ , abstract = {Video object detection (VOD) is the task of detecting objects in videos, a challenge due to the changing appearance of objects over time, leading to potential detection errors. Recent research has addressed this by aggregating features from neighboring frames and incorporating information from distant frames to mitigate appearance deterioration. However, relying solely on object candidate regions in distant frames, independent of object position, has limitations, as it depends heavily on the performance of these regions and struggles with deteriorated appearances. To overcome these challenges, we propose a novel Hybrid Multi-Attention Transformer (HyMAT) module as our main contribution. HyMAT enhances relevant correlations while suppressing flawed information by searching for an agreement between whole correlation vectors. This module is designed for flexibility and can be integrated into both self- and cross-attention blocks to significantly improve detection accuracy. Additionally, we introduce a simplified Transformer-based object detection framework, named Hybrid Multi-Attention Object Detection (HyMATOD), which leverages competent feature reprocessing and target-background embeddings to more effectively utilize temporal references. Our approach demonstrates state-of-the-art performance, as evaluated on the ImageNet video object detection benchmark (ImageNet VID) and the University at Albany DEtection and TRACking (UA-DETRAC) benchmarks. Specifically, our HyMATOD model achieves an impressive 86.7% mean Average Precision (mAP) on the ImageNet VID dataset, establishing its superiority and practicality for video object detection tasks. These results underscore the significance of our contributions to advancing the field of VOD.}, archive = {J_EAAI}, author = {Sathishkumar Moorthy and Sachin Sakthi K.S. and Sathiyamoorthi Arthanari and Jae Hoon Jeong and Young Hoon Joo}, doi = {10.1016/j.engappai.2024.109606}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109606}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid multi-attention transformer for robust video object detection}, volume = {139}, year = {2025}, }
• (2025). Metric learning-based generalized gaussian kernel for nonlinear classification. EAAI, 139, 109605. (www)

  @article{ , abstract = {Metric learning and kernel learning have attracted a lot of interest for nonlinear classifications in the artificial intelligence field. Most existing nonlinear classification methods are with preset kernel functions, while classification performances depend on kernel functions. To address these shortcomings, we propose a classification framework based on two-stage metric learning. Specifically, in the first stage, a robust metric is trained by applying a robust loss and the maximal margin principle. Then dual technology accompanied by an alternate iteration algorithm is adopted to optimize the problem, and the strong duality property is demonstrated theoretically. The resulting algorithm is convergent. In the second stage, a generalized Gaussian kernel is trained based on the learned metric. This is an adaptive kernel and its main properties are analyzed theoretically. Then with the learned kernel, a nonlinear classification framework is built to train the optimal separation hyperplane, as a classification decision rule. This can avoid effectively the limitation of k-nearest neighbor classifiers. In addition, we analyze the robustness and generalization error bound of the proposed method. Numerical experiments are conducted on two different databases. In the application database, experiments on different spectral regions show that the proposed framework outperforms significantly the traditional methods. Compared with traditional nonlinear methods, the proposed framework increases classification accuracy by an average of 5.36% in the application database. The statistical tests show the ranking of the proposed framework is the best of all methods. These experiments show the feasibility and effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Yidan Wang and Liming Yang}, doi = {10.1016/j.engappai.2024.109605}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109605}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Metric learning-based generalized gaussian kernel for nonlinear classification}, volume = {139}, year = {2025}, }
• (2025). MFLSCI: Multi-granularity fusion and label semantic correlation information for multi-label legal text classification. EAAI, 139, 109604. (www)

  @article{ , abstract = {Multi-label text classification tasks face challenges such as sample diversity, complexity, and the need for effective utilization of label correlations. In this paper, we propose a model that integrates multi-granularity fusion of text sequence features and label semantic correlation information. Our model leverages graph convolutional networks to extract label semantic correlation, which enhances classification performance for samples with similar labels and addresses label omission issues. Additionally, text convolutional neural networks are employed to extract multi-granularity sense group features from text sequences, calculate their similarity with semantic correlation label distributions, and dynamically adjust the similarity between text context and label information. This approach tackles the limitations of feature extraction in short texts and label confusion. We replace the original multi-hot label encoding in model training with a label distribution that fuses text multi-granularity sense group features and label correlation information, using a more precise encoding method for soft alignment based on label probability distributions. This enhances the model’s resilience to noisy data, avoiding the issue of assigning high-confidence probabilities to incorrect categories due to hard-coded supervision. Our model’s performance improvement on noisy datasets significantly surpasses that achieved by label smoothing. Extensive experiments on three legal text datasets and two generalized multi-label datasets demonstrate the model’s excellent performance. Our approach is applicable in various real-world scenarios, such as legal judgment prediction, news categorization, and recommendation systems, where accurate multi-label classification is crucial. Ablation and experiments on noisy datasets validate the model’s effectiveness and robustness.}, archive = {J_EAAI}, author = {Chunyun Meng and Yuki Todo and Cheng Tang and Li Luan and Zheng Tang}, doi = {10.1016/j.engappai.2024.109604}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109604}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MFLSCI: Multi-granularity fusion and label semantic correlation information for multi-label legal text classification}, volume = {139}, year = {2025}, }
• (2025). Self-learning guided residual shrinkage network for intelligent fault diagnosis of planetary gearbox. EAAI, 139, 109603. (www)

  @article{ , abstract = {The original vibration signals of the fault gear under different working conditions have a large distribution difference, and there will be insufficient feature extraction during fault diagnosis, which leads to the problem of low diagnostic accuracy. Therefore, a self-learning model based on residual shrinkage network (SLRSN) is proposed. The model constructs a deep residual shrinkage network as the main network for feature extraction of the original vibration signal to enhance the robustness of the model. Then self-believing loss and self-doubting loss are proposed to achieve self-confidence and suspicion of health status prediction. The first is self-confidence loss, which adopts sub-domain distribution adaptation to actively align learned cross-domain features. The second is self-doubt loss, which provides SLRSN with the ability to extricate from wrong experience. Finally, to mitigate the effects of negative transfer, a novel adaptative weight allocation mechanism is designed to recalibrate the weighting of each source domain sample. Through the experiment of two gearboxes, it is verified that the proposed SLRSN method has good diagnostic reliability under the condition of gear speed and load change.}, archive = {J_EAAI}, author = {Xingwang Lv and Jinrui Wang and Ranran Qin and Jihua Bao and Xue Jiang and Zongzhen Zhang and Baokun Han and Xingxing Jiang}, doi = {10.1016/j.engappai.2024.109603}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109603}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-learning guided residual shrinkage network for intelligent fault diagnosis of planetary gearbox}, volume = {139}, year = {2025}, }
• (2025). Advanced forecast models for the climate and energy crisis: The case of the california independent system operator. EAAI, 139, 109602. (www)

  @article{ , abstract = {Climate change across the globe, especially extreme temperature events, is increasing pressures on energy systems. The extraordinary situation that California and the West faced in late August and early September 2022, when record temperatures led to a spike in electricity demand, provided an important backdrop for the resilience and sustainability of clean energy technologies. The electricity market managed by the California Independent System Operator is considered in this study to examine the potential impacts on electricity demand spikes and system resilience. The methodology of the research involves analyzing the system operator's responses to electricity demand using advanced deep learning algorithms, convolutional neural network – long-short term memory and attention mechanism models. 1, 3 and 7-days forecasts of electricity demand were made using models in the day ahead market. In 1-day forecasts, while the former models have a mean absolute percentage error value of 12.40%, the latter model has a lower error rate of 10.36%. Overall findings obtained from various scenarios show that the long-short term memory – attention mechanism can more effectively understand complex patterns in energy demand and has the potential to increase system stability against such extreme weather events. The advanced horizon of the study offers an important perspective on how clean energy technologies, especially battery energy storage systems, can provide solutions to today's priority problems such as climate change and extreme temperature.}, archive = {J_EAAI}, author = {Merve Bulut and Hüseyin Aydilek and Mustafa Yasin Erten and Evrencan Özcan}, doi = {10.1016/j.engappai.2024.109602}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109602}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Advanced forecast models for the climate and energy crisis: The case of the california independent system operator}, volume = {139}, year = {2025}, }
• (2025). Decision-making systems improvement based on explainable artificial intelligence approaches for predictive maintenance. EAAI, 139, 109601. (www)

  @article{ , abstract = {To maintain the performance of the latest generation of onshore and offshore wind turbine systems, a new methodology must be proposed to enhance the maintenance policy. In this context, this paper introduces an approach to designing a decision support tool that combines predictive capabilities with anomaly explanations for effective IoT predictive maintenance tasks. Essentially, the paper proposes an approach that integrates a predictive maintenance model with an explicative decision-making system. The key challenge is to detect anomalies and provide plausible explanations, enabling human operators to determine the necessary actions swiftly. To achieve this, the proposed approach identifies a minimal set of relevant features required to generate rules that explain the root causes of issues in the physical system. It estimates that certain features, such as the active power generator, blade pitch angle, and the average water temperature of the voltage circuit protection in the generator’s sub-components, are particularly critical to monitor. Additionally, the approach simplifies the computation of an efficient predictive maintenance model. Compared to other deep learning models, the identified model provides up to 80% accuracy in anomaly detection and up to 96% for predicting the remaining useful life of the system under study. These performance metrics and indicators values are essential for enhancing the decision-making process. Moreover, the proposed decision support tool elucidates the onset of degradation and its dynamic evolution based on expert knowledge and data gathered through Internet of Things (IoT) technology and inspection reports. Thus, the developed approach should aid maintenance managers in making accurate decisions regarding inspection, replacement, and repair tasks. The methodology is demonstrated using a wind farm dataset provided by Energias De Portugal.}, archive = {J_EAAI}, author = {Lala Rajaoarisoa and Raubertin Randrianandraina and Grzegorz J. Nalepa and João Gama}, doi = {10.1016/j.engappai.2024.109601}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109601}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Decision-making systems improvement based on explainable artificial intelligence approaches for predictive maintenance}, volume = {139}, year = {2025}, }
• (2025). Integration of in-wheel motor sensorless systems and hierarchical direct yaw moment control for distributed drive electric vehicles. EAAI, 139, 109600. (www)

  @article{ , abstract = {Ensuring robust and reliable control of distributed vehicles powered by in-wheel motor systems poses a significant challenge due to the harsh operating environments and high costs of such motor systems. Poor motor control, parameter variations, and sensor malfunction under these conditions can compromise the vehicle yaw stability. Integrating permanent magnet synchronous motor (PMSM) sensorless systems with vehicle yaw moment control offers a cost-effective solution for this issue without wheel angular speed sensors while enhancing yaw stability. In this paper, a composite nonlinear feedback sliding mode controller that can enhance the PMSM speed response is proposed. The proposed scheme exhibits a rotor speed overshoot and transient time of only 0.64% and 0.07s, respectively, which are smaller and shorter compared with other methods under motor parameter changes. Subsequently, the key states and tire-road friction coefficients required for vehicle control were estimated using sensorless rotor speeds and unscented Kalman filters, enabling the integration of the PMSM sensorless system with the vehicle yaw moment control. Additionally, a fuzzy adaptive hybrid sliding mode method is presented for yaw moment control enhancement. This method maintained the smallest sideslip angle root mean square error during double lane changes (0.4192 deg) compared with other methods. Analysis results show that different motor controllers and parameter changes significantly affect the vehicle dynamics performance. The proposed integrated scheme is feasible and effectively enhances the yaw moment control via high-performance sensorless PMSM systems.}, archive = {J_EAAI}, author = {Xiaodong Wang and Maoping Ran and Xinglin Zhou}, doi = {10.1016/j.engappai.2024.109600}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109600}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integration of in-wheel motor sensorless systems and hierarchical direct yaw moment control for distributed drive electric vehicles}, volume = {139}, year = {2025}, }
• (2025). One test to predict them all: Rheological characterization of complex fluids via artificial neural network. EAAI, 139, 109598. (www)

  @article{ , abstract = {The rheological behavior of complex fluids, including thixotropy, viscoelasticity, and viscoplasticity, poses significant challenges in both measurement and prediction due to the transient nature of their stress responses. This study introduces an artificial neural network (ANN) designed to digitally characterize the rheology of complex fluids with unprecedented accuracy. By employing a data-driven approach, the ANN is trained using transient rheological tests with step inputs of shear rate. Once trained, the network adeptly captures the intricate dependencies of rheological properties on time and shear, enabling rapid and accurate predictions of various rheological tests. In contrast, traditional phenomenological structural kinetic constitutive models often fail to accurately describe the evolution of nonlinear rheological properties, particularly as material complexity increases. The ANN demonstrates high flexibility, reliability and robustness by accurately predicting transient rheology of varied materials with different shear histories. Our findings illustrate that ANNs can not only complement and validate traditional rheological characterization methods but also potentially replace them, thereby paving the way for more efficient material development and testing.}, archive = {J_EAAI}, author = {Ases Akas Mishra and Viney Ghai and Valentina Matovic and Dragana Arlov and Roland Kádár}, doi = {10.1016/j.engappai.2024.109598}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109598}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {One test to predict them all: Rheological characterization of complex fluids via artificial neural network}, volume = {139}, year = {2025}, }
• (2025). WIGNN: An adaptive graph-structured reasoning model for credit default prediction. EAAI, 139, 109597. (www)

  @article{ , abstract = {In credit default prediction, the main challenge is handling complex data structures and addressing data class imbalance. Given class imbalance and multi-dimensional data, general models find it difficult to fully explore the deep interdependencies within the data and the interaction effects between local and global. To overcome these challenges, this study proposes a Weighted Imbalanced Graph Neural Network (WIGNN) model that integrates adaptive graph structure inference with differential weight connectivity strategy, and the model solves the existing problems from the perspective of differential weight connectivity and graph balancing. Here, the weight connection uses the Gaussian kernel function to refine calculations and an adaptive percentile method to adjust sparsity, improving the understanding and efficiency of mining data connections. The weighted graph generated by this method can reflect the interaction between nodes and improve the model’s ability to analyse complex data structures. Based on this weighted graph, the graph imbalance module adopts a reinforcement learning-driven neighbour sampling strategy to adjust the sampling threshold automatically, optimizes the node embedding through message aggregation, and combines with a cost-sensitive matrix to improve classification accuracy and cost-effectiveness of the model on diverse credit datasets. We applied the WIGNN model to six real and class-imbalanced credit datasets, comparing it with 11 mainstream credit default prediction models. Evaluated using metrics Area Under the Curve (AUC), Geometric Mean (G-mean), and Accuracy. The results show that WIGNN significantly outperforms other models in handling class imbalance and graph sparsity, demonstrating its potential in financial credit applications.}, archive = {J_EAAI}, author = {Zhipeng Yan and Hanwen Qu and Chen Chen and Xiaoyi Lv and Enguang Zuo and Kui Wang and Xulun Cai}, doi = {10.1016/j.engappai.2024.109597}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109597}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {WIGNN: An adaptive graph-structured reasoning model for credit default prediction}, volume = {139}, year = {2025}, }
• (2025). Weather-aware energy management for unmannedaerial vehicles: A machine learning application with global data integration. EAAI, 139, 109596. (www)

  @article{ , abstract = {This study introduces a machine learning (ML) framework to predict unmanned aerial vehicle (UAV) energy requirements under diverse environmental conditions. The framework correlates UAV flight patterns with publicly accessible weather data, to yield an energy management tool applicable to a wide range of UAV configurations. The model employs the Cross-industry standard process for data mining and advanced feature engineering, offering an in-depth analysis of meteorological factors and UAV energy demands. The study assesses several multi-regression linear and ML models, whereby ensemble models gradient boosting (GB) and eXtreme gradient boosting demonstrate superior performance and accuracy. Specifically, the GB model achieved a test mean absolute error (MAE) of 0.0395 V (V) for voltage, 0.808 A (A) for current, and 9.758 mA-hours (mAh) for discharge, with prediction accuracy of over 99.9% for voltage and discharge, and 97% for current, derived from the coefficient of determination (R 2 ). A novel integration of real-world UAV logs and weather data underpins the development of a weather-aware ML prediction model for UAV energy consumption. Our framework is capable of concurrently predicting three components of energy and power with almost uniform accuracy, a feature not found in contemporary models. Empirical test flights show a discrepancy of only 0.005 W-hour (Wh) between total predicted and actual energy consumption. This work enhances both efficiency and safety in UAV operations. The resulting energy-predictive flight planning tool sets a new benchmark for artificial intelligence (AI) applications in intelligent automation for UAVs.}, archive = {J_EAAI}, author = {Abhishek G. Somanagoudar and Walter Mérida}, doi = {10.1016/j.engappai.2024.109596}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109596}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weather-aware energy management for unmannedaerial vehicles: A machine learning application with global data integration}, volume = {139}, year = {2025}, }
• (2025). Cooperative task assignment of heterogeneous unmanned aerial vehicles for simultaneous multi-directional attack on a moving target. EAAI, 139, 109595. (www)

  @article{ , abstract = {Multiple unmanned aerial vehicle (UAV) attacks, on moving targets with directional priorities, need careful task allocation especially when the UAVs have variable attacking power. In addition, the constraints of simultaneous attack from different directions multi-folds the complexity of the problem. In this work, an extended contract net protocol (ECNP) based autonomous and cooperative task assignment method is proposed to deal with the time-sensitive position assignment for multi-directional attack and uniform resource allocation. Initially, an optimization problem is formulated using distances between each UAV and expected attack positions (APs), arrival time, and attack power. In addition, for uniform resource allocation, a variable is introduced to monitor available resources at a given time. An agent-based model is built with UAV information, such as position and speed, along with the direction of the high-value moving target (HVMT). Each UAV identifies the possible arrival points based on its speed constraints, and current position, and the position and velocity of HVMT. Moreover, distance and expected arrival time to all APs are computed. Finally, the agents make attack point allocations using the proposed ECNP after making a consensus about simultaneous attack time. The proposed method ensures uniform resource allocation. The simulation results show the superiority of the proposed method (ECNP) in comparison with classical contract net protocol (CNP) and Genetic Algorithms (GA) in terms of uniform resource allocation and mission accomplishment.}, archive = {J_EAAI}, author = {Sami Shahid and Ziyang Zhen and Umair Javaid}, doi = {10.1016/j.engappai.2024.109595}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109595}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cooperative task assignment of heterogeneous unmanned aerial vehicles for simultaneous multi-directional attack on a moving target}, volume = {139}, year = {2025}, }
• (2025). Three-branch neural network for no-reference quality assessment of pan-sharpened images. EAAI, 139, 109594. (www)

  @article{ , abstract = {Pan-Sharpening (PS) techniques aim to enhance the spatial resolution of low-resolution multispectral images by leveraging data from high-resolution panchromatic images. Their comparison typically relies on the quality assessment of the resulting Full-Resolution (FS) pan-sharpened images. However, in the absence of a reference image, a dedicated No-Reference (NR) method must be employed. Therefore, this paper introduces a novel approach called the Three-Branch Neural Network for No-Reference Quality Assessment of Pan-Sharpened Images (TBN-PSI). The network consists of three subnetworks designed for perceptual processing of image channels, featuring shared extraction of low-level features and high-level semantics. Extensive experimental evaluation demonstrates the superiority of the approach over the state-of-the-art NR PS image quality assessment methods, using six datasets containing diverse satellite images that span urban areas, green vegetation, and water scenarios. Specifically, TBN-PSI outperforms the compared methods by 4% to 9% in terms of Spearman’s Rank-Order Correlation Coefficient (SRCC), Pearson’s Linear Correlation Coefficient (PLCC), and Kendall’s Rank Correlation Coefficient (KRCC) between the obtained scores and those of three representative full-reference methods.}, archive = {J_EAAI}, author = {Igor Stępień and Mariusz Oszust}, doi = {10.1016/j.engappai.2024.109594}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109594}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Three-branch neural network for no-reference quality assessment of pan-sharpened images}, volume = {139}, year = {2025}, }
• (2025). A flow rate estimation method for gas–liquid two-phase flow based on filter-enhanced convolutional neural network. EAAI, 139, 109593. (www)

  @article{ , abstract = {Accurate estimation of flow rate in gas–liquid two-phase flow is crucial for various industrial processes. How to accurately estimate flow rate remains a challenging problem. Previously, deep learning-based methods focused on a few human-set points with single task learning. In addition, the data were not denoised. In this study, a flow rate estimation method based on a filter-enhanced convolutional neural network (FECNN) is proposed for gas–liquid two-phase flow. The method leverages multimodal data from a Venturi tube and an electrical capacitance tomography (ECT) sensor as input, utilizing multilayer perceptron (MLP) to fuse data. Subsequently, a learnable filter module is employed to attenuate noise adaptively, followed by multiscale convolutional neural network (MSCNN) extraction of flow rate features at different scales. Finally, the method enables estimate each single-phase flow rate simultaneously through multi-task learning (MTL). The adaptive noise attenuation capabilities of the learnable filter module are demonstrated, and the ability of the proposed MSCNN to capture multiscale flow rate features through multiple comparative experiments is shown. Additionally, a qualitative comparison with recent flow rate estimation methods is provided. Overall, this study demonstrates the effectiveness and superiority of the proposed FECNN in flow rate estimation.}, archive = {J_EAAI}, author = {Yuxiao Jiang and Yinyan Liu and Lihui Peng and Yi Li}, doi = {10.1016/j.engappai.2024.109593}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109593}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A flow rate estimation method for gas–liquid two-phase flow based on filter-enhanced convolutional neural network}, volume = {139}, year = {2025}, }
• (2025). A distribution linguistic group decision-making method considering twin multiplicative data envelopment analysis regret-rejoice cross-efficiency. EAAI, 139, 109592. (www)

  @article{ , abstract = {Traditional approaches to group decision-making (GDM) often assume that decision-makers (DMs) are perfectly rational, which neglects the psychological attitudes exhibited by DMs during the decision-making process. To address this issue, this paper designs a novel method for GDM with multiplicative distribution linguistic preference relations (MDLPRs), which incorporates twin multiplicative data envelopment analysis (TMDEA) regret-rejoice cross-efficiency models and a weighting model with individual tolerance. Firstly, we provide a clear definition of MDLPRs based on multiplicative linguistic terms that capture the asymmetrical qualitative recognition of DMs. Then, a weighting model is constructed to obtain DM's weights. This model assumes that DMs are boundedly rational and have a certain tolerance level for group consensus measurement. Subsequently, TMDEA cross-efficiency models are developed based on different perspectives. On this basis, we further propose TMDEA regret-rejoice cross-efficiency models that consider the psychological risks of DMs. Furthermore, the method for GDM based on TMDEA regret-rejoice cross-efficiency models and individual tolerance with MDLPRs is given. Finally, we present a case study of the selection of the Ya'an ecological monitoring station in China to test the validity and applicability of the proposed method. The merits and robustness of the constructed method are highlighted by sensitive analysis and comparative analysis.}, archive = {J_EAAI}, author = {Jinpei Liu and Tianqi Shui and Longlong Shao and Feifei Jin and Ligang Zhou}, doi = {10.1016/j.engappai.2024.109592}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109592}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A distribution linguistic group decision-making method considering twin multiplicative data envelopment analysis regret-rejoice cross-efficiency}, volume = {139}, year = {2025}, }
• (2025). Cross-modal prompt-driven network for low-resource vision-to-language generation. EAAI, 139, 109591. (www)

  @article{ , abstract = {Image captioning is a classic vision-to-language generation task, which aims to generate a descriptive sentence to describe the input image, involving the understanding of the image and the generation of natural language. Conventional methods require a large-scale labeled dataset for training, which includes a large volume of image-caption pairs. However, for several application scenarios, e.g., medicine and non-English, such plenty of image-caption pairs are usually not available. In this work, we propose the Cross-modal Prompt-Driven Network (XProDNet) to perform low-resource image captioning, which can generate accurate and comprehensive image captioning, with extremely limited data for training. We conduct experiments on (1) six benchmark datasets; (2) three application scenarios, i.e. , conventional image captioning, medical image captioning, and non-English image captioning; (3) four target languages, i.e. , English, Chinese, German, and French; (4) two experimental settings, i.e. , fully-supervised learning and few-shot learning. The extensive experiments prove the effectiveness of our approach, which can not only generate high-quality and comprehensive image captions but also significantly surpass previous state-of-the-art methods under both the few-shot learning and fully-supervised learning settings. The improved results suggest that our method has great potential for improving image captioning in real-world applications.}, archive = {J_EAAI}, author = {Yuena Jiang and Yanxun Chang}, doi = {10.1016/j.engappai.2024.109591}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109591}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-modal prompt-driven network for low-resource vision-to-language generation}, volume = {139}, year = {2025}, }
• (2025). Etching process prediction based on cascade recurrent neural network. EAAI, 139, 109590. (www)

  @article{ , abstract = {Etching is one of the most critical processes in semiconductor manufacturing. Etch models have been developed to reveal the underlying etch mechanisms, which employs rigorous physical and chemical process simulation. Traditional simulation is very time consuming. The data-driven artificial intelligence model provides an alternative modeling approach. In this paper, a Cascade Recurrent Neural Networks (CRNN) is proposed to model and predict etching profiles. The etching profile is represented by polar coordinates and modeled by the recurrent neural networks, the corresponding etching parameters (e.g., pressure, power, temperature, and voltage) are integrated into the network through cascade combination layers. Experimental results on a dataset of 10,000 simulated etching profiles demonstrated the effectiveness of our method: compared with traditional etching simulation methods, CRNN can speedup 21,000 × with an average error of less than 0.7 nm for 1 step prediction. Furthermore, compared to simple deep neural networks, the Mean Absolute Errors (MAE) could be reduced from 1.7329 nm to 1.3845 nm for 10 steps prediction. Finally, the effectiveness and accuracy of CRNN etching predictor is validated through fine-tuning on experimental data.}, archive = {J_EAAI}, author = {Zhenjie Yao and Ziyi Hu and Panpan Lai and Fengling Qin and Wenrui Wang and Zhicheng Wu and Lingfei Wang and Hua Shao and Yongfu Li and Zhiqiang Li and Zhongming Liu and Junjie Li and Rui Chen and Ling Li}, doi = {10.1016/j.engappai.2024.109590}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109590}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Etching process prediction based on cascade recurrent neural network}, volume = {139}, year = {2025}, }
• (2025). Soft prompt-tuning with self-resource verbalizer for short text streams. EAAI, 139, 109589. (www)

  @article{ , abstract = {Short text streams such as real-time news and search snippets have attained vast amounts of attention and research in recent decades, the characteristics of high generation velocity, feature sparsity, and high ambiguity accentuate both the importance and challenges to language models. However, most of the existing short text stream classification methods can neither automatically select relevant knowledge components for arbitrary samples, nor expand knowledge internally instead of rely on external open knowledge base to address the inherent limitations of short text stream. In this paper, we propose a Soft Prompt-tuning with Self-Resource Verbalizer (SPSV for short) for short text stream classification, the soft prompt with self-resource knowledgeable expansion is conducted for updating label words space to address evolved semantic topics in the data streams. Specifically, the automatic constructed prompt is first generated to instruct the model prediction, which is optimized to address the problem of high velocity and topic drift in short text streams. Then, in each chunk, the projection between category names and label words space, i.e. verbalizer, is updated, which is constructed by internal knowledge expansion from the short text itself. Through comprehensive experiments on four well-known benchmark datasets, we validate the superb performance of our method compared to other short text stream classification and fine-tuning PLMs methods, which achieves up to more than 90% classification accuracy with the counts of data chunk increased.}, archive = {J_EAAI}, author = {Yi Zhu and Ye Wang and Yun Li and Jipeng Qiang and Yunhao Yuan}, doi = {10.1016/j.engappai.2024.109589}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109589}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Soft prompt-tuning with self-resource verbalizer for short text streams}, volume = {139}, year = {2025}, }
• (2025). A knowledge-guided reinforcement learning method for lateral path tracking. EAAI, 139, 109588. (www)

  @article{ , abstract = {Lateral Control algorithms in autonomous vehicles often necessitates an online fine-tuning procedure in the real world. While reinforcement learning (RL) enables vehicles to learn and improve the lateral control performance through repeated trial and error interactions with a dynamic environment, applying RL directly to safety-critical applications in real physical world is challenging because ensuring safety during the learning process remains difficult. To enable safe learning, a promising direction is to make use of previously gathered offline data, which is frequently accessible in engineering applications. In this context, this paper presents a set of knowledge-guided RL algorithms that can not only fully leverage the prior collected offline data without the need of a physics-based simulator, but also allow further online policy improvement in a smooth, safe and efficient manner. To evaluate the effectiveness of the proposed algorithms on a real controller, a hardware-in-the-loop and a miniature vehicle platform are built. Compared with the vanilla RL, behavior cloning and the existing controller, the proposed algorithms realize a closed-loop solution for lateral control problems from offline training to online fine-tuning, making it attractive for future similar RL-based controller to build upon.}, archive = {J_EAAI}, author = {Bo Hu and Sunan Zhang and Yuxiang Feng and Bingbing Li and Hao Sun and Mingyang Chen and Weichao Zhuang and Yi Zhang}, doi = {10.1016/j.engappai.2024.109588}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109588}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A knowledge-guided reinforcement learning method for lateral path tracking}, volume = {139}, year = {2025}, }
• (2025). Cognitive digital twins of the natural environment: Framework and application. EAAI, 139, 109587. (www)

  @article{ , abstract = {Digital Twin (DT) technology offers a method of creating digital models of natural systems to enhance their ability to withstand natural disasters. Currently, DT of the natural environment is in its initial phases, lacking adaptive capabilities and relying on human-assisted modeling. The key to endowing DT of the natural environment with greater autonomy lies in the integration of expert knowledge. Knowledge graphs can efficiently arrange and structurally store expert knowledge, thereby supporting the autonomous functionality of DT. This paper introduces the concept of Cognitive Digital Twin(CDT) derived from the industrial domain and presents a framework for CDT of the natural environment. This framework is centered around knowledge graph technology, aiming to provide more insights and guidance for system development. This framework integrates human cognition by constructing knowledge graphs of objects, models, events, and scene modes. Moreover, these knowledge graphs support agents for the dynamic adjustment of processes, as well as the adaptation and parameter optimization of related models. As a use case, we utilize this framework to implement digital twin watersheds. We develop appropriate ontologies and agents to facilitate the construction of cognitive digital watersheds for various regions. Cognitive digital watersheds effectively fulfill the application needs of integrated flood forecasting and control scheduling. This application validates the framework’s effectiveness and provides a reference for constructing CDTs of other natural systems.}, archive = {J_EAAI}, author = {Jun Feng and Hailin Tang and Siyuan Zhou and Yang Cai and Jianxin Zhang}, doi = {10.1016/j.engappai.2024.109587}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109587}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cognitive digital twins of the natural environment: Framework and application}, volume = {139}, year = {2025}, }
• (2025). On the use of synthetic data for body detection in maritime search and rescue operations. EAAI, 139, 109586. (www)

  @article{ , abstract = {Time is a critical factor in maritime Search And Rescue (SAR) missions, during which promptly locating survivors is paramount. Unmanned Aerial Vehicles (UAVs) are a useful tool with which to increase the success rate by rapidly identifying targets. While this task can be performed using other means, such as helicopters, the cost-effectiveness of UAVs makes them an effective choice. Moreover, these vehicles allow the easy integration of automatic systems that can be used to assist in the search process. Despite the impact of artificial intelligence on autonomous technology, there are still two major drawbacks to overcome: the need for sufficient training data to cover the wide variability of scenes that a UAV may encounter and the strong dependence of the generated models on the specific characteristics of the training samples. In this work, we address these challenges by proposing a novel approach that leverages computer-generated synthetic data alongside novel modifications to the You Only Look Once (YOLO) architecture that enhance its robustness, adaptability to new environments, and accuracy in detecting small targets. Our method introduces a new patch-sample extraction technique and task-specific data augmentation, ensuring robust performance across diverse weather conditions. The results demonstrate our proposal’s superiority, showing an average 28% relative improvement in mean Average Precision (mAP) over the best-performing state-of-the-art baseline under training conditions with sufficient real data, and a remarkable 218% improvement when real data is limited. The proposal also presents a favorable balance between efficiency, effectiveness, and resource requirements.}, archive = {J_EAAI}, author = {Juan P. Martinez-Esteso and Francisco J. Castellanos and Adrian Rosello and Jorge Calvo-Zaragoza and Antonio Javier Gallego}, doi = {10.1016/j.engappai.2024.109586}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109586}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {On the use of synthetic data for body detection in maritime search and rescue operations}, volume = {139}, year = {2025}, }
• (2025). Demand management of plug-in electric vehicle charging station considering bidirectional power flow using deep reinforcement learning. EAAI, 139, 109585. (www)

  @article{ , abstract = {The recent development in plug-in electric vehicle technology has increased its popularity. Plug-in electric vehicles are widely used for their environment-friendly nature and they contribute to the reduction in global warming. With the increasing number of plug-in electric vehicles, charging coordination becomes essential for managing the charging station demand. The random charging behaviour of plug-in electric vehicles makes it a difficult task. This paper proposes a novel demand management strategy of charging stations. The proposed strategy supports the grid and reduces its burden during peak load. Deep-Q network based deep reinforcement learning is used in the proposed strategy. It is a value based deep reinforcement learning algorithm that approximates the Q value function using deep neural network. The deep reinforcement schedules the charging and discharging of plug-in electric vehicles to optimize the cost and manage the charging station load. Deep reinforcement learning enhances charging coordination by dynamically optimizing charging schedules according to real-time conditions and user preferences, thereby increasing efficiency and better integration with the grid. The reward function in deep reinforcement learning is designed based on the power price and demand of the charging station. A discount factor is introduced based on the service time to make the charging coordination efficient. A case study using dynamic pricing is carried out to validate the proposed strategy. The results prove that the proposed strategy optimizes charging and discharging costs and manages charging station demand efficiently. It is also observed that the proposed strategy is fast and incurs less computational cost.}, archive = {J_EAAI}, author = {Durgesh Choudhary and Rabindra Nath Mahanty and Niranjan Kumar}, doi = {10.1016/j.engappai.2024.109585}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109585}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Demand management of plug-in electric vehicle charging station considering bidirectional power flow using deep reinforcement learning}, volume = {139}, year = {2025}, }
• (2025). A fault diagnosis framework using unlabeled data based on automatic clustering with meta-learning. EAAI, 139, 109584. (www)

  @article{ , abstract = {With the growth of the industrial internet of things, the poor performance of conventional deep learning models hinders the application of intelligent diagnosis methods in industrial situations such as lack of fault samples and difficulties in data labeling. To solve the above problems, we propose a fault diagnosis framework based on unsupervised meta-learning and contrastive learning, which is called automatic clustering with meta-learning (ACML). First, the amount of data is expanded through data augmentation approaches, and a feature generator is constructed to extract highly discriminative features from the unlabeled dataset using contrastive learning. Then, a cluster generator is used to automatically divide cluster partitions and add pseudo-labels for these. Finally, the classification tasks are derived through taking original samples in the partitions, which are embedded in the meta-learner for fault diagnosis. In the meta-learning stage, we split out two subsets from task and feed them into the inner and outer loops to maintain the class consistency of the real labels. After training, ACML transfers its prior expertise to the unseen task to efficiently complete the categorization of new faults. ACML is applied to two cases concerning a public dataset and a self-constructed dataset, demonstrate that ACML achieves good diagnostic performance, outperforming popular unsupervised methods.}, archive = {J_EAAI}, author = {Zhiqian Zhao and Yinghou Jiao and Yeyin Xu and Zhaobo Chen and Enrico Zio}, doi = {10.1016/j.engappai.2024.109584}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109584}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A fault diagnosis framework using unlabeled data based on automatic clustering with meta-learning}, volume = {139}, year = {2025}, }
• (2025). Cross-attention interaction learning network for multi-model image fusion via transformer. EAAI, 139, 109583. (www)

  @article{ , abstract = {Current image fusion techniques often fail to adequately consider the inherent correlations among different modalities, resulting in suboptimal integration of multi-modal information. Drawing inspiration from inter-modal interactions, this paper introduces a cross-attention interaction learning network, CrossATF, leveraging the transformer architecture. The cornerstone of CrossATF resides in a generator network equipped with dual encoders. The multi-modal encoder incorporates two transformer modules of comparable computational complexity, alongside a meticulously designed cross-modal transformer. This architectural choice empowers the model to effectively extract modality-specific features while simultaneously integrating complementary features from diverse modalities. Furthermore, an auxiliary encoder is enlisted to encode features from the entire input image, thereby enhancing the model's comprehensive understanding of the image. Significantly, the loss function is tailored to selectively preserve a more targeted set of information from the source images, endowing the network with heightened feature extraction capabilities. Comprehensive experimental results across various datasets substantiate the promising performance of the proposed approach when compared to both task-specific methodologies and unified fusion frameworks.}, archive = {J_EAAI}, author = {Jing Wang and Long Yu and Shengwei Tian}, doi = {10.1016/j.engappai.2024.109583}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109583}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Cross-attention interaction learning network for multi-model image fusion via transformer}, volume = {139}, year = {2025}, }
• (2025). SinkFlow: Fast and traceable root-cause localization for multidimensional anomaly events. EAAI, 139, 109582. (www)

  @article{ , abstract = {With the development of various artificial intelligence (AI)–based applications, detecting anomalies and analyzing the root causes from massive data are critical to increasing the usability of AI. Fast, accurate root-cause analysis (RCA) that finds the main reason for an anomaly, as well as reasonable explanations, helps in solving problems effectively. Thus, RCA plays an important role in troubleshooting and fault diagnosis, making its application in data analysis crucial. Previous root-cause-localization approaches for multidimensional anomaly events encompass various techniques to reduce search space and have improved the localization performance. However, they do not effectively balance the requirements in terms of performance, compatibility, and interpretability. To solve these problems, we propose a new root-cause-localization method called SinkFlow . It provides a unified framework event-aggregation Graph (EAG) to describe the constraints of event aggregation and relations between events, so it can be easily generalized to various domains. SinkFlow introduces an applicable measure evaluation method for both fundamental and derived measures to quantify the impact of events. Also, it utilizes an optimal search strategy to reduce the search space based on the anomaly behavioral consistency and deviation significance. Our experimental results on semisynthetic datasets show that SinkFlow achieved better performance than other baselines and ran much faster, achieving a 1.88% increase of the F1-score and only 25% of the time cost of the second best localization method. In addition, SinkFlow offered clear, visible explanations of the localization results to answer the questions of why they are root causes and how the anomaly is formed.}, archive = {J_EAAI}, author = {Zhichao Hu and Likun Liu and Lina Ma and Xiangzhan Yu}, doi = {10.1016/j.engappai.2024.109582}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109582}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {SinkFlow: Fast and traceable root-cause localization for multidimensional anomaly events}, volume = {139}, year = {2025}, }
• (2025). Lightweight distributed deep learning on compressive measurements for internet of things. EAAI, 139, 109581. (www)

  @article{ , abstract = {In this work, we investigate the problem of distributed deep learning in Internet of Things (IoT). The proposed learning framework is constructed in a fog–cloud computing architecture, so as to overcome the limitation of resource constrained IoT end device. Compressive Sensing (CS) is used as a lightweight encryption in the framework to preserve the privacy of training data. Specifically, a chaotic-based CS measurement matrix construction mechanism is applied in the system to save the storage and transmission costs. With this design, the computation overhead of the learning framework in IoT can be successfully offloaded from IoT end device to the fog nodes. Theoretical analysis demonstrates that our system can guarantee security of the raw data against chosen plaintext attack (CPA). Experimental and analysis results show that our privacy-preserving proposal can significantly reduce the communication costs and computation costs with only a negligible accuracy penalty (with classification accuracy 91% testing on MNIST dataset under compression rate 0.5) compared to traditional non-private federated learning schemes. Notably, due to the chaotic-based CS measurement matrix construction mechanism, the memory requirement of end device side can be significantly reduced. This makes our framework be very suitable for the IoT applications in which end devices are equipped with low-spec chips.}, archive = {J_EAAI}, author = {Guiqiang Hu and Yong Hu and Tao Wu and Yushu Zhang and Shuai Yuan}, doi = {10.1016/j.engappai.2024.109581}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109581}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Lightweight distributed deep learning on compressive measurements for internet of things}, volume = {139}, year = {2025}, }
• (2025). Federated reinforcement learning for smart and privacy-preserving energy management of residential microgrids clusters. EAAI, 139, 109579. (www)

  @article{ , abstract = {Real-time energy management optimizes energy utilization and manages electrical loads, which is crucial for improving the operational efficiency of residential microgrids. However, existing management methods suffer from model complexity and slow training speed. To solve this problem, we introduce Federated Reinforcement Learning to manage residential microgrids by training a control strategy in a decentralized and privacy-preserving manner. Specifically, a residential microgrid energy optimization management model is first established based on the Proximal Policy Optimization (PPO) method. Then, we propose a cooperative training strategy for multiple Residential microgrids based on Federated Reinforcement Learning (RFRL). The proposed method improves the training speed of residential microgrid models by sharing parameter information, such as network weights, while protects users’ usage data. Finally, clustering analysis is introduced in the case of heterogeneous residential microgrid data. Extensive experimental evaluation shows that our method outperforms the alternative residential microgrid management methods in terms of cost efficiency.}, archive = {J_EAAI}, author = {Mao Tan and Jie Zhao and Xiao Liu and Yongxin Su and Ling Wang and Rui Wang and Zhuocen Dai}, doi = {10.1016/j.engappai.2024.109579}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109579}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Federated reinforcement learning for smart and privacy-preserving energy management of residential microgrids clusters}, volume = {139}, year = {2025}, }
• (2025). Anomaly detection in smart-manufacturing era: A review. EAAI, 139, 109578. (www)

  @article{ , abstract = {Manufacturing downtime due to faults is costly and disruptive. With the increasing availability of real-time data in modern Smart Manufacturing (SM) environments, effective anomaly detection (AD) has become crucial but challenging due to diverse scenarios and methods. This paper aims to present a comprehensive review of state-of-the-art AD methods tailored for SM, to facilitate their implementation in real manufacturing environments while providing a foundation for future research. First, it introduces a structured SM classification framework highlighting recent and successful AD algorithms applied in real-world scenarios with a valuable repository of over 100 manufacturing datasets to support further research. Second, an extensive experimental evaluation of 16 AD algorithms across 29 SM datasets covering a broad and diverse spectrum of techniques, including supervised, unsupervised, and semi-supervised approaches, encompassing classic, deep, and ensemble methods. Finally, insights gained from these experiments are presented providing practical guidance on the most suitable methods for various manufacturing contexts, identifying key challenges and opportunities for future developments.}, archive = {J_EAAI}, author = {Iñaki Elía and Miguel Pagola}, doi = {10.1016/j.engappai.2024.109578}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109578}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Anomaly detection in smart-manufacturing era: A review}, volume = {139}, year = {2025}, }
• (2025). Fault diagnosis in electric machines and propellers for electrical propulsion aircraft: A review. EAAI, 139, 109577. (www)

  @article{ , abstract = {The present work aims to conduct an extensive literature review on the fault diagnosis and classification in electric machines, especially those with permanent magnets, for aeronautical propulsion applications. The main contribution of this research is to assess how intelligent systems focused on fault detection and diagnosis in electric propulsion systems have evolved over the past five years, what are the main types of algorithms used, and how the rapid advancement of machine learning techniques has impacted this research area. Initially, an introduction to the main diagnostic methods is provided, including techniques based on mathematical models, signal analysis, as well as the use of machine learning and deep learning. Subsequently, a detailed study of the main references found in recent years for each type of fault, whether electrical, magnetic, or mechanical, is undertaken. Regarding aeronautical applications, a study of faults in rotating blades and on coupling systems between an electric motor and a set of propellers is conducted. Throughout the text, some of the main datasets found during the research are presented. These datasets include characteristics of healthy operation and fault of windings, bearings, as well as other mechanical components that can be connected to the machine’s shaft, such as gearboxes. Finally, some statistics from this research are presented showing results regarding the annual distribution of publication of all reviewed references, the proportion of faults addressed in all articles, as well as a detailed analysis of the proportion in which each type of algorithm appears in the cited references.}, archive = {J_EAAI}, author = {Leonardo Duarte Milfont and Gabriela Torllone de Carvalho Ferreira and Mateus Giesbrecht}, doi = {10.1016/j.engappai.2024.109577}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109577}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis in electric machines and propellers for electrical propulsion aircraft: A review}, volume = {139}, year = {2025}, }
• (2025). Machine learning based state observer for discrete time systems evolving on lie groups. EAAI, 139, 109576. (www)

  @article{ , abstract = {In this paper, a machine learning based observer for systems evolving on manifolds is designed such that the state of the observer is restricted to the Lie group on which the system evolves. Designing machine learning based observers for systems evolving on Lie groups using charts would require training a machine learning based observer for each chart of the Lie group, and switching between the trained models based on the state of the system. We propose a novel deep learning based technique whose predictions are restricted to certain measure 0 subsets of the Euclidean space without using charts. Using this network, we design an observer ensuring that the state of the observer is restricted to the Lie group, and predicting the state using only one trained algorithm. The deep learning network predicts an error term on the Lie algebra of the Lie group, uses the map from the Lie algebra to the group, the group operation, and the present state to estimate the state at the next epoch. This approach, being purely data driven, does not require a model of the system. The proposed algorithm provides a novel framework for constraining the output of machine learning networks to certain measure 0 subsets of a Euclidean space without training on each specific chart and without requiring switching. We show the validity of this method using Monte Carlo simulations performed of the rigid body rotation and translation system.}, archive = {J_EAAI}, author = {Soham Shanbhag and Dong Eui Chang}, doi = {10.1016/j.engappai.2024.109576}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109576}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Machine learning based state observer for discrete time systems evolving on lie groups}, volume = {139}, year = {2025}, }
• (2025). Spatio-temporal attention based collaborative local–global learning for traffic flow prediction. EAAI, 139, 109575. (www)

  @article{ , abstract = {Traffic flow prediction is crucial for intelligent transportation systems (ITS), providing valuable insights for traffic control, route planning, and operation management. Existing work often separately models the spatial and temporal dependencies and primarily relies on predefined graphs to represent spatio-temporal dependencies, neglecting the traffic dynamics caused by unexpected events and the global relationships among road segments. Unlike previous models that primarily focus on local feature extraction, we propose a novel collaborative local–global learning model (LOGO) that employs spatio-temporal attention (STA) and graph convolutional networks (GCN). Specifically, LOGO simultaneously extracts hidden traffic features from both local and global perspectives. In local feature extraction, a novel STA is devised to directly attend to spatio-temporal coupling interdependencies instead of separately modeling temporal and spatial dependencies, and to capture in-depth spatio-temporal traffic context with an adaptive graph focusing on the dynamics in traffic flow. In global feature extraction, a global correlation matrix is constructed and GCNs are utilized to propagate messages on the obtained matrix to achieve interactions between both adjacent and similar road segments. Finally, the obtained local and global features are concatenated and fed into a gated aggregation to forecast future traffic flow. Extensive experiments on four real-world traffic datasets sourced from the Caltrans Performance Measurement System (PEMS03, PEMS04, PEMS07, and PEMS08) demonstrate the effectiveness of our proposed model. LOGO achieves the best performance over 18 state-of-the-art baselines and the best prediction performance with the highest improvement of 6.06% on the PEMS07 dataset. Additionally, two real-world case studies further substantiate the robustness and interpretability of LOGO.}, archive = {J_EAAI}, author = {Haiyang Chi and Yuhuan Lu and Can Xie and Wei Ke and Bidong Chen}, doi = {10.1016/j.engappai.2024.109575}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109575}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Spatio-temporal attention based collaborative local–global learning for traffic flow prediction}, volume = {139}, year = {2025}, }
• (2025). Designing deep neural networks for driver intention recognition. EAAI, 139, 109574. (www)

  @article{ , abstract = {Driver intention recognition (DIR) studies increasingly rely on deep neural networks. Deep neural networks have achieved top performance for many different tasks. However, apart from image classifications and semantic segmentation for mobile phones, it is not a common practice for components of advanced driver assistance systems to explicitly analyze the complexity and performance of the network’s architecture. Therefore, this paper applies neural architecture search to investigate the effects of the deep neural network architecture on a real-world safety critical application with limited computational capabilities. We explore a pre-defined search space for three deep neural network layer types that are capable to handle sequential data (a long-short term memory, temporal convolution, and a time-series transformer layer), and the influence of different data fusion strategies on the driver intention recognition performance. A set of eight search strategies are evaluated for two driver intention recognition datasets. For the two datasets, we observed that there is no search strategy clearly sampling better deep neural network architectures. However, performing an architecture search improves the model performance compared to the original manually designed networks. Furthermore, we observe no relation between increased model complexity and better driver intention recognition performance. The result indicate that multiple architectures can yield similar performance, regardless of the deep neural network layer type or fusion strategy. However, the optimal complexity, layer type and fusion remain unknown upfront.}, archive = {J_EAAI}, author = {Koen Vellenga and H. Joe Steinhauer and Alexander Karlsson and Göran Falkman and Asli Rhodin and Ashok Koppisetty}, doi = {10.1016/j.engappai.2024.109574}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109574}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Designing deep neural networks for driver intention recognition}, volume = {139}, year = {2025}, }
• (2025). An end-to-end deep convolutional neural network-based data-driven fusion framework for identification of human induced pluripotent stem cell-derived endothelial cells in photomicrographs. EAAI, 139, 109573. (www)

  @article{ , abstract = {Deep learning is a very powerful analytic tool to recognize the patterns in data to make appropriate predictions. It has tremendous potential in data analyses, particularly for cell biology domain, caused by the growing scale and inherent complexity of biological data. The core purpose of this research work is to design, implement, and calibrate an efficient deep convolutional neural network (DCNN) architecture in the context of binary-class classification problem. This diversified network is developed to precisely identify human induced pluripotent stem cell-derived endothelial cells (hiPSC-derived EC) based on photomicrograph. The proposed architecture is cerebrally developed with numerous convolutional modules, multiple kernel sizes, various pooling layers, activation functions and strides, nevertheless fewer trainable parameters to strengthen the network and enhance its performance. The proposed feature fusion framework is compared with the classifier fusion approach in terms of Matthews’s correlation coefficient (MCC), training time, inference time, number of layers, number of parameters, graphics processing unit (GPU) memory utilization, and floating-point operations (FLOPS). Specifically, it achieves 94.6% sensitivity, 94.5% specificity, and 94.7% precision. Induced pluripotent stem cell (iPS) dataset is also introduced in this research work that has 16278 images which are labelled by three independent and experienced human experts of cell biology domain to facilitate future research. Experimental results show that the proposed framework offers an innovative and attainable algorithm for accelerating and systematizing the classification task along with saving time and effort.}, archive = {J_EAAI}, author = {Imran Iqbal and Imran Ullah and Tingying Peng and Weiwei Wang and Nan Ma}, doi = {10.1016/j.engappai.2024.109573}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109573}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An end-to-end deep convolutional neural network-based data-driven fusion framework for identification of human induced pluripotent stem cell-derived endothelial cells in photomicrographs}, volume = {139}, year = {2025}, }
• (2025). Korean football in-game conversation state tracking dataset for dialogue and turn level evaluation. EAAI, 139, 109572. (www)

  @article{ , abstract = {Recent research in dialogue state tracking has made significant progress in tracking user goals through dialogue-level and turn-level approaches, but existing research primarily focused on predicting dialogue-level belief states. In this study, we present the KICK : K orean football I n-game C onversation state trac K ing dataset, which introduces a conversation-based approach. This approach leverages the roles of casters and commentators within the self-contained context of sports broadcasting to examine how utterances impact the belief state at both the dialogue-level and turn-level. Towards this end, we propose a task that aims to track the states of a specific time turn and understand conversations during the entire game. The proposed dataset comprises 228 games and 2463 events over one season, with a larger number of tokens per dialogue and turn, making it more challenging than existing datasets. Experiments revealed that the roles and interactions of casters and commentators are important for improving the zero-shot state tracking performance. By better understanding role-based utterances, we identify distinct approaches to the overall game process and events at specific turns.}, archive = {J_EAAI}, author = {Sangmin Song and Juhyoung Park and Juhwan Choi and Junho Lee and Kyohoon Jin and YoungBin Kim}, doi = {10.1016/j.engappai.2024.109572}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109572}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Korean football in-game conversation state tracking dataset for dialogue and turn level evaluation}, volume = {139}, year = {2025}, }
• (2025). Explained fire resistance machine learning models for compressed steel members of trusses and bracing systems. EAAI, 139, 109571. (www)

  @article{ , abstract = {Trusses and bracing systems are usually constructed from monosymmetric and built-up cross-sections, which under compression stresses may buckle in torsional or flexural–torsional modes. In fire, this phenomenon is utterly important as failure in bracing systems or trusses may cause the collapse of buildings and result in loss of lives or severe economic impacts. Machine learning models, including neural networks, random forests and support vector machines, are developed considering a dataset with 21879 samples and are further assessed in this study as an alternative with greater accuracy and ease of application over existing design methods, namely the Eurocode 3 Part 1-2 and a recent proposal for its improvement. The machine learning models are explained using a combination of domain knowledge inference, partial dependence plots and SHapley Additive exPlanations. The accuracy versus safety trade-off is discussed for a better-informed model selection. The proposed approach and discussion create an additional confidence layer for applying these techniques for the fire design.}, archive = {J_EAAI}, author = {Luca Possidente and Carlos Couto}, doi = {10.1016/j.engappai.2024.109571}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109571}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explained fire resistance machine learning models for compressed steel members of trusses and bracing systems}, volume = {139}, year = {2025}, }
• (2025). Multi-scale motion-based relational reasoning for group activity recognition. EAAI, 139, 109570. (www)

  @article{ , abstract = {Existing group activity recognition methods generally use optical flow image to represent motion within videos, which often struggle to capture the movements of individuals inaccurately. In this paper, we explore the effectiveness of more kinds of motion information for group activity recognition. We propose a novel multi-scale MOtion-based relational reasoning framework for Group Activity Recognition (MOGAR). It combines joint motion (intra-individual level) with trajectory (individual-level) and individual position (inter-individual level) to acquire richer activity representation. Specifically, it involves two branches: the trajectory branch utilizes individuals’ trajectories and positions to extract the motion feature at the individual and inter-individual levels. The joint branch extracts the motion features at the intra-individual level. Furthermore, the gated recurrent units (GRU) and Transformers are employed to enhance the corresponding features through gating mechanism and self-attention mechanism. The features from the two branches are concatenated for group activity recognition. The experiments on two public datasets demonstrate that our method achieves competitive performance and has potential benefits in terms of computational complexity.}, archive = {J_EAAI}, author = {Yihao Zheng and Zhuming Wang and Ke Gu and Lifang Wu and Zun Li and Ye Xiang}, doi = {10.1016/j.engappai.2024.109570}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109570}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale motion-based relational reasoning for group activity recognition}, volume = {139}, year = {2025}, }
• (2025). Kinematic matrix: One-shot human action recognition using kinematic data structure. EAAI, 139, 109569. (www)

  @article{ , abstract = {One-shot action recognition, which refers to recognizing human-performed actions using only a single training example, holds significant promise in advancing video analysis, particularly in domains requiring rapid adaptation to new actions. However, existing algorithms for one-shot action recognition face multiple challenges, including high computational complexity, limited accuracy, and difficulties in generalization to unseen actions. To address these issues, we propose a novel kinematic-based skeleton representation that effectively reduces computational demands while enhancing recognition performance. This representation leverages skeleton locations, velocities, and accelerations to formulate the one-shot action recognition task as a metric learning problem, where a model projects kinematic data into an embedding space. In this space, actions are distinguished based on Euclidean distances, facilitating efficient nearest-neighbour searches among activity reference samples. Our approach not only reduces computational complexity but also achieves higher accuracy and better generalization compared to existing methods. Specifically, our model achieved a validation accuracy of 78.5%, outperforming state-of-the-art methods by 8.66% under comparable training conditions. These findings underscore the potential of our method for practical applications in real-time action recognition systems.}, archive = {J_EAAI}, author = {Mohammad Hassan Ranjbar and Ali Abdi and Ju Hong Park}, doi = {10.1016/j.engappai.2024.109569}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109569}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Kinematic matrix: One-shot human action recognition using kinematic data structure}, volume = {139}, year = {2025}, }
• (2025). Region-guided spatial feature aggregation network for vehicle re-identification. EAAI, 139, 109568. (www)

  @article{ , abstract = {In the context of the advancement of smart city management, re-identification technology has emerged as an area of particular interest and research in the field of artificial intelligence, especially vehicle re-identification (re-ID), which aims to identify target vehicles in multiple non-overlapping fields of view. Most existing methods rely on fine-grained cues in the salient regions. Although impressive results have been achieved, these methods typically require additional auxiliary networks to localize the salient regions containing fine-grained cues. Meanwhile, changes in state such as illumination, viewpoint and occlusion can affect the position of the salient regions. To solve the above problems, this paper proposes a Region-guided Spatial Feature Aggregation Network (RSFAN) for vehicle re-ID, which forces the model to learn the latent information in the minor salient regions. Firstly, a Regional Localization (RL) module is proposed to automatically locate the salient regions without additional auxiliary networks. In addition, to mitigate the misguidance caused by the inaccurate salient regions, a Spatial Feature Aggregation (SFA) module is designed to weaken and enhance the expression of the salient and minor salient regions, respectively. Meanwhile, to enhance the diversity of the minor salient region-related information, a Cross-level Channel Attention (CCA) module is designed to implement cross-level interactions through the channel attention mechanism across different levels. Finally, to constrain the distributional differences between the salient regions and minor salient regions feature, a Distributional Variance (DV) loss is proposed. The extensive experiments show that the RSFAN has a good performance on VeRi-776, VehicleID, VeRi-Wild and Market1501 datasets.}, archive = {J_EAAI}, author = {Yanzhen Xiong and Jinjia Peng and Zeze Tao and Huibing Wang}, doi = {10.1016/j.engappai.2024.109568}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109568}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Region-guided spatial feature aggregation network for vehicle re-identification}, volume = {139}, year = {2025}, }
• (2025). Seafloor topography inversion from multi-source marine gravity data using multi-channel convolutional neural network. EAAI, 139, 109567. (www)

  @article{ , abstract = {Seafloor topography is extremely important for marine scientific surveys and research. Current physical methods have difficulties in integrating multi-source marine gravity data and recovering non-linear features. To overcome this challenge, a multi-channel convolutional neural network (MCCNN) is employed to establish the seafloor topography model. Firstly, the MCCNN model is trained using the input data from the 64 × 64 grid points centered around the control points. The input data includes the differences in position between calculation points and surrounding grid points, gravity anomaly, vertical gravity gradient, east component of deflection of the vertical and north component of deflection of the vertical, as well as the reference terrain information. Then, the data from the 64 × 64 grid points centered around the predicted points is inputted into the trained MCCNN model to obtain the predicted depth at those points. Finally, the predicted depth is utilized to establish the seafloor topography model of the study area. This method is tested in a local area located in the southern part of the Emperor Seamount Chain in the Northwest Pacific (31°N −37°N, 169°E −175°E). The root mean square of the differences between the resultant seafloor topography model and ship-borne bathymetric values at the check points is 88.48 m. This performance is commendable compared to existing models.}, archive = {J_EAAI}, author = {Bangzhuang Ge and Jinyun Guo and Qiaoli Kong and Chengcheng Zhu and Lingyong Huang and Heping Sun and Xin Liu}, doi = {10.1016/j.engappai.2024.109567}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109567}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Seafloor topography inversion from multi-source marine gravity data using multi-channel convolutional neural network}, volume = {139}, year = {2025}, }
• (2025). The docking control system of an autonomous underwater vehicle combining intelligent object recognition and deep reinforcement learning. EAAI, 139, 109565. (www)

  @article{ , abstract = {This study develops a visual-based docking system (VDS) for an autonomous underwater vehicle (AUV), significantly enhancing docking performance by integrating intelligent object recognition and deep reinforcement learning (DRL). The system overcomes traditional navigation limitations in complex and unpredictable environments by using a variable information dock (VID) for precise multi-sensor docking recognition in the AUV. Employing image-based visual servoing (IBVS) technology, the VDS efficiently converts 2D visual data into accurate 3D motion control commands. It integrates the YOLO (short for You Only Look Once) algorithm for object recognition and the deep deterministic policy gradient (DDPG) algorithm, improving continuous motion control, docking accuracy, and adaptability. Experimental validation at the National Cheng Kung University towing tank demonstrates that the VDS enhances control stability and operational reliability, reducing the mean absolute error (MAE) in depth control by 42.03% and pitch control by 98.02% compared to the previous method. These results confirm the VDS's reliability and its potential for transforming AUV docking.}, archive = {J_EAAI}, author = {Chao-Ming Yu and Yu-Hsien Lin}, doi = {10.1016/j.engappai.2024.109565}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109565}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {The docking control system of an autonomous underwater vehicle combining intelligent object recognition and deep reinforcement learning}, volume = {139}, year = {2025}, }
• (2025). Dynamic flame feature-driven prediction model for basic oxygen furnace steelmaking endpoint carbon content based on three-dimensional multi-layer complex networks. EAAI, 139, 109564. (www)

  @article{ , abstract = {Accurate prediction of carbon content at the endpoint is crucial for the endpoint management of Basic Oxygen Furnace (BOF) steelmaking. The carbon content in the molten pool is closely related to the dynamic and static characteristics of the flame at the furnace’s mouth. However, the flame’s texture change exhibits multidirectional and multiscale properties, posing challenges for existing algorithms to effectively extract dynamic color texture features. To address this issue, this paper proposes a dynamic texture feature extraction model based on a three-dimensional multi-layer complex network (3D-MLCN). The model constructs an unbounded complex network for a single-frame flame picture by integrating spatiotemporal position information of the image region’s centroid with color information, thereby quantizing the single-frame image into a complex network with spatiotemporal properties. Subsequently, a multi-scale multi-direction weighted dynamic color texture complex network is built for the flame video at the furnace mouth, utilizing the temporal index of the video frames in combination with vertex color values to capture the time-varying features of the flame video. The proposed method quantifies network characteristics through vertex degree distribution features to obtain dynamic color texture feature descriptors. These descriptors are then combined with static color texture features and color features to construct dynamic and static feature descriptors for the flame video, enabling the prediction of the endpoint carbon content using a regression model. By analyzing the actual production data of BOF steelmaking, the prediction accuracy of carbon content within the error range of ±0.02% is 87.91%, the R 2 value is 0.8547, and the RMSE value is 2.0959, which verifies the effectiveness of the proposed method.}, archive = {J_EAAI}, author = {JianXun Liu and Hui Liu and FuGang Chen and YunKe Su and Heng Li and XiaoJun Xue}, doi = {10.1016/j.engappai.2024.109564}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109564}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic flame feature-driven prediction model for basic oxygen furnace steelmaking endpoint carbon content based on three-dimensional multi-layer complex networks}, volume = {139}, year = {2025}, }
• (2025). Self-supervised contrastive learning for implicit collaborative filtering. EAAI, 139, 109563. (www)

  @article{ , abstract = {Recommendation systems are a critical application of artificial intelligence (AI), driving personalized user experiences across various platforms. Recent advancements in contrastive learning-based recommendation algorithms have led to significant progress in self-supervised recommendation. A key method in this field is Bayesian Personalized Ranking (BPR), which has become a dominant approach for implicit collaborative filtering. However, the challenge of false-positive and false-negative examples in implicit feedback continues to hinder accurate preference learning. In this study, we introduce an efficient self-supervised contrastive learning framework that enhances the supervisory signal by incorporating positive feature augmentation and negative label augmentation. Our theoretical analysis reveals that this approach is equivalent to maximizing the likelihood estimation with latent variables representing user interest centers. Additionally, we present a novel negative label augmentation technique that selects unlabeled examples based on their relative ranking positions, enabling efficient augmentation with constant time complexity. Validation on the MovieLens-100k, MovieLens-1M, Yahoo!-R3, Yelp2018, and Gowalla datasets demonstrates that our method achieves over a 5% improvement in precision compared to the widely used BPR optimization objective, while maintaining comparable runtime efficiency.}, archive = {J_EAAI}, author = {Shipeng Song and Bin Liu and Fei Teng and Tianrui Li}, doi = {10.1016/j.engappai.2024.109563}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109563}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Self-supervised contrastive learning for implicit collaborative filtering}, volume = {139}, year = {2025}, }
• (2025). Fine-tuning language model embeddings to reveal domain knowledge: An explainable artificial intelligence perspective on medical decision making. EAAI, 139, 109561. (www)

  @article{ , abstract = {Integrating large language models (LLMs) to retrieve targeted medical knowledge from electronic health records enables significant advancements in medical research. However, recognizing the challenges associated with using LLMs in healthcare is essential for successful implementation. One challenge is that medical records combine unstructured textual information with highly sensitive personal data. This, in turn, highlights the need for explainable Artificial Intelligence (XAI) methods to understand better how LLMs function in the medical domain. In this study, we propose a novel XAI tool to accelerate data-driven cancer research. We apply the Bidirectional Encoder Representations from Transformers (BERT) model to German language pathology reports examining the effects of domain-specific language adaptation and fine-tuning. We demonstrate our model on a real-world pathology dataset, analyzing the contextual representations of diagnostic reports. By illustrating decisions made by fine-tuned models, we provide decision values that can be applied in medical research. To address interpretability, we conduct a performance evaluation of the classifications generated by our fine-tuned model, as assessed by an expert pathologist. In domains such as medicine, inspection of the medical knowledge map in conjunction with expert evaluation reveals valuable information about how contextual representations of key disease features are categorized. This ultimately benefits data structuring and labeling and paves the way for even more advanced approaches to XAI, combining text with other input modalities, such as images which are then applicable to various engineering problems.}, archive = {J_EAAI}, author = {Ceca Kraišniković and Robert Harb and Markus Plass and Wael Al Zoughbi and Andreas Holzinger and Heimo Müller}, doi = {10.1016/j.engappai.2024.109561}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109561}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fine-tuning language model embeddings to reveal domain knowledge: An explainable artificial intelligence perspective on medical decision making}, volume = {139}, year = {2025}, }
• (2025). A deep learning ensemble approach for malware detection in internet of things utilizing explainable artificial intelligence. EAAI, 139, 109560. (www)

  @article{ , abstract = {The Internet of Things (IoT) has been popularized these days due to digitization and automation. It is deployed in various applications, i.e., smart homes, smart agriculture, smart transportation, smart healthcare, and industrial monitoring. In an IoT network, many IoT devices communicate with servers, or users access IoT devices through an open channel via a certain exchange of messages. Besides providing many benefits like efficiency, automation, and convenience, IoT presents significant security challenges due to a lack of proper standard security measures. Thus, malicious actors may be able to infect the network with malware. They may launch destructive attacks with the goal of stealing data or causing damage to the systems’ resources. This can be mitigated by introducing intrusion detection and prevention mechanisms in the network. An intelligent intrusion detection system is required to put preventative measures in place for secure communication and a malware-free network. In this article, we propose a deep learning based ensemble approach for IoT malware attack detection (in short, we call it as DLEX-IMD) trained and tested against benchmark datasets. The important measures, including accuracy, precision, recall, and F1-score, are used to evaluate the performance of the proposed DLEX-IMD. The performance of the proposed scheme is explained utilizing benchmark Explainable Artificial Intelligence (AI) method–LIME (Local Interpretable Model-Agnostic Explanations), which justifies the reliability of the proposed model training. The DLEX-IMD is also compared with a range of other closely related existing schemes and has shown better performance than those schemes with 99.96% accuracy and F1-score of 0.999.}, archive = {J_EAAI}, author = {Saksham Mittal and Mohammad Wazid and Devesh Pratap Singh and Ashok Kumar Das and M. Shamim Hossain}, doi = {10.1016/j.engappai.2024.109560}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109560}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep learning ensemble approach for malware detection in internet of things utilizing explainable artificial intelligence}, volume = {139}, year = {2025}, }
• (2025). A lightweight and explainable model for driver abnormal behavior recognition. EAAI, 139, 109559. (www)

  @article{ , abstract = {With the advancement of intelligent transportation systems, accurate identification of driver abnormal behavior is crucial for enhancing road safety. However, the limited computing power of vehicular systems poses a challenge for running efficient and explainable behavior recognition models. This paper proposes a lightweight and explainable driver abnormal behavior recognition model based on an improved You Only Look Once version 8 (YOLOv8). Firstly, a Spatial and Channel Reconstruction Convolution (SCConv) module is introduced to optimize the Convolution to Feature (C2f) structure, enhancing the model's feature extraction capabilities while reducing parameter redundancy. Secondly, a Spatial Pyramid Pooling with Fast Large Separable Kernel Attention (SPPF-LSKA) module is designed to better capture image context and integrate global information. Additionally, a Dynamic upsample (Dysample) module is introduced to improve the model's ability to capture subtle driver movements. Lastly, a Lightweight Shared Group Normalization Convolution Detection Head (LSGCDH) is designed to enhance the model's generalization ability, significantly reducing the model's computational load, parameter count, and size. Experimental results demonstrate that our approach has significant advantages for edge device deployment compared to mainstream algorithms. The visualization results effectively corroborate the role of each improved structure, enhancing the explainability of the abnormal behavior recognition model, which is beneficial for deployment in vehicular systems and contributes to improving road traffic safety.}, archive = {J_EAAI}, author = {Jingbin Hao and Xiaokai Sun and Xinhua Liu and Dezheng Hua and Jianhua Hu}, doi = {10.1016/j.engappai.2024.109559}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109559}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A lightweight and explainable model for driver abnormal behavior recognition}, volume = {139}, year = {2025}, }
• (2025). A novel anomaly detection and classification algorithm for application in tuyere images of blast furnace. EAAI, 139, 109558. (www)

  @article{ , abstract = {Traditional relying on manual experience to assess the tuyere status consumes significant human resources. In the era of intelligent blast furnaces and intensified smelting, this approach struggles to meet the demands for accuracy and real-time assessment, posing challenges to safety and efficiency of blast furnace production. Tuyere images exhibit high feature similarity, and the number of samples is often limited. Therefore, if a simple convolution operation is only used, it will be difficult to discern differences across various images. To address this challenge and cater to the requirements of intelligent tuyere status recognition across different steel enterprises, we designed a novel deep neural network algorithm called ES-SFRNet (Enhanced Sequential: Feature Fusion and Recognition Network), building upon our prior research. The algorithm concurrently modeled tuyere images alongside relevant time series data, comprising three components: Feature pre-extraction, Tuyere status recognition, and Generalization & Robustness. The first two modules focus on feature extraction and fusion of tuyere images, while leveraging edge detection information from the image, we developed a mathematical index A r (Area Ratio) to serve as an auxiliary criterion for tuyere status recognition. Given the model's future scalability and multi-scenario application, the final module focuses on knowledge integration and parameter control. Test results reveal an overall accuracy rate of 99.3% for the ES-SFRNet algorithm, effectively capturing key parameters to facilitate on-site operations. In comparison to other mainstream object detection algorithms, our algorithm framework excels in tuyere image feature extraction and recognition, which can offer broad applications to Chinese blast furnace ironmaking industry.}, archive = {J_EAAI}, author = {Yifan Duan and Xiaojie Liu and Ran Liu and Xin Li and Hongwei Li and Hongyang Li and Yanqin Sun and Yujie Zhang and Qing Lv}, doi = {10.1016/j.engappai.2024.109558}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109558}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel anomaly detection and classification algorithm for application in tuyere images of blast furnace}, volume = {139}, year = {2025}, }
• (2025). An effective multi-agent-based graph reinforcement learning method for solving flexible job shop scheduling problem. EAAI, 139, 109557. (www)

  @article{ , abstract = {Flexible job shop scheduling problem (FJSP) is a complex optimization problem in intelligent manufacturing and plays a key role in improving productivity, which is characterized by that each operation can be processed by multiple machines. Most current research into FJSP focuses on finding a higher-quality scheduling scheme in a shorter time. However, existing studies are hard to optimize the operation sequencing and machine assignment strategies simultaneously, which is critical for making the optimal scheduling decision. Therefore, a multi-agent-based graph reinforcement learning (MAGRL) method is proposed to effectively solve FJSP. Firstly, the FJSP is modeled into two Markov decision processes (MDPs), where the operation and machine agents are adopted to control the operation sequencing and machine assignment respectively. Secondly, to effectively predict the operation sequencing and machine assignment strategies, an encoder-double-decoder architecture is designed, including an improved graph attention network (IGAT)-based encoder, an operation strategy network-based decoder, and a machine strategy network-based decoder. Thirdly, an automatic entropy adjustment multi-agent proximal policy optimization (AEA-MAPPO) algorithm is proposed for effectively training the operation and machine strategy networks to optimize the operation sequencing and machine assignment strategies simultaneously. Finally, the effectiveness of MAGRL is verified through experimental comparisons with the classical scheduling rules and state-of-the-art methods to solve FJSP. The results achieved on the randomly generated FJSP instances and two common benchmarks indicate that MAGRL can consume less solution time to achieve higher solution quality in solving different-sized FJSP instances, and the overall performance of MAGRL is superior to that of the comparison methods.}, archive = {J_EAAI}, author = {Lanjun Wan and Long Fu and Changyun Li and Keqin Li}, doi = {10.1016/j.engappai.2024.109557}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109557}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An effective multi-agent-based graph reinforcement learning method for solving flexible job shop scheduling problem}, volume = {139}, year = {2025}, }
• (2025). Residual-based attention physics-informed neural networks for spatio-temporal ageing assessment of transformers operated in renewable power plants. EAAI, 139, 109556. (www)

  @article{ , abstract = {Transformers are crucial for reliable and efficient power system operations, particularly in supporting the integration of renewable energy. Effective monitoring of transformer health is critical to maintain grid stability and performance. Thermal insulation ageing is a key transformer failure mode, which is generally tracked by monitoring the hotspot temperature (HST). However, HST measurement is complex, costly, and often estimated from indirect measurements. Existing HST models focus on space-agnostic thermal models, providing worst-case HST estimates. This article introduces a spatio-temporal model for transformer winding temperature and ageing estimation, which leverages physics-based partial differential equations (PDEs) with data-driven Neural Networks (NN) in a Physics Informed Neural Networks (PINNs) configuration to improve prediction accuracy and acquire spatio-temporal resolution. The computational accuracy of the PINN model is improved through the implementation of the Residual-Based Attention (PINN-RBA) scheme that accelerates the PINN model convergence. The PINN-RBA model is benchmarked against self-adaptive attention schemes and classical vanilla PINN configurations. For the first time, PINN based oil temperature predictions are used to estimate spatio-temporal transformer winding temperature values, validated through PDE numerical solution and fiber optic sensor measurements. Furthermore, the spatio-temporal transformer ageing model is inferred, which supports transformer health management decision-making. Results are validated with a distribution transformer operating on a floating photovoltaic power plant.}, archive = {J_EAAI}, author = {Ibai Ramirez and Joel Pino and David Pardo and Mikel Sanz and Luis del Rio and Alvaro Ortiz and Kateryna Morozovska and Jose I. Aizpurua}, doi = {10.1016/j.engappai.2024.109556}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109556}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Residual-based attention physics-informed neural networks for spatio-temporal ageing assessment of transformers operated in renewable power plants}, volume = {139}, year = {2025}, }
• (2025). Adaptive masked network for ultra-short-term photovoltaic forecast. EAAI, 139, 109555. (www)

  @article{ , abstract = {In recent years, power grid companies have faced increasingly stringent requirements for accurate prediction of photovoltaic (PV) power generation with the rapid development of PV technologies. In ultra-short-term forecasting, PV power generation exhibits strong temporal correlations, leading to high data redundancy. To address this issue, we propose an adaptive masked network (ASMNet) to enhance the accuracy of ultra-short-term PV forecasting. Specifically, this method improves the feature extraction of short-term fluctuations within historical time periods by down-weighting less significant temporal segments during the learning process. It captures the uncertain effects of environmental changes and provides a better understanding of the impacts of ultra-short-term fluctuations. We test our model on three public PV power generation datasets, and it achieves the best performance with a root mean square error of 21.42, 0.2824 and 23.36 for the Belgian, American National Renewable Energy Laboratory, and Desert Knowledge Australia Solar Center datasets, respectively. Additionally, the proposed model demonstrates a 0.01%–0.50% improvement in coefficient of determination compared to baseline models across all datasets, highlighting its superior performance and effectiveness in ultra-short-term PV forecasting.}, archive = {J_EAAI}, author = {Qiaoyu Ma and Xueqian Fu and Qiang Yang and Dawei Qiu}, doi = {10.1016/j.engappai.2024.109555}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109555}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive masked network for ultra-short-term photovoltaic forecast}, volume = {139}, year = {2025}, }
• (2025). Picture fuzzy complex proportional assessment approach with step-wise weight assessment ratio analysis and criteria importance through intercriteria correlation. EAAI, 139, 109554. (www)

  @article{ , abstract = {The concept of the picture fuzzy set (PiFS) significantly enhances the multi-criteria decision-making (MCDM) process by incorporating membership value (MV), non-membership value (NMV), and a neutral component. PiFS extends the capabilities of traditional fuzzy sets (FSs), intuitionistic fuzzy sets (IFSs), and other fuzzy models. This paper introduces a novel MCDM approach, the picture fuzzy SWARA-CRITIC-COPRAS (PiF-SCC) method, specifically designed to assist decision-makers (DMs) in evaluating and selecting dynamic digital marketing (DDM) technologies within PiFS settings. The proposed method integrates the strengths of PiFS with step-wise weight assessment ratio analysis (SWARA), criteria importance through intercriteria correlation (CRITIC), and complex proportional assessment (COPRAS), aiming to improve the precision and effectiveness of technology evaluations. To validate the approach, a case study is conducted on DDM technology assessment within a specific business context. The PiF-SCC technique is applied to rank technological options using linguistic terms (LTs), PiFS numbers, an accuracy function (AF), and a score function (SF). Additionally, a comprehensive sensitivity analysis is performed to evaluate the robustness of the proposed method under different input scenarios and uncertainties. A thorough comparison with existing techniques is also provided, demonstrating the superior decision-making capability of the new approach, which leads to more accurate and dependable technology selection results. The manuscript also discusses marginal implications and limitations, along with potential future research directions to further enhance the applicability and effectiveness of the proposed approach.}, archive = {J_EAAI}, author = {Ayesha Razzaq and Zareen A. Khan and Khalid Naeem and Muhammad Riaz}, doi = {10.1016/j.engappai.2024.109554}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109554}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Picture fuzzy complex proportional assessment approach with step-wise weight assessment ratio analysis and criteria importance through intercriteria correlation}, volume = {139}, year = {2025}, }
• (2025). Word-sequence entropy: Towards uncertainty estimation in free-form medical question answering applications and beyond. EAAI, 139, 109553. (www)

  @article{ , abstract = {Uncertainty estimation is crucial for the reliability of safety-critical human and artificial intelligence (AI) interaction systems, particularly in the domain of healthcare engineering. However, a robust and general uncertainty measure for free-form answers has not been well-established in open-ended medical question-answering (QA) tasks, where generative inequality introduces a large number of irrelevant words and sequences within the generated set for uncertainty quantification (UQ), which can lead to biases. This paper proposes Word-Sequence Entropy ( WSE ), which calibrates uncertainty at both the word and sequence levels based on semantic relevance, highlighting keywords and enlarging the generative probability of trustworthy responses when performing UQ. We compare WSE with six baseline methods on five free-form medical QA datasets, utilizing seven popular large language models (LLMs), and demonstrate that WSE exhibits superior performance in accurate UQ under two standard criteria for correctness evaluation. Additionally, in terms of the potential for real-world medical QA applications, we achieve a significant enhancement (e.g., a 6.36% improvement in model accuracy on the COVID-QA dataset) in the performance of LLMs when employing responses with lower uncertainty that are identified by WSE as final answers, without requiring additional task-specific fine-tuning or architectural modifications.}, archive = {J_EAAI}, author = {Zhiyuan Wang and Jinhao Duan and Chenxi Yuan and Qingyu Chen and Tianlong Chen and Yue Zhang and Ren Wang and Xiaoshuang Shi and Kaidi Xu}, doi = {10.1016/j.engappai.2024.109553}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109553}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Word-sequence entropy: Towards uncertainty estimation in free-form medical question answering applications and beyond}, volume = {139}, year = {2025}, }
• (2025). Data-driven drift detection and diagnosis framework for predictive maintenance of heterogeneous production processes: Application to a multiple tapping process. EAAI, 139, 109552. (www)

  @article{ , abstract = {The rise of Industry 4.0 technologies has revolutionized industries, enabled seamless data access, and fostered data-driven methodologies for improving key production processes such as maintenance. Predictive maintenance has notably advanced by aligning decisions with real-time system degradation. However, data-driven approaches confront challenges such as data availability and complexity, particularly at the system level. Most approaches address component-level issues, but system complexity exacerbates problems. In the realm of predictive maintenance, this paper proposes a framework for addressing drift detection and diagnosis in heterogeneous manufacturing processes. The originality of the paper is twofold. First, this paper proposes algorithms for handling drift detection and diagnosing heterogeneous processes. Second, the proposed framework leverages several machine learning techniques (e.g., novelty detection, ensemble learning, and continuous learning) and algorithms (e.g., K-Nearest Neighbors, Support Vector Machine, Random Forest and Long-Short Term Memory) for enabling the concrete implementation and scalability of drift detection and diagnostics on industrial processes. The effectiveness of the proposed framework is validated through metrics such as accuracy, precision, recall, F1-score, and variance. Furthermore, this paper demonstrates the relevance of combining machine learning and deep learning algorithms in a production process of SEW USOCOME, a French manufacturer of electric gearmotors and a market leader. The results indicate a satisfactory level of accuracy in detecting and diagnosing drifts, and the adaptive learning loop effectively identifies new drift and nominal profiles, thereby validating the robustness of the framework in real industrial settings.}, archive = {J_EAAI}, author = {Julien Chapelin and Alexandre Voisin and Bertrand Rose and Benoît Iung and Lionel Steck and Ludovic Chaves and Mathieu Lauer and Olivier Jotz}, doi = {10.1016/j.engappai.2024.109552}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109552}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven drift detection and diagnosis framework for predictive maintenance of heterogeneous production processes: Application to a multiple tapping process}, volume = {139}, year = {2025}, }
• (2025). Combining model-based and learning-based anomaly detection schemes for increased performance and safety of aircraft braking controllers. EAAI, 139, 109551. (www)

  @article{ , abstract = {In aircraft, the braking system is a safety-critical and heavily used component of the landing gear, prone to significant wear. Anomalies arising in the wear dynamics can degrade the performance of the braking system and compromise the safety of ground handling maneuvers. In this work, we tackle the problem of detecting incipient anomalies in aircraft brakes in a tightly coupled implementation with the Brake Control Unit (BCU). Two complementary approaches are presented. The first one is an observer-based architecture designed on the longitudinal aircraft dynamics that returns physically interpretable outputs connected to the wear process and allows us to improve braking performance online. The second one is an end-to-end convolutional autoencoder-based architecture that returns an anomaly score computed on data collected by the BCU with inherent robustness to modeling uncertainty, which the model-based one does not. A combined architecture that allows one to exploit the features of both model-based and learning-based approaches is proposed, which shows its capability of optimally blending the two. The approaches are evaluated in a MATLAB/Simulink multibody simulation environment that is able to replicate the braking actuator wear dynamics, demonstrating remarkable performances in anomaly detection, anti-skid control performance, and safety improvement.}, archive = {J_EAAI}, author = {José Joaquín Mendoza Lopetegui and Mara Tanelli}, doi = {10.1016/j.engappai.2024.109551}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109551}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Combining model-based and learning-based anomaly detection schemes for increased performance and safety of aircraft braking controllers}, volume = {139}, year = {2025}, }
• (2025). Adversarial-causal representation learning networks for machine fault diagnosis under unseen conditions based on vibration and acoustic signals. EAAI, 139, 109550. (www)

  @article{ , abstract = {To address the challenges of obtaining diverse data, domain generalization (DG) methods for fault diagnosis have been developed. Domain adversarial methods are currently the most popular, due to their ability to handle data from unknown domains without requiring target domain information. However, their capacity to extract domain-irrelevant features remains challenging, often resulting in accuracy below 90% in many DG scenarios. This limitation stems from their inability to fully capture global dependencies, causing feature entanglement and redundant dependencies. To address these issues, we proposed a novel intelligent fault diagnosis method called Adversarial-Causal Representation Learning Networks (ACRLN), which is based on causal learning. By spatial mask domain adversarial method, ACRLN can significantly enhance data utilization by fully capturing the global dependency that are often ignored by domain adversarial algorithms. At the same time, causal learning is integrated into the ACRLN to further accomplish feature decoupling and the reduction of redundant dependency. This is achieved through channel feature orthogonality method combined with a loss function rooted in correlation analysis. Moreover, it adeptly addresses the spill-over effect often encountered in causal learning. Finally, ACRLN achieves better results and proves its effectiveness by comparison with several state-of-the-art fault diagnosis and DG algorithms on multiple datasets.}, archive = {J_EAAI}, author = {Fei Wu and Zhuohang Xiang and Dengyu Xiao and Yaodong Hao and Yi Qin and Huayan Pu and Jun Luo}, doi = {10.1016/j.engappai.2024.109550}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109550}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adversarial-causal representation learning networks for machine fault diagnosis under unseen conditions based on vibration and acoustic signals}, volume = {139}, year = {2025}, }
• (2025). Data-driven nonmodel seismic assessment of eccentrically braced frames with soil-structure interaction. EAAI, 139, 109549. (www)

  @article{ , abstract = {This study presents a nonmodel-based machine learning framework for estimating engineering demand parameters (EDPs) of eccentrically braced frames with soil-structure interaction effects. The objective is to estimate residual and peak story drift ratio, peak floor acceleration, and develop fragility curves using traditional regression equations and advanced machine-learning techniques. Correction coefficients are developed to improve prediction accuracy by accounting for soil-structure interaction. A comprehensive database, including incremental dynamic analysis results of 4- and 8-story frames, is developed, consisting of 109,841 data points. The database includes fixed-base models and models with various soil-structure interaction values, subjected to 44 far-field ground motions. Four scenarios are introduced considering various input variables to compare the impact of soil-structure interaction. Findings reveal the effects of soil-structure interaction features on the performance of machine learning algorithms, increasing by up to 17.61% of the coefficient of determination. Utilizing the predicted story drift ratio, two types of fragility curves indicate more precise predictions, emphasizing the impact of soil-structure interaction effects at lower damage levels. A graphical user interface has been developed to predict fragility curves based on various inputs to promote the practical use of machine learning in engineering. Two new 4-story frames are used as case studies, subjected to unseen ground motions to assess the application of trained machine learning algorithms. Prediction errors in input-output scenarios considering soil-structure interaction range from 3% to 18% for new frames. The proposed approach for predicting EDPs is further acknowledged by evaluating a real instrumented five-story steel frame office building.}, archive = {J_EAAI}, author = {Mahshad Jamdar and Kiarash M. Dolatshahi and Omid Yazdanpanah}, doi = {10.1016/j.engappai.2024.109549}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109549}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Data-driven nonmodel seismic assessment of eccentrically braced frames with soil-structure interaction}, volume = {139}, year = {2025}, }
• (2025). Bearing fault diagnosis for variable working conditions via lightweight transformer and homogeneous generalized contrastive learning with inter-class repulsive discriminant. EAAI, 139, 109548. (www)

  @article{ , abstract = {As indispensable components of rolling axle boxes, the condition of the bearings affects the safety of the traveling train. Therefore, bearing fault diagnosis is an imperative prerequisite for train safety. However, the diagnosis performance under variable working conditions is degraded owing to the large difference in the sample distribution and fewer samples. Although unsupervised domain adaptation models can solve these problems, environmental noise causes the fault features extracted from the two domains to overlap. Ultimately, the discriminative properties of the different samples remain insufficient. Therefore, we propose a rolling fault diagnosis approach for variable working conditions via lightweight Transformer and homogeneous generalized contrastive learning with inter-class repulsive discriminant (HGCL-ICRD). First, a deformable Transformer with lightweight manner is constructed to extract fault features from historical working conditions. Then, the source domain clustering cluster points are used to construct the positive and negative samples of the target domain to achieve the redistribution of the number. On this basis, the homogeneous generalized contrastive learning approach is built to make the samples to be tested have better classifiability. Finally, an inter-class repulsive discriminant term is constructed to minimize the sample distributional difference between the two domains. Furthermore, we construct an improved gray wolf algorithm to optimize the HGCL-ICRD. Extensive experiments on three datasets demonstrate that our model can perform high-precision and high-efficiency diagnosis under variable working conditions.}, archive = {J_EAAI}, author = {Qiang Zhou and Wengang Ma and Yadong Zhang and Jin Guo}, doi = {10.1016/j.engappai.2024.109548}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109548}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Bearing fault diagnosis for variable working conditions via lightweight transformer and homogeneous generalized contrastive learning with inter-class repulsive discriminant}, volume = {139}, year = {2025}, }
• (2025). TopoPIS: Topology-constrained pipe instance segmentation via adaptive curvature convolution. EAAI, 139, 109547. (www)

  @article{ , abstract = {Precise and fast pipe instance segmentation is a critical component in industrial automatic assembly, facilitating accurate object detection and pose estimation, optimizing and supervising the assembly process. However, this problem is challenging due to topological errors on fine-scale structures caused by the pipes being complex and slender. To address these challenges, we propose a topology-constrained pipe instance segmentation network (TopoPIS) for complex stacking scene to achieve accurate segmentation with topological correctness. To better extract the features of complex and variable morphological pipes, adaptive curvature convolution is introduced to dynamically adapt to the slender pipe structure and capture critical features. To handle topological errors like broken connections, we propose a novel topological constraint loss function based on persistent homology, which greatly improves the topological continuity of the segmentation. Experimental results on real-world and unseen datasets demonstrate that our TopoPIS outperforms other methods regrading segmentation accuracy and topological continuity.}, archive = {J_EAAI}, author = {Jia Hu and Jianhua Liu and Shaoli Liu}, doi = {10.1016/j.engappai.2024.109547}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109547}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {TopoPIS: Topology-constrained pipe instance segmentation via adaptive curvature convolution}, volume = {139}, year = {2025}, }
• (2025). Constrained multi-objective optimization assisted by convergence and diversity auxiliary tasks. EAAI, 139, 109546. (www)

  @article{ , abstract = {In the field of constrained multi-objective optimization, constructing auxiliary tasks can guide the algorithm to achieve efficient search. Different forms of auxiliary tasks have their own advantages, and a reasonable combination can effectively improve the performance of the algorithm. Inspired by this, a Constrained Multi-objective Optimization Evolutionary Algorithm based on Convergence and Diversity auxiliary Tasks (CMOEA-CDT) is proposed. This algorithm achieves efficient search through simultaneous optimization and knowledge transfer of the main task, convergence auxiliary task, and diversity auxiliary task. Specifically, the main task is to find feasible Pareto front, which improves the global exploration and local exploitation of the algorithm through knowledge transfer from the convergence and diversity auxiliary tasks. In addition, the convergence auxiliary task helps the main task population traverse infeasible obstacles by ignoring constraints to achieve global search. The diversity auxiliary task aims to provide local diversity to the regions around the main task population to exploit promising search regions. The convergence and diversity of the algorithm are significantly improved by knowledge transfer between the convergence auxiliary task, diversity auxiliary task, and main task. CMOEA-CDT is compared with five state-of-the-art constrained multi-objective evolutionary optimization algorithms on 37 benchmark problems and a disc brake engineering design problem. The experimental results indicate that the proposed CMOEA-CDT respectively obtains 19 and 20 best results on the two indicators, and achieves the best performance on disc brake engineering design problem.}, archive = {J_EAAI}, author = {Qianlong Dang and Wutao Shang and Zhengxin Huang and Shuai Yang}, doi = {10.1016/j.engappai.2024.109546}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109546}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Constrained multi-objective optimization assisted by convergence and diversity auxiliary tasks}, volume = {139}, year = {2025}, }
• (2025). Integrated metaheuristic approaches for estimation of fracture porosity derived from fullbore formation micro-imager logs: Reaping the benefits of stand-alone and ensemble machine learning models. EAAI, 139, 109545. (www)

  @article{ , abstract = {Fracture porosity is one of the most effective parameters for reservoir productivity and recovery efficiency. This study aims to predict and improve the accuracy of fracture porosity estimation through the application of advanced machine learning (ML) algorithms. A novel approach was introduced for the first time to estimate fracture porosity by reaping the benefits of petrophysical and fullbore formation micro-imager (FMI) data based on employing various stand-alone, ensemble, optimisation and multi-variable linear regression (MVLR) algorithms. This study proposes a ground-breaking two-step committee machine (CM) model. Petrophysical data containing compressional sonic-log travel time, deep resistivity, neutron porosity and bulk density (inputs), along with FMI-derived fracture porosity values (outputs), were employed. Nine stand-alone ML algorithms, including back-propagation neural network, Takagi and Sugeno fuzzy system, adaptive neuro-fuzzy inference system, decision tree, radial basis function, extreme gradient boosting, least-squares boosting, least squares support vector regression and k-nearest neighbours, were trained for initial estimation. To improve the efficacy of stand-alone algorithms, their outputs were combined in CM structures using optimisation algorithms. This integration was applied through five optimisation algorithms, including genetic algorithm, ant colony, particle swarm, covariance matrix adaptation evolution strategy (CMA-ES) and Coyote optimisation algorithm. Considering the lowest error, the CM with CMA-ES showed superior performance. Subsequently, MVLR was employed to improve the CMs further. Employing MVLR to combine the CMs yielded a 57.85% decline in mean squared error and a 4.502% improvement in the correlation coefficient compared to the stand-alone algorithms. The results of the benchmark analysis validated the efficacy of this approach.}, archive = {J_EAAI}, author = {Ali Gholami Vijouyeh and Maha Raoof Hamoudi and Dyana Aziz Bayz and Ali Kadkhodaie}, doi = {10.1016/j.engappai.2024.109545}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109545}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Integrated metaheuristic approaches for estimation of fracture porosity derived from fullbore formation micro-imager logs: Reaping the benefits of stand-alone and ensemble machine learning models}, volume = {139}, year = {2025}, }
• (2025). Investigation of hybrid modeling and its transferability in building load prediction used for district heating systems. EAAI, 139, 109544. (www)

  @article{ , abstract = {In the district heating systems, the historical operation data of the buildings in those areas would be partially or entirely missing. The traditional data-driven model is hard to predict the ground truth results because the historical data is not available for model training. However, utilizing the physics-based methods for load calculation takes a long time to process and encounters low accuracy issues. This paper investigates several hybrid models that integrate the data-driven model and the physics-based models with different fusion methods. The physics-based models calculate envelope load and infiltration load, based on Fourier's law and the grand canonical ensemble theory, respectively. After undergoing load processing, features fusion, and residual connection, the best advanced hybrid models generate 21.35%, 16.35%, and 12.73% better prediction results compared with the data-driven model. Moreover, the advanced hybride models also perform strong transferability across all the data quantity groups. In terms of practical application, the advanced hybrid models could be deployed with effective generalization in limited data scenarios and robust transfer capabilities. The selected best model constructed by hybrid modeling displays the highest performance and saves the total training costs with strong transferability.}, archive = {J_EAAI}, author = {Ning Zhang and Wei Zhong and Xiaojie Lin and Liuliu Du-Ikonen and Tianyue Qiu}, doi = {10.1016/j.engappai.2024.109544}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109544}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Investigation of hybrid modeling and its transferability in building load prediction used for district heating systems}, volume = {139}, year = {2025}, }
• (2025). Real-time joint recognition of weather and ground surface conditions by a multi-task deep network. EAAI, 139, 109543. (www)

  @article{ , abstract = {Climate change and the occurrence of intense and unexpected weather events highlighted the need for real-time weather warning systems, especially in smart roads and isolated scenarios like rural areas. In this work, we propose to jointly recognize the weather and the ground surface conditions using existing video surveillance systems. Previous works separately tackled these two tasks even if they are correlated to each other. We propose a convolutional neural network with shared weights in the lower layers and two separate classification branches on top to exploit the correlation between the tasks and, at the same time, learn diverse high-level features for each task. Moreover, the network architecture implements attention mechanisms allowing the classification branches to focus on diverse image regions. The method is versatile and allows us to train the network on partially labeled data. The experimental analysis on real data demonstrate the effectiveness of the proposed method on both tasks, confirmed by the accuracy comparison with existing methods for the recognition of weather and ground surface conditions. The multi-task solution improves the inference speed (50 frames per second) and reduces the required memory (less than 1 GB) with respect to a system with two different single-task approaches; these results confirm that the proposed solution is ready for video surveillance applications to support smart cities.}, archive = {J_EAAI}, author = {Diego Gragnaniello and Antonio Greco and Carlo Sansone and Bruno Vento}, doi = {10.1016/j.engappai.2024.109543}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109543}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Real-time joint recognition of weather and ground surface conditions by a multi-task deep network}, volume = {139}, year = {2025}, }
• (2025). A graph convolutional neural network model based on fused multi-subgraph as input and fused feature information as output. EAAI, 139, 109542. (www)

  @article{ , abstract = {The graph convolution neural network (GCN)-based node classification model tackles the challenge of classifying nodes in graph data through learned feature representations. However, most existing graph neural networks primarily focus on the same type of edges, which might not accurately reflect the intricate real-world graph structure. This paper introduces a novel graph neural network model, MF-GCN, which integrates subgraphs with various edge types as input and combines feature information from each graph convolutional neural network layer to produce the final output. This model learns node feature representations by separately feeding subgraphs with different edge types into the graph convolutional layer. It then computes the weight vectors for fusing various edge type subgraphs based on the learned node features. Additionally, to efficiently extract feature information, the outputs of each graph convolution layer, without an activation function, are weighted and summed to obtain the final node features. This approach resolves the challenges of determining fusion weights and effectively extracting feature information during subgraph fusion. Experimental results show that the proposed model significantly improves performance on all three datasets, highlighting its effectiveness in node representation learning tasks.}, archive = {J_EAAI}, author = {Junwen Lu and Zeji Chen and XiaoLong Deng}, doi = {10.1016/j.engappai.2024.109542}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109542}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A graph convolutional neural network model based on fused multi-subgraph as input and fused feature information as output}, volume = {139}, year = {2025}, }
• (2025). Guided deep reinforcement learning framework using automated curriculum scheme for accurate motion planning. EAAI, 139, 109541. (www)

  @article{ , abstract = {Collaborative robotic arms in smart factories should ensure the safety and interactivity during their operation such as reaching and grasping objects. Especially, the advanced motion planner including the path planning and the motion control functions is essential for human-machine co-working. Since the traditional physics-based motion planning approaches require extreme computational resources to obtain near-optimal solutions, deep reinforcement learning algorithms have been actively adopted and have effectively solved the limitation. They, however, have the easy task preference problem, primarily taking the simpler ways for the more rewards, due to randomly training the agents how to reach the target points in the large-scale search spaces. Therefore, we propose a novel curriculum-based deep reinforcement learning framework that makes the agents learn the motion planning tasks in unbiased ways from the ones with the low complexities to the others with the high complexities. It uses the unsupervised learning algorithms to cluster the target points with the similar task complexities for generating the effective curriculum. In addition, the review and buffer flushing mechanisms are integrated into the framework to mitigate the catastrophic forgetting problem where the agent abruptly lose the previous learned knowledge upon learning new one in the curriculum. The evaluation results of the proposed framework show that the curriculum significantly enhances the success rate on the task with the highest complexity from 12% to 56% and the mechanisms improve the success rate on the tasks with the easier complexities from an average of 66% to 76.5%, despite requiring less training time.}, archive = {J_EAAI}, author = {Deun-Sol Cho and Jae-Min Cho and Won-Tae Kim}, doi = {10.1016/j.engappai.2024.109541}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109541}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Guided deep reinforcement learning framework using automated curriculum scheme for accurate motion planning}, volume = {139}, year = {2025}, }
• (2025). Compact representation for memory-efficient storage of images using genetic algorithm-guided key pixel selection. EAAI, 139, 109540. (www)

  @article{ , abstract = {In the past few years, we have observed rapid growth in digital content. Even in the biological domain, the arrival of microscopic and nanoscopic images and videos captured for biological investigations increases the need for space to store them. Hence, storing these data in a storage-efficient manner is a pressing need. In this work, we have introduced a compact image representation technique with an eye on preserving the shape that can shrink the memory requirement to store. The compact image representation is different from image compression since it does not include any encoding mechanism. Rather, the idea is that this mechanism stores the positions of key pixels, and when required, the original image can be regenerated. The genetic algorithm is used to select key pixels, while the Gaussian kernel performs the reconstruction task with the help of the positions of the selected key pixels. The model is tested on four different datasets. The proposed technique shrinks the memory requirement by 87% to 98% while evaluated using the bit reduction rate. However, the reconstructed images’ quality is a bit low when evaluated using metrics like structural similarity index (ranges between 0.81 to 0.94), or root means squared error (ranges between 0.06 to 0.08). To investigate the impact of quality reduction in reconstructed images in real-life applications, we performed image classification using reconstructed samples and found 0.13% to 2.30% classification accuracy reduction compared to when classification is done using original samples. The proposed model’s performance is comparable to state-of-the-art’s similar solutions.}, archive = {J_EAAI}, author = {Samir Malakar and Nirwan Banerjee and Dilip K. Prasad}, doi = {10.1016/j.engappai.2024.109540}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109540}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Compact representation for memory-efficient storage of images using genetic algorithm-guided key pixel selection}, volume = {139}, year = {2025}, }
• (2025). Estimation of the legs’ state of a mobile robot based on long short-term memory network. EAAI, 139, 109539. (www)

  @article{ , abstract = {In this paper, we propose a model-free method for estimating the height and the Ground Reaction Force (GRF) for the legs of mobile robots using the Long Short-Term Memory network (LSTM). The method does not require the presence of a force sensor at each foot, and it is proven to be robust to the changes that may occur in the dynamics. First, we generated a dataset to estimate the state of the legs for the non-damaged robot and for various types of damage situations; a disabled leg with working joints’ encoders, a fully disabled leg, and a removed leg. The network was tuned to obtain the highest stable R 2 score. Then, we studied the effect of the available sensors on the results of estimation which proved the sufficiency of using just the joint encoders which led to reducing the computational time by 17%. The sequence length required for estimation is also optimized to less than half of the gait period. The estimation results on a simulated hexapod robot and on a dataset recorded using a real four-legged robot proved the effectiveness and reliability of the proposed method as the R 2 score reached 94% with the damaged hexapod robot and 92% with the real four-legged robot, and that also proved the ability of our proposed method to be generalized to different types of robots.}, archive = {J_EAAI}, author = {Ahed Albadin and Chadi Albitar and Michel Alsaba}, doi = {10.1016/j.engappai.2024.109539}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109539}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Estimation of the legs’ state of a mobile robot based on long short-term memory network}, volume = {139}, year = {2025}, }
• (2025). A deep sequence-to-sequence model for power swing blocking of distance protection in power transmission lines. EAAI, 139, 109538. (www)

  @article{ , abstract = {As the primary protection method for transmission lines, distance relays are prone to malfunction during power swings. In fact, the inability of distance relays to differentiate between power swings and short-circuit faults imposes a significant risk to power system stability that can result in blackouts. In recent years, there has been increasing interest in leveraging machine learning techniques to identify various types of faults and power swings in electrical systems. However, previous works mainly focus on fault classification, which is mostly done after a long period from the moment of fault initiation. This is the reason for requiring extensive post-fault data for diagnosis. To address this challenge, this study proposes a predictive protection strategy utilizing deep learning methodologies, specifically a sequence-to-sequence model, to monitor electrical power systems continuously. The objective is to effectively detect power swings from short-circuit faults with minimal reliance on post-fault data and accurately identify short-circuit faults during power swings. In the proposed approach, features are extracted from grid current signals using the Hilbert transform and empirical mode decomposition algorithms. These features are then fed into the sequence-to-sequence model, which issues block/unblock commands upon confirming the presence of a power swing or fault during the power swing. Results from various simulations conducted on an IEEE 39-bus grid in DIgSILENT and MATLAB environments demonstrate that the proposed scheme outperforms baseline methods in the detection of short-circuit faults, power swings, and short-circuit faults occurring during the power swings. The timely and correct operation of the proposed protection scheme contributes to the stability of transmission lines and power systems.}, archive = {J_EAAI}, author = {Amin Mehdipour Birgani and Mohammadreza Shams and Mohsen Jannati and Farhad Hatami Aloghareh}, doi = {10.1016/j.engappai.2024.109538}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109538}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A deep sequence-to-sequence model for power swing blocking of distance protection in power transmission lines}, volume = {139}, year = {2025}, }
• (2025). A chinese named entity recognition method for landslide geological disasters based on deep learning. EAAI, 139, 109537. (www)

  @article{ , abstract = {Landslide Named Entity Recognition (LNER) involves extracting specific entities from Chinese unstructured landslide disaster texts, which is crucial for constructing a knowledge graph and supporting landslide prevention efforts. This study proposes a deep learning-based LNER model that utilizes Bidirectional Encoder Representations from Transformer (BERT) for word embeddings and integrates the Conditional Random Fields (CRF) algorithm and projected gradient descent (PGD) adversarial neural networks to enhance sequence labeling accuracy. The practical implications of this research lie in improving the efficiency and precision of disaster information extraction, aiding in real-time decision-making and risk mitigation strategies. Experiments on the constructed dataset show that the model effectively identifies eight types of landslide entities, achieving a highest F1 score of 89.7%.}, archive = {J_EAAI}, author = {Banghui Yang and Chunlei Zhou and Suju Li and Yuzhu Wang}, doi = {10.1016/j.engappai.2024.109537}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109537}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A chinese named entity recognition method for landslide geological disasters based on deep learning}, volume = {139}, year = {2025}, }
• (2025). TSD-DETR: A lightweight real-time detection transformer of traffic sign detection for long-range perception of autonomous driving. EAAI, 139, 109536. (www)

  @article{ , abstract = {The key to accurate perception and efficient decision making of autonomous driving is the long-range detection of traffic signs. Long-range detection of traffic signs has the problems of small traffic sign size and complex background. In order to solve these problems, this paper proposes a lightweight model for traffic sign detection based on real-time detection transformer (TSD-DETR). Firstly, the feature extraction module is constructed using multiple types of convolutional modules. The model extracts multi-scale features of different levels to enhance feature extraction ability. Then, small object detection module and detection head are designed to extract and detect shallow features. It can improve the detection of small traffic signs. Finally, Efficient Multi-Scale Attention is introduced to adjust the channel weights. It aggregates the output features of three parallel branches interactively. TSD-DETR achieves a mean average precision (mAp) of 96.8% on Tsinghua-Tencent 100K dataset. It is improved by 2.5% compared with real-time detection transformer. In small object detection, mAp improved by 9%. TSD-DETR achieves 99.4% mAp on the Changsha University of Science and Technology Chinese Traffic Sign Detection Benchmark dataset, with an improvement of 0.6%. The experimental results show that TSD-DETR reduces the number of parameters by 9.06M by optimizing the model structure. On the premise of ensuring the real-time performance of the model, the detection accuracy of the model is improved greatly. The results of ablation experiments show that the feature extraction module and small object detection module proposed in this paper are conducive to improving the detection accuracy.}, archive = {J_EAAI}, author = {Lili Zhang and Kang Yang and Yucheng Han and Jing Li and Wei Wei and Hongxin Tan and Pei Yu and Ke Zhang and Xudong Yang}, doi = {10.1016/j.engappai.2024.109536}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109536}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {TSD-DETR: A lightweight real-time detection transformer of traffic sign detection for long-range perception of autonomous driving}, volume = {139}, year = {2025}, }
• (2025). Dual-path aggregation transformer network for super-resolution with images occlusions and variability. EAAI, 139, 109535. (www)

  @article{ , abstract = {While Transformer-based approaches have recently achieved notable success in super-resolution, their extensive computational requirements impede widespread practical adoption. High-resolution meteorological satellite cloud imagery is essential for weather analysis and forecasting. Enhancing image resolution through super-resolution techniques facilitates the accurate identification and localization of geographic features by meteorological systems. However, current super-resolution methods fail to restore the intricacies of cloud formations and complex regions fully. This research introduces a novel dual-path aggregation Transformer network (DPAT) tailored to enhance the super-resolution of meteorological satellite cloud images. The DPAT network adeptly captures cloud imagery's subtle details and textures, effectively addressing occlusions and the variability inherent in satellite imagery. It bolsters the model's ability to manage the complex attributes of cloud images through the introduction of the Dual-path Aggregation Self-Attention (DASA) mechanism and the Multi-scale Feature Aggregation Block (MFAB), thereby enhancing performance in processing intricate cloud features. The DASA mechanism synthesizes features across spatial, depth, and channel dimensions via a dual-path approach, thoroughly exploiting feature correlations. The MFAB, designed to supplant the multilayer perceptron, incorporates shift convolution and a multi-scale interaction block to augment feature information, compensating for the deficiency in local information absorption due to fixed receptive fields. Experimental outcomes indicate that DPAT delivers superior super-resolution outcomes. With a parameter count of only 32% of the Enhanced Deep Residual Network (EDSR) or 77% of the Image Restoration using Shift Window Transformer (SwinIR), DPAT matches SwinIR's performance on the satellite cloud dataset. Moreover, DPAT balances accuracy and parameter economy across various datasets. This technology is expected to improve image super-resolution capabilities in multiple fields such as human action recognition and industrial recognition, and indirectly improve the accuracy of image perception tasks.}, archive = {J_EAAI}, author = {Qinghui Chen and Lunqian Wang and Zekai Zhang and Xinghua Wang and Weilin Liu and Bo Xia and Hao Ding and Jinglin Zhang and Sen Xu and Xin Wang}, doi = {10.1016/j.engappai.2024.109535}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109535}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dual-path aggregation transformer network for super-resolution with images occlusions and variability}, volume = {139}, year = {2025}, }
• (2025). Chin electromyography-based motor unit decomposition for alternative screening of obstructive sleep apnea events: A comprehensive analysis. EAAI, 139, 109534. (www)

  @article{ , abstract = {Obstructive Sleep Apnea-Hypopnea Syndrome (OSAHS) is a prevalent sleep disorder characterized by recurrent episodes of obstructed breathing due to the relaxation of muscles in the upper airway during sleep, often linked with neuromuscular and cardiovascular disorders. This study introduces a novel method using traditional machine learning classifiers and surface electromyography (SEMG) features extracted from motor units (MUs) decomposed from chin electromyography (EMG) signals to screen for OSA events in OSAHS subjects. SEMG features were extracted from individual MUs decomposed from chin EMG segments using a novel dataset. An apnea detection algorithm was designed to label these events for OSAHS subjects across sleep stages. Analysis of motor neuron firing patterns in OSAHS subjects revealed lower activation during OSA events and higher activation during non-OSA segments. Additionally, we evaluated the proposed system on a publicly available dataset, achieving a maximum accuracy of 72% for OSAHS subjects in the midlife phase age group (40–59 years) and 72.5% for subjects in the severe phase of OSAHS using Support Vector Machines (SVM). The random forest (RF) classifier demonstrated robust performance, achieving 97% accuracy, 93.2% sensitivity, 100% specificity, 100% precision, a 96.48% F1-score, and an area under the curve (AUC) of 0.996. This system facilitates early differentiation between OSA and non-OSA events, enabling timely intervention in the mild apnea phase to prevent progression to severe OSAHS. Moreover, it offers a convenient alternative to conventional polysomnography (PSG), enhancing diagnostic accessibility and clinical management.}, archive = {J_EAAI}, author = {Adil Rehman and Mostafa Moussa and Hani Saleh and Ali Khraibi and Ahsan H. Khandoker}, doi = {10.1016/j.engappai.2024.109534}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109534}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Chin electromyography-based motor unit decomposition for alternative screening of obstructive sleep apnea events: A comprehensive analysis}, volume = {139}, year = {2025}, }
• (2025). An image-to-image adversarial network to generate high resolution wind data over complex terrains from weather predictions. EAAI, 139, 109533. (www)

  @article{ , abstract = {In this work, we propose a Machine Learning method to predict detailed wind fields over extensive, complex terrains. The ability to predict local wind fields is becoming increasingly important for a range of applications, including sports in Nature, large outdoors events, light-aircraft flying, or the management of natural disasters. The intricate nature of wind dynamics, particularly in regions with complex orography such as a mountain range, presents a major challenge to traditional forecasting models. This work presents an efficient way to predict local wind conditions with a high resolution, similar to that of Computational Fluid Dynamics (CFD), in large geographical areas with complex terrain, using the results from relatively coarse (and therefore economical) data from Numerical Weather Prediction (NWP). To achieve this goal, we developed a conditional Generative Adversarial Neural network model (cGAN) to convert NWP data into CFD-like simulations. We apply the method to a rugged region in the Pyrenees mountain range in Spain. The results show that the proposed model outperforms traditional Machine Learning methods, such as Support Vector Machines (SVM), in terms of accuracy and computational efficiency. The method is four orders of magnitude faster than traditional CFD. Mean Average Errors of 1 . 36 m/s for wind speed and 18.73°for wind direction are obtained with the proposed approach.}, archive = {J_EAAI}, author = {Jaime Milla-Val and Carlos Montañés and Norberto Fueyo}, doi = {10.1016/j.engappai.2024.109533}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109533}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {An image-to-image adversarial network to generate high resolution wind data over complex terrains from weather predictions}, volume = {139}, year = {2025}, }
• (2025). Code-switching finetuning: Bridging multilingual pretrained language models for enhanced cross-lingual performance. EAAI, 139, 109532. (www)

  @article{ , abstract = {In recent years, the development of pre-trained models has significantly propelled advancements in natural language processing. However, multilingual sequence-to-sequence pretrained language models (Seq2Seq PLMs) are pretrained on a wide range of languages (e.g., 25 languages), yet often finetuned for specific bilingual tasks (e.g., English–German), leading to domain and task discrepancies between pretraining and finetuning stages, which may lead to sub-optimal downstream performance. In this study, we first illustratively reveal such domain and task discrepancies, and then conduct an in-depth investigation into the side effects that these discrepancies may have on both training dynamic and downstream performance. To alleviate those side effects, we introduce a simple and effective code-switching restoration task (namely code-switching finetuning ) into the standard pretrain-finetune pipeline. Specifically, in the first stage, we recast the downstream data as the self-supervised format used for pretraining, in which the denoising signal is the code-switched cross-lingual phrase. Then, the model is finetuned on downstream task as usual in the second stage. Experiments spanning both natural language generation (12 supervised translations, 30 zero-shot translations, and 2 cross-lingual summarization tasks) and understanding (7 cross-lingual natural language inference tasks) tasks demonstrate that our model consistently and significantly surpasses the standard finetuning strategy. Analyses show that our method introduces negligible computational cost and reduces cross-lingual representation gaps. We have made the code publicly available at: https://github.com/zanchangtong/CSR4mBART .}, archive = {J_EAAI}, author = {Changtong Zan and Liang Ding and Li Shen and Yu Cao and Weifeng Liu}, doi = {10.1016/j.engappai.2024.109532}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109532}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Code-switching finetuning: Bridging multilingual pretrained language models for enhanced cross-lingual performance}, volume = {139}, year = {2025}, }
• (2025). Enhancing particulate matter risk assessment with novel machine learning-driven toxicity threshold prediction. EAAI, 139, 109531. (www)

  @article{ , abstract = {Airborne particulate matter (PM) poses significant health risks, necessitating accurate toxicity threshold determination for effective risk assessment. This study introduces a novel machine-learning (ML) approach to predict PM toxicity thresholds and identify the key physico-chemical and exposure characteristics. Five machine learning algorithms — logistic regression, support vector classifier, decision tree, random forest, and extreme gradient boosting — were employed to develop predictive models using a comprehensive dataset from existing studies. We developed models using the initial dataset and a class weight approach to address data imbalance. For the imbalanced data, the Random Forest classifier outperformed others with 87% accuracy, 81% recall, and the fewest false negatives (23). In the class weight approach, the Support Vector Classifier minimized false negatives (21), while the Random Forest model achieved superior overall performance with 86% accuracy, 80% recall, and an F1-score of 82%. Furthermore, eXplainable Artificial Intelligence (XAI) techniques, specifically SHAP (SHapley Additive exPlanations) values, were utilized to quantify feature contributions to predictions, offering insights beyond traditional laboratory approaches. This study represents the first application of machine learning for predicting PM toxicity thresholds, providing a robust tool for health risk assessment. The proposed methodology offers a time- and cost-effective alternative to classical laboratory tests, potentially revolutionizing PM toxicity threshold determination in scientific and epidemiological research. This innovative approach has significant implications for shaping regulatory policies and designing targeted interventions to mitigate health risks associated with airborne PM.}, archive = {J_EAAI}, author = {Idriss Jairi and Amelle Rekbi and Sarah Ben-Othman and Slim Hammadi and Ludivine Canivet and Hayfa Zgaya-Biau}, doi = {10.1016/j.engappai.2024.109531}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109531}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhancing particulate matter risk assessment with novel machine learning-driven toxicity threshold prediction}, volume = {139}, year = {2025}, }
• (2025). Time-lagged relation graph neural network for multivariate time series forecasting. EAAI, 139, 109530. (www)

  @article{ , abstract = {Recently, Graph Neural Network-based approaches (GNNs) have been widely studied in Multivariate Time Series (MTS) prediction, which could extract information from the closely related variables for prediction. The variables contained in MTS data are lagged correlated, and the future trends of the lagging variables are guided by the leading variables. However, as the existing approaches only focus on delay-free relations, they cannot utilize the guidance information in leading variables to achieve accurate prediction. To address this issue, we propose a novel frame called the Time-Lagged Relation Graph Neural Network (TLGNN) including two key components: the time-lagged relation graph and the time-lagged relation graph learning. The time-lagged relation graph could explicitly model the time-delay relations among MTS variables by connecting variable nodes at lag intervals. The graph learning module could adaptively extract the time-delay relations among MTS variables. Based on the novel designed graph structure, the TLGNN could extract the guidance information from previous values of leading variables to generate more efficient feature representations for prediction. In experiments, the prediction accuracy is significantly improved due to the full exploration of the time-delay relations. Compared with existing methods, the TLGNN achieves the best results in both the single-step prediction and the multi-step prediction tasks.}, archive = {J_EAAI}, author = {Xing Feng and Hongru Li and Yinghua Yang}, doi = {10.1016/j.engappai.2024.109530}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109530}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Time-lagged relation graph neural network for multivariate time series forecasting}, volume = {139}, year = {2025}, }
• (2025). Multi-source information fused loose particle localization and material identification method for sealed electronic equipment. EAAI, 139, 109529. (www)

  @article{ , abstract = {Sealed electronic equipment are1 an important component of aerospace defense systems, and loose particles pose a significant threat to their reliable operation. Loose particle detection is crucial. For sealed electronic equipment with large scale and complex structure, loose particle detection should not only include the judgment of existence, but also obtain location and material information to facilitate the cleaning and control work. In this paper, the authors proposed a multi-source information fused loose particle localization and material identification method. Firstly, the equipment model was designed, the loose particle samples were made, and loose particle signals were collected. Secondly, the two-stage adaptive energy threshold pulse extraction algorithm was newly proposed to extract effective pulses, and the threshold-judgement-search pulse matching algorithm was improved to match the effective pulse groups. Next, spectrograms were transformed from effective pulses to create the localization and material image set. The time-domain, frequency-domain and gray-level co-occurrence matrix features were used to construct the joint feature library. Then, the channel-weighting feature selection method was used to create the localization and material data set. Finally, PReLU-VGG19-Plus was trained on the localization and material image set to obtain the optimal localization and material neural network, while parameter-optimized XGBoost was trained on the localization and material data set to obtain the optimal localization and material classifier. On this basis, combined with the triple majority voting process, the combined localization and material framework were constructed. Extensive test results show that, the location-identification achieved by combined localization framework and the material-identification accuracy achieved by combined material framework are all 100%. The feasibility, stability, and superiority of the method proposed in this paper have been fully verified. It is an important supplement to the existing loose particle detection research, providing important reference for signal detection and classification research in similar fields, and effectively improving the reliability of sealed electronic equipment.}, archive = {J_EAAI}, author = {Zhigang Sun and Qi Liang and Guofu Zhai and Guotao Wang and Min Zhang and Jingting Sun}, doi = {10.1016/j.engappai.2024.109529}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109529}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-source information fused loose particle localization and material identification method for sealed electronic equipment}, volume = {139}, year = {2025}, }
• (2025). A novel multi-criteria decision making method to evaluate green innovation ecosystem resilience. EAAI, 139, 109528. (www)

  @article{ , abstract = {Green innovation ecosystem resilience evaluation is crucial for ensuring sustainable development in dynamic environments. By comprehensively assessing resilience, it helps identify risks and facilitate effective responses, thus promoting sustainable development. To effectively evaluate the resilience of green innovation ecosystems, this paper proposes a model based on the interval-valued intuitionistic fuzzy multi-criteria decision-making (MCDM) method, designed to support decision-makers (DMs) in resilience evaluation. In this model, we improve the Criteria Importance Through Intercriteria Correlation (CRITIC) method by introducing Spearman's correlation coefficient to more accurately determine the objective weights of various indicators based on their interrelationships. Furthermore, we propose an Interval-Valued Intuitionistic Fuzzy Hybrid Average Aggregation (IVIFHAA) operator, which considers both the subjective and objective weight order for aggregating expert evaluation results. Additionally, a scoring function is introduced to generate ranked evaluation results. We also establish a resilience evaluation criteria system covering interconnectivity, sustainability, and diversity within green innovation ecosystems to assist DMs in resilience evaluation. Finally, to validate this method, we apply it to assess the resilience of regional green innovation ecosystems and compare it with other methods. The results demonstrate that the improved operator effectively reduces information loss and improves evaluation efficiency, while the improved CRITIC method handles outliers during the evaluation process.}, archive = {J_EAAI}, author = {Jiafu Su and Hongyu Liu and Yijun Chen and Na Zhang and Junjun Li}, doi = {10.1016/j.engappai.2024.109528}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109528}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel multi-criteria decision making method to evaluate green innovation ecosystem resilience}, volume = {139}, year = {2025}, }
• (2025). Ensemble methods with feature selection and data balancing for improved code smells classification performance. EAAI, 139, 109527. (www)

  @article{ , abstract = {Code smells are software flaws that make it challenging to comprehend, develop, and maintain the software. Identifying and removing code smells is crucial for software quality. This study examines the effectiveness of several machine-learning models before and after applying feature selection and data balancing on code smell datasets. Extreme Gradient Boosting, Gradient Boosting, Adaptive Boosting, Random Forest, Artificial Neural Network (ANN), and Ensemble model of Bagging, and the two best-performing Boosting techniques are used to predict code smell. This study proposes an enhanced approach, which is an ensemble model of the Bagging and Boosting classifier (EMBBC) that incorporates feature selection and data balancing techniques to predict code smells. Four publicly available code smell datasets, Blob Class, Data Class, Long Parameter List, and Switch Statement, were considered for the experimental work. Classes of datasets are balanced using the Synthetic Minority Over-Sampling Technique (SMOTE). A feature selection method called Recursive Feature Elimination with Cross-Validation (RFECV) is used. This study shows that the ensemble model of Bagging and the two best-performing Boosting techniques performs better in Blob Class, Data Class, and Long Parameter List datasets with the highest accuracy of 99.21%, 99.21%, and 97.62%, respectively. In the Switch Statement dataset, the ANN model provides a higher accuracy of 92.86%. Since the proposed model uses only seven features and still provides better results than others, it could be helpful to detect code smells for software engineers and practitioners in less computational time, improving the system's overall performance.}, archive = {J_EAAI}, author = {Pravin Singh Yadav and Rajwant Singh Rao and Alok Mishra and Manjari Gupta}, doi = {10.1016/j.engappai.2024.109527}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109527}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Ensemble methods with feature selection and data balancing for improved code smells classification performance}, volume = {139}, year = {2025}, }
• (2025). Attention-based hand pose estimation with voting and dual modalities. EAAI, 139, 109526. (www)

  @article{ , abstract = {Hand pose estimation has recently emerged as a compelling topic in the robotic research community, because of its usefulness in learning from human demonstration or safe human–robot interaction. Although deep learning-based methods have been introduced for this task and have shown promise, it remains a challenging problem. To address this, we propose a novel end-to-end architecture for hand pose estimation using red-green-blue (RGB) and depth (D) data (RGB-D). Our approach processes the two data sources separately and utilizes a dense fusion network with an attention module to extract discriminative features. The features extracted include both spatial information and geometric constraints, which are fused to vote for the hand pose. We demonstrate that our voting mechanism in conjunction with the attention mechanism is particularly useful for solving the problem, especially when hands are heavily occluded by objects or are self-occluded. Our experimental results on benchmark datasets demonstrate that our approach outperforms state-of-the-art methods by a significant margin.}, archive = {J_EAAI}, author = {Dinh-Cuong Hoang and Anh-Nhat Nguyen and Thu-Uyen Nguyen and Ngoc-Anh Hoang and Van-Duc Vu and Duy-Quang Vu and Phuc-Quan Ngo and Khanh-Toan Phan and Duc-Thanh Tran and Van-Thiep Nguyen and Quang-Tri Duong and Ngoc-Trung Ho and Cong-Trinh Tran and Van-Hiep Duong and Anh-Truong Mai}, doi = {10.1016/j.engappai.2024.109526}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109526}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Attention-based hand pose estimation with voting and dual modalities}, volume = {139}, year = {2025}, }
• (2025). Fault diagnosis of driving gear in battery swapping system based on auditory bionics. EAAI, 139, 109525. (www)

  @article{ , abstract = {Rack and pinion drives (RPD) are widely used in battery swapping system (BSS) for electric heavy trucks (EHT), and due to the continuous heavy-load and high-intensity operation, along with the electric erosion, the gears in the RPD are always damaged, which causes unexpected consequences such as downtime or safety incidents. The working conditions of the RPD in BSS include uncertain noises, fluctuant and low speed, which pose steep challenges to accurate fault diagnosis. Considering the auditory resistance of interference, the low-frequency sensitivity of auditory perception, and the auditory saliency mechanism, to leverage the advantages of auditory perceptual mechanism in addressing the above challenges, as the contribution in artificial intelligence, we propose an entire vibration signal processing scheme based on auditory bionics, including some mathematical models for auditory mechanisms. For the application in engineering, the proposed scheme is employed for fault diagnosis of RPD in BSS in unique working conditions. First, adaptive resampling is used to smooth the speed fluctuation, then, Gammatone filters are employed to transform vibration signals to cochleograms, after that, based on auditory stream segregation and selective attention mechanisms, effective frequency channels and salient features are extracted from the cochleograms, besides, to improve the diagnosis accuracy, binaural features are also extracted, finally, based on (sectional) sparse representation and fusion, fault diagnosis is achieved. The effectiveness of the fault diagnosis scheme is demonstrated using a BSS prototype system.}, archive = {J_EAAI}, author = {Hang Yuan and Hao Wu and Jiacheng Li and Kai Zhang and Huijuan Zhang and Xiaowen You and Xianglong You}, doi = {10.1016/j.engappai.2024.109525}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109525}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Fault diagnosis of driving gear in battery swapping system based on auditory bionics}, volume = {139}, year = {2025}, }
• (2025). Adaptive multimodal control of trans-media vehicle based on deep reinforcement learning. EAAI, 139, 109524. (www)

  @article{ , abstract = {To solve the problem that the control system is prone to instability due to the sudden change of physical characteristics, strong interference, and nonlinear in the process of multimodal movement of trans-media vehicle, an adaptive control method combining the Deep Deterministic Policy Gradient (DDPG) and traditional Proportional-Integral-Derivative (PID) controller is proposed in this paper. In this approach, the upper-level DDPG controller continuously monitors the vehicle's state and environmental conditions, dynamically adjusting the PID parameters in real-time. The lower-level PID controller then utilizes these updated parameters to modulate the output thrust of the vehicle's motors, thereby achieving excellent control over the vehicle's entire movement. Firstly, according to the hydrodynamic analysis, the kinematics and dynamics mathematical model of the self-designed trans-media vehicle is constructed. This model includes the multi-stage motion modal process of aerial flight, underwater navigation, and cross-media motion, which is suitable for the simulation and verification of the control method. Then, an adaptive controller called RL-PID combining DDPG and PID is built, so that PID can adjust parameters in real-time according to the changes in the external environment. Finally, after theoretical stability proof, a comparison study is performed across three approaches, namely the novel RL-PID, Fuzzy PID, and PID. The experimental results illustrate the superiority of the proposed approach over the competing ones and the generalization of the proposed approach under different interference.}, archive = {J_EAAI}, author = {Jingkang Wang and Shuang Liang and Mingming Guo and Heng Wang and Hua Zhang}, doi = {10.1016/j.engappai.2024.109524}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109524}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive multimodal control of trans-media vehicle based on deep reinforcement learning}, volume = {139}, year = {2025}, }
• (2025). A pre-trained multi-step prediction informer for ship motion prediction with a mechanism-data dual-driven framework. EAAI, 139, 109523. (www)

  @article{ , abstract = {The advancement of autonomous maritime surface ships has increased the need for accurate and rapid multi-step prediction of ship motion for decision-making, motion planning, and real-time control tasks. This paper proposes a multi-step prediction method based on Informer with a pre-trained strategy to achieve accurate and fast motion prediction for ships, which substitutes generative inference for rolling prediction to avoid the cumulative error caused by the increasing time horizon. Due to the difference in temporal features from long-term control actions and short-term state sequences, heterogeneous inputs of encoder and decoder are designed to respectively capture their information without information redundancy. To address the bottleneck between the high cost of real data acquisition and the high demand for deep learning methods for data, we propose a mechanism-data dual-driven framework. This framework utilizes a prior mechanism model to generate virtual data incorporating a range of excitation signals designed in accordance with the results of free-running model tests. To reduce the need for real data and increase interpretability, the improved Informer is pre-trained by virtual data from the mechanism model before being trained by real data. Our experiments for multi-step ship motion prediction demonstrate that the proposed method respectively reduces the error and time to 41.36% and 13.20% on average compared to state-of-the-art and classical methods.}, archive = {J_EAAI}, author = {Wenhe Shen and Xinjue Hu and Jialun Liu and Shijie Li and Hongdong Wang}, doi = {10.1016/j.engappai.2024.109523}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109523}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A pre-trained multi-step prediction informer for ship motion prediction with a mechanism-data dual-driven framework}, volume = {139}, year = {2025}, }
• (2025). Optimizing and predicting swarming collective motion performance for coverage problems solving: A simulation-optimization approach. EAAI, 139, 109522. (www)

  @article{ , abstract = {Algorithms using swarming collective motion can solve coverage problems in unknown environments by reacting to unknown obstacles in real-time when they are encountered. However, these algorithms face two key challenges when deployed on real robots. First, hand-tuning efficient collective motion parameters is both time-consuming and difficult. Second, predicting the time required for a swarm to solve a particular problem is not straightforward. This paper introduces a novel evolutionary framework to address both problems by proposing a methodology that autonomously tunes collective motion parameters for coverage problems while predicting the time required for real robots to complete the task. Our approach utilizes a simulation–optimization framework that employs a genetic algorithm to optimize the parameters of a frontier-led swarming algorithm. Results indicate that the optimized parameters are transferable to real robots, achieving 100% coverage while maintaining 84% connectivity between them. Compared to state-of-the-art swarm methods, our system reduced turnaround time by 50% and 57% in different environments while maintaining collective motion. It also achieved a 55% reduction in turnaround time on average across five scenarios compared to budget-constrained path planning, with a 10% increase in coverage. Furthermore, our framework outperformed both hand-tuned and learned collective motion approaches, reducing turnaround time by 73% in non-collective motion scenarios and by 63% while maintaining 85% connectivity in collective motion scenarios. This approach effectively combines the adaptability of swarm behavior with the predictive reliability of planning methods.}, archive = {J_EAAI}, author = {Reda Ghanem and Ismail M. Ali and Shadi Abpeikar and Kathryn Kasmarik and Matthew Garratt}, doi = {10.1016/j.engappai.2024.109522}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109522}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimizing and predicting swarming collective motion performance for coverage problems solving: A simulation-optimization approach}, volume = {139}, year = {2025}, }
• (2025). Color-aware fusion of nighttime infrared and visible images. EAAI, 139, 109521. (www)

  @article{ , abstract = {Pixel-level fusion of visible and infrared images has demonstrated promise in enhancing information representation. However, nighttime image fusion remains challenging due to low and uneven lighting. Existing fusion methods neglect the preservation of color-related information at night, resulting in unsatisfactory outcomes with insufficient brightness. This paper presents a novel color image fusion framework to prevent color distortion, thus generating results more aligned with human perception. Firstly, we design an image fusion network to retain color information from visible images under low-light conditions. Secondly, we incorporate mature low-light enhancement technology into the network as a flexible component to produce fusion results under normal illumination. The training process is carefully designed to address potential issues of overexposure or noise amplification. Finally, we utilize knowledge distillation to create a lightweight end-to-end network that directly generates fusion results under normal lighting conditions from pairs of low-light images. Experimental results demonstrate that our proposed framework outperforms existing methods in nighttime scenarios.}, archive = {J_EAAI}, author = {Jiaxin Yao and Yongqiang Zhao and Yuanyang Bu and Seong G. Kong and Xun Zhang}, doi = {10.1016/j.engappai.2024.109521}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109521}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Color-aware fusion of nighttime infrared and visible images}, volume = {139}, year = {2025}, }
• (2025). Denoising diffusion probabilistic model-enabled data augmentation method for intelligent machine fault diagnosis. EAAI, 139, 109520. (www)

  @article{ , abstract = {Bearing fault is one of the main causes of rotating machinery failure. However, collecting sufficient failure data proves challenging in real-world industrial environments due to their complexity. Owing to this constraint, the majority of current methods fail to accurately identify fault types with limited data, thereby impeding timely maintenance efforts. To address this issue, we propose a bearing fault diagnosis method utilizing diffusion model data enhancement in this study. Following data collection, we employ the continuous wavelet transform to convert various one-dimensional vibration data into two-dimensional time series graphs. Subsequently, these feature graphs are partitioned into sets of fault samples. Then, the diffusion model is employed to augment the training samples. These augmented samples are then fed into the convolutional neural network for fault diagnosis, and their diagnostic accuracy is compared with that of the original dataset. The comparative analysis demonstrates that the data augmentation technique, founded on the diffusion model, enhances fault diagnosis accuracy within a small sample training dataset. Ultimately, the efficacy of the proposed method is validated utilizing the Paderborn bearing dataset and juxtaposed against alternative data augmentation techniques. The findings indicate that the diffusion model-based data augmentation method outperforms other techniques in terms of both training accuracy and stability, particularly in scenarios with small sample.}, archive = {J_EAAI}, author = {Pengcheng Zhao and Wei Zhang and Xiaoshan Cao and Xiang Li}, doi = {10.1016/j.engappai.2024.109520}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109520}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Denoising diffusion probabilistic model-enabled data augmentation method for intelligent machine fault diagnosis}, volume = {139}, year = {2025}, }
• (2025). A novel ensemble method based on residual convolutional neural network with attention module for transient stability assessment considering operational variability. EAAI, 139, 109519. (www)

  @article{ , abstract = {Data-driven methods have been extensively applied in the field of power system transient stability assessment (TSA) owing to their robust capabilities to excavate valuable features. However, TSA methods still face significant challenges in predictive accuracy and generalization ability under variable operation conditions with fluctuating loads or power generations. To address this, a data-driven ensemble TSA method which integrates convolutional block attention module (CBAM) with residual network (ResNet) is proposed to enhance the prediction accuracy. Meanwhile, the traditional cross entropy loss function is replaced by the focal loss function, aiming to reduce the misclassification of unstable samples. Moreover, a rapid updating strategy integrating active learning and fine turning techniques is suggested. It can renew the classifier quickly with limited labeled samples and less time when the network topology changes substantially and makes the pre-trained TSA model unavailable, thus ensuring optimal performance on the new topology. Finally, case studies conducted on the New England 10-machine 39-bus system and the Western Electricity Coordinating Council (WECC) 29-machine 179-bus system validate the effectiveness and robustness of the proposed TSA method. The accuracy of the proposed TSA method achieves 99.56% on 10-machine system and 99.47% on 29-machine system separately, demonstrating the superiority of the proposed TSA method.}, archive = {J_EAAI}, author = {Wensheng Liu and Song Han and Na Rong}, doi = {10.1016/j.engappai.2024.109519}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109519}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel ensemble method based on residual convolutional neural network with attention module for transient stability assessment considering operational variability}, volume = {139}, year = {2025}, }
• (2025). MLTPED-BFC: Machine learning-based trust prediction for edge devices in the blockchain enabled fog computing environment. EAAI, 139, 109518. (www)

  @article{ , abstract = {The utilization of edge devices in fog computing services is increasing every day to achieve effective communication between edge devices as it reduces the latency and processing time. When the number of edge devices increases and operate in various applications, it is seen an increase in malfunctioning of devices due to compromises in security aspects. An increase in the number of un-trustworthy activities leads to loosing of end users to any service provider. So all edge devices must be labeled as trustworthy or not, based on their previous transactions, leading to effective communications. Finding and maintaining the trust score of edge devices is the most pressing concern in the distributed communication environment. Considering all the issues, this paper propose a Machine Learning-based Trust Prediction for Edge Devices in the Blockchain enabled Fog Computing Environment (MLTPED-BFC). The proposed scheme uses an ensemble of Support Vector Regression (SVR) and Multivariable Logistic Regression (MLR) for predicting the trust score of each edge device and updates it after every successful communication. The prediction and updating of the trust score is carried out by the fog server without any biasing. This Artificial Intelligence driven approach enhances communication effectiveness and security by classifying devices as trustworthy or not, improving the overall reliability of the distributed system. The proposed scheme is proved to be secured based on informal security analysis. Extensive simulations are carried out to validate the proposed scheme's effectiveness and compare it with existing schemes. The proposed MLTPED-BFC mechanism have attained 98.91% of accuracy, 0.0048 loss rate, 98.92% of precision, 98.32% of recall, 98.96% of F-Measure and took 356 s for 100 iterations.}, archive = {J_EAAI}, author = {Naveen Chandra Gowda and A. Bharathi Malakreddy and Y. Vishwanath and K.R. Radhika}, doi = {10.1016/j.engappai.2024.109518}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109518}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {MLTPED-BFC: Machine learning-based trust prediction for edge devices in the blockchain enabled fog computing environment}, volume = {139}, year = {2025}, }
• (2025). Weak saliency ensemble network for person re-identification using infrared light images. EAAI, 139, 109517. (www)

  @article{ , abstract = {In recent years, person re-identification (re-id) has primarily been studied using visible light (VL) images. However, the challenges of employing VL images in nighttime environments have prompted research into using infrared light (IR) images. Yet, the utilization of both VL and IR images in person re-id has resulted in increased computational cost and processing time in multi-modality systems, leading to studies focusing solely on IR images. Nevertheless, IR images, lacking color and texture information, generally yield lower recognition performance in existing person re-id studies. In addition, previous studies have shown that person re-id performance suffers in the presence of complex background noise. To tackle these challenges, this study proposes a new weak saliency ensemble network (WSE-Net) for person re-id using IR images. WSE-Net incorporates a channel reduction of feature (CRF) method to reduce computational cost in the ensemble network, a technique for converting input images into group of patch images and feeding them into the ensemble model to enhance the reduced feature information, and a grouped convolution ensemble network (GCE-Net) that enables the fusion of features extracted from original and attention-guided ensemble models. The performance of person re-id using WSE-Net was evaluated on the Dongguk body-based person recognition database version 1 (DBPerson-Recog-DB1) and the Sun Yat-sen university multiple modality re-identification version 1 (SYSU-MM01). Experimental results demonstrated that on DBPerson-Recog-DB1, WSE-Net achieved 93.65% in rank 1, 95.28% in mean average precision (mAP), and 93.52% in the harmonic mean of precision and recall. Additionally, on SYSU-MM01, WSE-Net achieved 86.85% in rank 1, 44.58% in mAP, and 40.06% in the harmonic mean of precision and recall. Furthermore, the accuracy of WSE-Net on both datasets surpassed that of state-of-the-art methods.}, archive = {J_EAAI}, author = {Min Su Jeong and Seong In Jeong and Dong Chan Lee and Seung Yong Jung and Kang Ryoung Park}, doi = {10.1016/j.engappai.2024.109517}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109517}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Weak saliency ensemble network for person re-identification using infrared light images}, volume = {139}, year = {2025}, }
• (2025). Unsupervised domain adaptation for drive-by condition monitoring of multiple railway tracks. EAAI, 139, 109516. (www)

  @article{ , abstract = {Monitoring railway tracks through drive-by vibration data collected by in-service trains offers a cost-effective and adaptable solution for inspecting multiple railway lines. However, numerous existing drive-by monitoring methods rely on supervised learning models, necessitating extensive labelled data for each line. In this paper, a novel framework is proposed based on Unsupervised Domain Adaptation (UDA) concept which facilitates the transfer of a geometric defects diagnosis model learned from one line to a new line without the need for any labelled data from the new line. The proposed framework learns the dynamic-based features that are sensitive to damage and also invariant to different railway tracks. It comprises three components: data pre-processing, UDA implementation, and damage diagnosis. The framework uses the data from the source domain, including corresponding labels, as well as the unlabelled data from the target domain as input. The outputs of the framework consist of the predicted labels for the target domain. The performance of the proposed framework is evaluated using a comprehensive dataset of field measurements of a high-speed train passing 4 different lines within the French high-speed rail network. The proposed UDA framework is implemented using four common UDA algorithms including Information-Theoretical Learning (ITL), Geodesic Flow Kernel (GFK), Transfer Component Analysis (TCA), and Subspace Alignment (SA). The results show that the proposed framework has a 14% increase in the anomaly detection accuracy compared to traditional unsupervised learning methods in which UDA is not used. Furthermore, this study investigates the impact of incorporating a percentage of target data labels during training (semi-supervised domain adaptation), along with various sensor layouts and different tuning parameters, on the accuracy of the proposed approach. The results show that the proposed framework can significantly facilitate the monitoring of railway track conditions using the data collected by in-service trains which could be great interest of railway owners.}, archive = {J_EAAI}, author = {Ramin Ghiasi and Nicolas Lestoille and Cassandre Diaine and Abdollah Malekjafarian}, doi = {10.1016/j.engappai.2024.109516}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109516}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Unsupervised domain adaptation for drive-by condition monitoring of multiple railway tracks}, volume = {139}, year = {2025}, }
• (2025). Hybrid neural network-based fractional-order sliding mode controller for tracking control problem of reconfigurable robot manipulators using fast terminal type switching law. EAAI, 139, 109515. (www)

  @article{ , abstract = {A hybrid neural network-based fractional-order sliding mode controller for the position/force tracking control problem of a reconfigurable robot manipulator system is presented in this work. Due to interchangeable link modules, modeling uncertainties, coupled interconnected states, etc., the control of reconfigurable robot manipulators is very complicated in variable circumstances. So, to handle these dynamical systems, initially, a stable fractional-order sliding manifold is introduced to facilitate accurate and faster system state responses. Subsequently, a neural network-based fractional-order fast terminal sliding mode controller is designed to manage the consequences of external disruptions and parametric uncertainties effectively. In the controller’s design, a hybrid combination of radial basis function neural network and adaptive compensator with fast terminal type switching law is opted for robust performance of the dynamical system. The novelty of the work lies in the combination of the hybrid intelligent sliding mode control scheme with fractional calculus for tracking control problems of reconfigurable robot manipulator systems. The proposed scheme improves the transient response of the controller with a fast terminal-type switching law, and addresses the robustness, fixed-time convergence of system states along with an explicit assessment of the settling time. Finally, the asymptotic stability of the closed-loop dynamical system is validated through Lyapunov’s stability criteria and Fractional-order Barbalat’s lemma, and the simulation results along with a comparative study with some quantitative statistical evaluations confirm the contribution of the presented work.}, archive = {J_EAAI}, author = {Km Shelly Chaudhary and Naveen Kumar}, doi = {10.1016/j.engappai.2024.109515}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109515}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Hybrid neural network-based fractional-order sliding mode controller for tracking control problem of reconfigurable robot manipulators using fast terminal type switching law}, volume = {139}, year = {2025}, }
• (2025). A new spatiotemporal long-term prediction method for continuous annealing processes. EAAI, 139, 109514. (www)

  @article{ , abstract = {Accurately and consistently predicting the heating state is essential to maintain stable operation in continuous annealing processes (CAPs). However, long-term prediction biases often arise due to unmodeled dynamics associated with high-dimensional, time-varying, and strongly coupled variables. This study introduces a spatiotemporal-based forecast model designed to extend the prediction horizon while significantly reducing bias accumulation. The model leverages the spatial characteristics derived from classified process parameters by analyzing the internal structure and dynamics of the process. Additionally, it captures the temporal features of each parameter type through deep learning techniques that preserve and learn from historical data, enabling the model to account for the autocorrelation of multiple variables, including the output, and their correlation with the output. We conducted experiments with real process data, confirming the model’s accuracy and consistency in real-world settings. Additionally, ablation experiments validated the need to integrate both temporal and spatial features for long-term prediction accuracy. Compared to existing methods, the proposed model significantly reduces prediction bias and enhances forecast robustness.}, archive = {J_EAAI}, author = {Wenshuo Song and Weihua Cao and Yan Yuan and Kang-Zhi Liu}, doi = {10.1016/j.engappai.2024.109514}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109514}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A new spatiotemporal long-term prediction method for continuous annealing processes}, volume = {139}, year = {2025}, }
• (2025). Reference-based image super-resolution of hyperspectral and red-green-blue image for determination of wheat kernel quality using deep learning networks. EAAI, 139, 109513. (www)

  @article{ , abstract = {In the process of cultivation and harvest, wheat kernel quality is highly susceptible to various factors, such as disease, mildew, atrophy and impurities, and detection of kernel quality is essential to avoid hazard proliferation, facilitate product grading, and ensure food safety. Possessing abundant image and spectral characteristics, hyperspectral imaging (HSI) has gained impressive achievements in kernel quality analysis, but its low spatial resolution limits its detection accuracy. In this study, reference-based image super-resolution (RefSR) of HSI and Red-Green-Blue image was adopted to improve resolution to determine wheat kernel quality using deep learning networks. Firstly, RefSR was conducted by the improved transformer network with dual-branch feature extraction and weighted fusion operation and achieved excellent RefSR with significant resolution improvement, peak signal to noise ratio of 35.521 and structural similarity index of 0.97, outweighing the existing state-of-the-art networks. Then, the reflectance images (RIs) of effective wavelengths (EWs) from generated HSI images were combined with the residual network with a spatial, channel attention and multi-scale residual to determine wheat kernel quality. Precise analysis was achieved with the accuracy in calibration, validation and prediction sets of 100.00%, 95.26% and 92.78%. RefSR provides a novel and efficient approach for obtaining HSI images of high spatial resolution and facilitates the application of HSI in analysis of crop kernels. RIs of several sporadic EWs can be easily acquired and processed, achieving field and rapid kernel detection. Therefore, the proposed method furnishes the efficient, accurate and applicable determination of wheat kernel quality.}, archive = {J_EAAI}, author = {Shizhuang Weng and Qiaoqiao Zhang and Kaixuan Han and Meijing Pan and Yujian Tan and Qun Chen and Feihong Wu and Cong Wang and Ling Zheng and Yu Lei and Wen Sha}, doi = {10.1016/j.engappai.2024.109513}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109513}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reference-based image super-resolution of hyperspectral and red-green-blue image for determination of wheat kernel quality using deep learning networks}, volume = {139}, year = {2025}, }
• (2025). A robust accent classification system based on variational mode decomposition. EAAI, 139, 109512. (www)

  @article{ , abstract = {State-of-the-art automatic speech recognition models often struggle to capture nuanced features inherent in accented speech, leading to sub-optimal performance in speaker recognition based on regional accents. Despite substantial progress in the field of automatic speech recognition, ensuring robustness to accents and generalization across dialects remains a persistent challenge, particularly in real-time settings. In response, this study introduces a novel approach leveraging Variational Mode Decomposition (VMD) to enhance accented speech signals, aiming to mitigate noise interference and improve generalization on unseen accented speech datasets. Our method employs decomposed modes of the VMD algorithm for signal reconstruction, followed by feature extraction using Mel-Frequency Cepstral Coefficients (MFCC). These features are subsequently classified using machine learning models such as 1D Convolutional Neural Network (1D-CNN), Support Vector Machine (SVM), Random Forest, and Decision Trees, as well as a deep learning model based on a 2D Convolutional Neural Network (2D-CNN). Experimental results demonstrate superior performance, with the SVM classifier achieving an accuracy of approximately 87.5% on a standard dataset and 99.3% on the AccentBase dataset. The 2D-CNN model further improves the results in multi-class accent classification tasks. This research contributes to advancing automatic speech recognition robustness and accent-inclusive speaker recognition, addressing critical challenges in real-world applications.}, archive = {J_EAAI}, author = {Darshana Subhash and Jyothish Lal G. and Premjith B. and Vinayakumar Ravi}, doi = {10.1016/j.engappai.2024.109512}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109512}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A robust accent classification system based on variational mode decomposition}, volume = {139}, year = {2025}, }
• (2025). A single-residual partial mutual information (SPMI) approach to learning discrete-time inputs of stable nonlinear dynamic systems. EAAI, 139, 109511. (www)

  @article{ , abstract = {The selection of input variables and their discrete time delays are fundamentally important in developing robust data-driven dynamic models for use in applied engineering settings, such as for controller design or system level calibration/optimisation. This work is not trivial, especially in the case of complex multivariate and non-linear dynamic systems. There are an array of model-free approaches to input selection explored in the literature, including Multivariate Mutual Information (MMI), Gamma Tests (GT), Self-Organising Maps (SOM) and Partial Mutual Information (PMI). Such a filter-based approach has advantages in exploring feature correlations and their associated information content in a data set directly, agnostic to the constraint of any specific model structure or form. This paper investigates and expands upon the application of a PMI-based Input Selection (PMI-IS) methodology for resulting in a modified version of the algorithm. The modifications are: (1) Selection of input Dead Time (DT) using Mutual Information of First and Second Difference terms of input delays with the output. (2) The Number of Delayed Outputs (NDO) is selected based on the PMI incorporating the previously selected time delays; (3) The Number of Delayed Inputs (NDI) is selected based on the PMI incorporating the identified delay times and NDO; (4) The established Dual-residual PMI (DPMI) algorithm for input selection is simplified to a Single-residual PMI (SPMI) algorithm. Three benchmark discrete-time non-linear dynamic systems and one practical demonstration are used in the case study to demonstrate the effectiveness of this learning algorithm for data-driven identification of time delays, in addition to the implementation details of this modified SPMI-IS methodology.}, archive = {J_EAAI}, author = {Zhijia Yang and Jack Prior and Byron Mason and Edward Winward}, doi = {10.1016/j.engappai.2024.109511}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109511}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A single-residual partial mutual information (SPMI) approach to learning discrete-time inputs of stable nonlinear dynamic systems}, volume = {139}, year = {2025}, }
• (2025). Crude oil price forecasting with multivariate selection, machine learning, and a nonlinear combination strategy. EAAI, 139, 109510. (www)

  @article{ , abstract = {Crude oil price forecasting has been one of the research hotspots in the field of energy economics, which plays a crucial role in energy supply and economic development. However, numerous influencing factors bring serious challenges to crude oil price forecasting, and existing research has room for further improvement in terms of an integrated research roadmap that combines impact factor analysis with predictive modelling. This study aims to examine the impact of financial market factors on the crude oil market and to propose a nonlinear combined forecasting framework based on common variables. Four types of daily exogenous financial market variables are introduced: commodity prices, exchange rates, stock market indices, and macroeconomic indicators for ten indicators. First, various variable selection methods generate different variable subsets, providing more diversity and reliability. Next, common variables in the subset of variables are selected as key features for subsequent models. Then, four models predict crude oil prices using common features as inputs and obtain the prediction results for each model. Finally, the nonlinear mechanism of the deep learning technology is introduced to combine above single prediction results. Experimental results reveal that commodity and foreign exchange factors in financial markets are critical determinants of crude oil market volatility over the long term, as observed in experiments conducted on the West Texas Intermediate and Brent oil price datasets. The proposed model demonstrates strong performance regarding average absolute percentage error, recorded at 2.9962% and 2.4314%, respectively, indicating high forecasting accuracy and robustness. This forecasting framework offers an effective methodology for predicting crude oil prices and enhances understanding the crude oil market.}, archive = {J_EAAI}, author = {Yan Xu and Tianli Liu and Qi Fang and Pei Du and Jianzhou Wang}, doi = {10.1016/j.engappai.2024.109510}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109510}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Crude oil price forecasting with multivariate selection, machine learning, and a nonlinear combination strategy}, volume = {139}, year = {2025}, }
• (2025). Multi-view deep embedded clustering: Exploring a new dimension of air pollution. EAAI, 139, 109509. (www)

  @article{ , abstract = {Clustering is essential for uncovering hidden patterns and relationships in complex datasets. Its importance reveals when labeled data is scarce, expensive, time-consuming to obtain. Real-world applications often exhibit heterogeneity due to the diverse nature of the encapsulated data. This heterogeneity poses a significant challenge in data analysis, modeling, and makes traditional clustering methods ineffective. By adopting a hybrid architecture based on two promising techniques, multi-view and deep clustering, our method achieved better results, outperforming several existing methods including K -means, deep embedded clustering, deep clustering network, deep embedded K -means among many others. Multiple experiments conducted across diverse publicly accessible datasets validate the effectiveness of our proposed method based on well established evaluation metrics such as Accuracy and Normalized Mutual Information (NMI). Furthermore, we applied our method on the air pollution data of Luxembourg, a country with sparse sensor coverage. Our method demonstrated promising results, and unveil a new dimension that pave way for future work in air pollution’s level prediction and hotspots detection, crucial steps towards effective pollution reduction strategies.}, archive = {J_EAAI}, author = {Hassan Kassem and Sally El Hajjar and Fahed Abdallah and Hichem Omrani}, doi = {10.1016/j.engappai.2024.109509}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109509}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-view deep embedded clustering: Exploring a new dimension of air pollution}, volume = {139}, year = {2025}, }
• (2025). Multi-view deep reciprocal nonnegative matrix factorization. EAAI, 139, 109508. (www)

  @article{ , abstract = {Multi-view deep matrix factorization has recently gained popularity for extracting high-quality representations from multi-view data to improve the processing performance of multi-view data in pattern recognition, data mining, and machine learning. It explores the hierarchical semantics of data by performing a multi-layer decomposition on representation matrices after decomposing the data into basis and representation matrices but ignoring the basis matrices, which also contain valuable information about data. Extracting high-quality bases during the deep representation learning process can facilitate the learning of high-quality representations for multi-view data. To this end, this paper proposes a novel deep nonnegative matrix factorization architecture, named M ulti-view D eep R eciprocal N onnegative M atrix F actorization ( MDRNMF ), that collaborates with high-quality basis extraction, allowing the deep representation learning and high-quality basis extraction to promote each other. Based on the representations at the top layer, this paper adaptively learns the intrinsic local similarities of data within each view to capture the view-specific information. In addition, to explore the high-order data consistency across views, this paper introduces a Schatten p -norm-based low-rank regularization on the similarity tensor stacked by the view-specific similarity matrices. In this way, the proposed method can effectively explore and leverage the view-specific and consistent information of multi-view data simultaneously. Finally, extensive experiments demonstrate the superiority of the proposed model over several state-of-the-art methods.}, archive = {J_EAAI}, author = {Bo Zhong and Jun-Yun Wu and Jian-Sheng Wu and Weidong Min}, doi = {10.1016/j.engappai.2024.109508}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109508}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-view deep reciprocal nonnegative matrix factorization}, volume = {139}, year = {2025}, }
• (2025). High-order correlation preserved multi-view unsupervised feature selection. EAAI, 139, 109507. (www)

  @article{ , abstract = {Multi-view unsupervised feature selection (MUFS) has attracted considerable attention as an efficient dimensionality reduction technique. Data usually exhibit certain correlations, and in multi-view data there are more complex high-order correlations. However, some MUFS methods neglect to explore the high-order correlations. In addition, existing methods focus only on the high-order correlation between views or between samples. To tackle these shortcomings, this paper proposes a high-order correlation preserved MUFS (HCFS) method, which fully preserves both the high-order correlation between views and between samples. Specifically, HCFS embeds the energy preservation into the self-representation learning for multi-view data, which preserves the global structure while performing feature selection. Meanwhile, HCFS uses the adaptive weighting strategy to fuse the self-representation matrices of each view into a consistent graph, and constructs a hypergraph based on it to maintain the high-order correlation in the consistent information. Furthermore, the high-order correlation between views is preserved by low-rank tensor learning, and the local structure of data is preserved by using the hyper-Laplacian regularization. Extensive experimental results on eight public datasets demonstrate that the proposed method outperforms several existing state-of-the-art methods, which validates the effectiveness of the proposed method.}, archive = {J_EAAI}, author = {Meng Duan and Peng Song and Shixuan Zhou and Yuanbo Cheng and Jinshuai Mu and Wenming Zheng}, doi = {10.1016/j.engappai.2024.109507}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109507}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {High-order correlation preserved multi-view unsupervised feature selection}, volume = {139}, year = {2025}, }
• (2025). Autonomous smart palm tree harvesting with deep learning-enabled date fruit type and maturity stage classification. EAAI, 139, 109506. (www)

  @article{ , abstract = {This work proposes an innovative autonomous system based on intelligent deep-transfer learning for the sustainable harvesting of palm trees. The machine learning-based autonomous robot detects and captures the date fruit bunches on palm trees in the natural farm environment using the lightweight you only look once (YOLO)v8 algorithm. Five different types of fruit bunches are further classified using a deep transfer learning system based on the type (Khalas, Barhi, Sullaj, Meneifi, and Naboot Saif) and the maturity stage of date fruit (Immature, Khalal, Khalal with Rutab, Pre-Tamar, and Tamar) for their efficient, faster, and accurate harvesting. Five deep convolutional neural network (CNN) models, the Alex Krizhevsky network (AlexNet), the visual geometry group (VGG-16), the residual network (ResNet-50), Inception-v3 and Efficient Net, were trained on around 12,000 images at the bunch level for the two classification tasks. The findings of various performed experiments suggested that the VGG-16 network outperforms the compared models with maximum achieved testing accuracies of 98.89% and 98.17% for date type and maturity stage classification, respectively. The obtained testing accuracies of AlexNet, ResNet-50, Efficient Net, and Inception-v3 models are 97.33%, 97.87%, 98.39%, 96.61% 98%, 93%, and 86.5% for both date type/maturity stage predictions. These obtained accuracies are superior than the state-of-the-art legacy models. Autonomous robotic vehicle front and top cameras are used to localize the quick response (QR)-labeled palm trees using canny edge detection and hough transformation, and date bunch detection and capturing using the trained YOLOv8 algorithm. The robotic vehicle transfers all captured images using Firebase after the completion of the farm journey. The developed and integrated front-end user interface (UI) provides ease to farmers for the two classification tasks of the retrieved images, along with the harvesting decision for each image. The use of proposed sustainable smart harvesting robots to classify and analyze date bunches in the natural environment can significantly improve the yield and global supply chain of this fruit.}, archive = {J_EAAI}, author = {Jawad Yousaf and Zainab Abuowda and Shorouk Ramadan and Nour Salam and Eqab Almajali and Taimur Hassan and Abdalla Gad and Mohammad Alkhedher and Mohammed Ghazal}, doi = {10.1016/j.engappai.2024.109506}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109506}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Autonomous smart palm tree harvesting with deep learning-enabled date fruit type and maturity stage classification}, volume = {139}, year = {2025}, }
• (2025). A novel time-delay multivariable grey model and its application in predicting oil production. EAAI, 139, 109505. (www)

  @article{ , abstract = {An accurate prediction of oil production can provide a scientific basis for planning the production of the Qinghai oilfield and help in rationally arranging resources. To address the time-delay of related factors in the oil production system and how this problem affects oil production, this paper classifies the different degrees of time-delay of related factors and establishes a time-delay multivariable grey model with multiple parameters. This model not only reflects the characteristics of a Logistic model with strong historical recurrence ability and high prediction accuracy in the short and medium terms but also compensates for the defects of existing grey models that do not consider the time-delay of related factors; this is a new idea for grey modelling. Moreover, the parameter estimation of the new model is obtained via the least squares technique, the time response of the new model is obtained via a mathematical method, and the modelling steps of the new model are also obtained. Finally, the new model is applied to the prediction of production of the Qinghai oilfield in China, the effectiveness of the model is analysed according to two different correlation sequences, and six types of the same modelling objects are tested. Results of two types of twelve experiments each show that the total mean absolute percentage error is less than 5%, the lowest is 1.2580%, and the highest is only 4.0087%. This shows that the new model has a good effect and results of six technical indicators show that the new model is better than the other five multivariable grey models.}, archive = {J_EAAI}, author = {Huiming Duan and Guan Wang and Yuxin Song and Hongli Chen}, doi = {10.1016/j.engappai.2024.109505}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109505}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel time-delay multivariable grey model and its application in predicting oil production}, volume = {139}, year = {2025}, }
• (2025). A novel integrated prediction method using adaptive mode decomposition, attention mechanism and deep learning for coking products prices. EAAI, 139, 109504. (www)

  @article{ , abstract = {Accurate prediction of coking product prices is crucial for enhancing production efficiency, cost optimization, and profit maximization in smart coking facilities. To address the volatility caused by nonlinear factors such as raw material costs, substitutes, macroeconomic indicators, sudden events, policy changes, and market behaviors, we propose a novel integrated prediction method for coking product price prediction. This method combines Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) for signal decomposition, Bidirectional Encoder Representations from Transformers (BERT) for natural language processing, attention mechanisms (AT) to weigh feature importance, and an ensemble of Bidirectional Gated Recurrent Unit, Bidirectional Long Short-Term Memory, and Gated Recurrent Unit, abbreviated BBG, for robust feature extraction. We design a feature selection strategy to avoid data leakage and improve the predictive ability of the model, and describe a method to maintain textual data information integrity when combining data from different sources. Experimental results on coke and methanol datasets show that our approach retains multi-source text richness improves predictive capability, and outperforms other state-of-the-art methods, providing an effective tool for developing smart coke plants.}, archive = {J_EAAI}, author = {Xuhui Zhu and Chenggong Ma and Hao Lei and Pingfan Xia and Zhanglin Peng}, doi = {10.1016/j.engappai.2024.109504}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109504}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel integrated prediction method using adaptive mode decomposition, attention mechanism and deep learning for coking products prices}, volume = {139}, year = {2025}, }
• (2025). Explainable artificial intelligence of tree-based algorithms for fault detection and diagnosis in grid-connected photovoltaic systems. EAAI, 139, 109503. (www)

  @article{ , abstract = {A grid-connected photovoltaic system integrates solar panels with the utility grid through a power inverter unit, allowing them to operate in parallel with the grid. Commonly known as grid-tied or on-grid solar systems, these configurations enable panels to feed electrical energy back into the grid, offering simplicity, low operating and maintenance costs, and reduced electricity bills. Despite these advantages, this environmentally friendly energy solution is still susceptible to downtimes and faults. This study utilizes advanced machine learning tree-based algorithms for fault detection and diagnosis in such systems with the goal of maintaining reliability, improving performance, and ensuring optimal energy generation. Specifically, the research investigates the effectiveness of Extra Trees as a fault detection and diagnosis algorithm through an efficient two-phase framework that consists of a binary fault detection phase followed by a multi-class fault diagnosis phase, achieving respective accuracies of 99.5% and 98.7%. In addition, the study underscores the importance of oversampling in improving results, particularly for imbalanced datasets. Moreover, explainable artificial intelligence is employed to enhance transparency in the model’s output and sensitivity to specific features in a given order. Remarkably, the findings align directly with results obtained from techniques such as feature importance averaging and incremental feature accuracy tracking. The research unveils a highly scalable, lightweight, and simple framework for fault detection and diagnosis in grid-connected photovoltaic systems.}, archive = {J_EAAI}, author = {Hassan N. Noura and Zaid Allal and Ola Salman and Khaled Chahine}, doi = {10.1016/j.engappai.2024.109503}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109503}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Explainable artificial intelligence of tree-based algorithms for fault detection and diagnosis in grid-connected photovoltaic systems}, volume = {139}, year = {2025}, }
• (2025). Search task extraction using k-contour based recurrent deep graph clustering. EAAI, 139, 109501. (www)

  @article{ , abstract = {Search engines must accurately predict the implicit intent of users to effectively guide their online search experience and assist them in completing their tasks. Users create time-ordered query logs by performing various queries on search engines to access desired information. Search task extraction groups queries with the same intent into unique clusters, whether these queries come from different tasks within the same session or from the same task across different sessions. Accurate identification of user intent improves the performance of search-guiding processes, including query suggestion, personalized search, and advertisement retrieval. Many existing methods focus on creating graphs that show relationships between queries. However, these methods typically cluster the graph using simple threshold-based techniques rather than leveraging graph topological structure features. Recent studies have introduced deep clustering layers to prevent the model size from growing as the number of queries increases. However, these models rely on labeled data and overlook modern embeddings from language models. We propose a novel k-contour-based graph convolutional network connective proximity clustering layer (CoGCN-C-CL) architecture that clusters graphs without requiring labeled data by leveraging graph topological properties. CoGCN-C-CL simultaneously learns query representations and search tasks. The k-contours identify distinct regions of the graph, while the graph convolutional network (GCN) exploits interactions between nodes within these regions. Experimental results demonstrate that CoGCN-C-CL outperforms existing state-of-the-art search task clustering methods on frequently used search task datasets.}, archive = {J_EAAI}, author = {Nurullah Ates and Yusuf Yaslan}, doi = {10.1016/j.engappai.2024.109501}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109501}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Search task extraction using k-contour based recurrent deep graph clustering}, volume = {139}, year = {2025}, }
• (2025). Dynamic model-based intelligent fault diagnosis method for fault detection in a rod fastening rotor. EAAI, 139, 109499. (www)

  @article{ , abstract = {A complete fault sample database is of great significance for the intelligent fault diagnosis method of rod fastening rotor. However, the lack of fault samples makes the fault diagnosis results unbelievable. To solve this issue, the dynamic model-based intelligent fault diagnosis method is established for a rod fastening rotor, and the fault sample database is enriched by numerical simulations. First, the lumped parameter model of the rod fastening rotor system is constructed and further updated using Euclidean Distance between measurement and numerical simulation of the intact system. Second, mathematical models of various fault types are incorporate into the updated model to obtain numerical simulation fault samples. Thirdly, the utilization of numerical simulation fault samples is severed as training data to the artificial intelligence (AI) models and the unknown measurement test samples will be finally classified. In this paper, Support Vector Machine, Random Forest, Bayesian Network and Decision Tree are selected as the typical AI models. Subsequently, the feasibility of classification is validated by the test bench of the rod fastening rotor system, and the problem of insufficient fault samples can be solved.}, archive = {J_EAAI}, author = {Wuhui Xu and Hui Wang and Jiabin Jin and Ronggang Yang and Jiawei Xiang}, doi = {10.1016/j.engappai.2024.109499}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109499}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamic model-based intelligent fault diagnosis method for fault detection in a rod fastening rotor}, volume = {139}, year = {2025}, }
• (2025). Navigation of autonomous mobile robots in dynamic unknown environments based on dueling double deep q networks. EAAI, 139, 109498. (www)

  @article{ , abstract = {This study focuses on applying the algorithm of Dueling Double Deep Q Networks to create a robust and adaptable navigation system for autonomous robots. The main objective of the study is to propose a network model that is capable of making a robot explore unknown areas, avoid static and dynamic obstacles efficiently, recognize predefined targets, and achieve them with high accuracy. Toward this, three different network models have been designed and trained with depth images obtained from a depth camera and directional and distance information from an RGB camera. First, these models were trained and tested in simple and complex simulated environments. The D3QN-C model demonstrated strong performance, achieving a success rate of 89% in the simple environment and of 87% in the complex environment. Tests were further extended by adding real-world data with different obstacle densities in order to prove the strength of the model in increasingly difficult and realistic scenarios. During all tests, the D3QN-C model could sustain high performance, showing 90% success rates in low-density, 85% in medium-density, and 82% in high-density environments. These results are evidence of the efficiency, robustness, and flexibility of this model and underline the potential of the algorithm of the Dueling Double Deep Q Networks as a main tool in using robots within real-world scenarios characterized by dynamics and complexity.}, archive = {J_EAAI}, author = {Koray Ozdemir and Adem Tuncer}, doi = {10.1016/j.engappai.2024.109498}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109498}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Navigation of autonomous mobile robots in dynamic unknown environments based on dueling double deep q networks}, volume = {139}, year = {2025}, }
• (2025). Consistency-guided multi-source-free domain adaptation. EAAI, 139, 109497. (www)

  @article{ , abstract = {Deep neural networks suffer from severe performance degradation when facing a distribution shift between the labeled source domain and unlabeled target domain. Domain adaptation addresses this issue by aligning the feature distributions of both domains. Conventional methods assume that the labeled source samples are drawn from a single data distribution (domain) and can be fully accessed during training. However, in real applications, multiple source domains with different distributions often exist, and source samples may be unavailable due to privacy and storage constraints. To address multi-source and data-free challenges, Multi-Source-Free Domain Adaptation (MSFDA) uses only diverse pre-trained source models without requiring any source data. Most existing MSFDA methods adapt each source model to the target domain individually, making them ineffective in leveraging the complementary transferable knowledge from different source models. In this paper, we propose a novel COnsistency-guided multi-source-free Domain Adaptation (CODA) method, which leverages the label consistency criterion as a bridge to facilitate the cooperation among source models. CODA applies consistency regularization on the soft labels of weakly- and strongly-augmented target samples from each pair of source models, allowing them to supervise each other. To achieve high-quality pseudo-labels, CODA also performs a consistency-based denoising to unify the pseudo-labels from different source models. Finally, CODA optimally combines different source models by maximizing the mutual information of the predictions of the resulting target model. Extensive experiments on four benchmark datasets demonstrate the effectiveness of CODA compared to the state-of-the-art methods.}, archive = {J_EAAI}, author = {Ziyi Liu and Chaoran Cui and Chunyun Zhang and Fan’an Meng and Shuai Gong and Muzhi Xi and Lei Li}, doi = {10.1016/j.engappai.2024.109497}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109497}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Consistency-guided multi-source-free domain adaptation}, volume = {139}, year = {2025}, }
• (2025). Transformer-enabled weakly supervised abnormal event detection in intelligent video surveillance systems. EAAI, 139, 109496. (www)

  @article{ , abstract = {Video Anomaly Detection (VAD) for weakly supervised data operates with limited video-level annotations. It also holds the practical significance to play a pivotal role in surveillance and security applications like public safety, patient monitoring, autonomous vehicles, etc. Moreover, VAD extends its utility to various industrial settings, where it is instrumental in safeguarding workers' safety, enabling real-time production quality monitoring, and predictive maintenance. These diverse applications highlight the versatility of VAD and its potential to transform processes across various industries, making it an essential tool along with traditional surveillance applications. The majority of the existing studies have been focused on mitigating critical aspects of VAD, such as reducing false alarm rates and misdetection. These challenges can be effectively addressed by capturing the intricate spatiotemporal pattern within video data. Therefore, the proposed work named Swin Transformer-based Hybrid Temporal Adaptive Module (ST-HTAM) Abnormal Event Detection introduces an intuitive temporal module along with leveraging the strengths of the Swin (Shifted window-based) Transformers for spatial analysis. The novel aspect of this work lies in the hybridization of global self-attention and Convolutional-Long Short Term Memory (C-LSTM) Networks are renowned for capturing both global and local temporal dependencies. By extracting these spatial and temporal components, the proposed method, ST-HTAM, offers a comprehensive understanding of anomalous events. Altogether, it enhances the accuracy and robustness of Weakly Supervised VAD (WS-VAD). Finally, an anomaly scoring mechanism is employed in the classification step to facilitate effective anomaly detection from test video data. The proposed system is tailored to operate in real-time and highlights the dual focus on sophisticated Artificial Intelligence (AI) techniques and their impactful use cases across diverse domains. Comprehensive experiments are conducted on benchmark datasets that clearly show the substantial superiority of the ST-HTAM over state-of-the-art approaches. Code is available at https://github.com/Shalmiyapaulraj78/STHTAM-VAD .}, archive = {J_EAAI}, author = {Shalmiya Paulraj and Subramaniyaswamy Vairavasundaram}, doi = {10.1016/j.engappai.2024.109496}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109496}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Transformer-enabled weakly supervised abnormal event detection in intelligent video surveillance systems}, volume = {139}, year = {2025}, }
• (2025). A novel policy for coordinating a hurricane monitoring system using a swarm of buoyancy-controlled balloons trading off communication and coverage. EAAI, 139, 109495. (www)

  @article{ , abstract = {This paper introduces a novel architecture for hurricane monitoring aimed at maximizing the collection of critical data to enhance the accuracy of weather predictions. The proposed system deploys a swarm of controllable balloons equipped with meteorological sensors within the hurricane environment. A key challenge in this setup is managing the trade-off between maximizing area coverage for data collection and maintaining robust communication links among the balloons. To address this challenge, we propose a cost function with two conflicting components: one prioritizes area coverage, and the other focuses on repositioning to maintain communication. This cost function is optimized using an adaptive neural network-based model predictive control strategy, which enables the system to dynamically balance these competing requirements in real-time. Quantitative results from extensive simulations demonstrate the versatility and effectiveness of the proposed architecture, showing that it can achieve comprehensive communication connectivity and increased area coverage across various configurations, including different numbers of balloons and operational periods.}, archive = {J_EAAI}, author = {Bruno R.O. Floriano and Benjamin Hanson and Thomas Bewley and João Y. Ishihara and Henrique C. Ferreira}, doi = {10.1016/j.engappai.2024.109495}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109495}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A novel policy for coordinating a hurricane monitoring system using a swarm of buoyancy-controlled balloons trading off communication and coverage}, volume = {139}, year = {2025}, }
• (2025). Optimization of underground mining production layouts considering geological uncertainty using deep reinforcement learning. EAAI, 139, 109493. (www)

  @article{ , abstract = {Mineral extraction plays a key role in the global raw materials supply chain, however the exhaustion of shallow deposits and typical scarcity of sampled data during exploration activities creates challenges in mine planning and design, where decision-making is highly sensitive to uncertainty in geology and mineral grade prediction. Geostatistical techniques are commonly used to generate a set of equiprobable simulated numerical models to capture these uncertainties, however incorporating these simulated models within a mine planning and design framework remains a major challenge. The purpose of this paper is to propose a novel approach to decision-making in underground mine design that can use information from an ensemble of numerical realizations of a mineral resource to improve the financial performance of the asset. A deep reinforcement learning (DRL) framework, using the proximal policy optimization (PPO) algorithm, is developed for the design of underground mining production level layouts. A case study is presented using a gold mineral resource characterized by an ensemble of 100 numerical realizations to verify the advantages of the proposed method, considering a baseline consisting of an industry standard automated design method. The DRL approach achieved an 8.3% higher expected profit, a 3.4% more gold mined than the baseline, and has the added functionality of considering uncertainty in mineral grades.}, archive = {J_EAAI}, author = {Roberto Noriega and Jeff Boisvert}, doi = {10.1016/j.engappai.2024.109493}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109493}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Optimization of underground mining production layouts considering geological uncertainty using deep reinforcement learning}, volume = {139}, year = {2025}, }
• (2025). Learning to generate and evaluate fact-checking explanations with transformers. EAAI, 139, 109492. (www)

  @article{ , abstract = {In an era increasingly dominated by digital platforms, the spread of misinformation poses a significant challenge, highlighting the need for solutions capable of assessing information veracity. Our research contributes to the field of Explainable Artificial Antelligence (XAI) by developing transformer-based fact-checking models that contextualise and justify their decisions by generating human-accessible explanations. Importantly, we also develop models for automatic evaluation of explanations for fact-checking verdicts across different dimensions such as (self)-contradiction , hallucination , convincingness and overall quality . By introducing human-centred evaluation methods and developing specialised datasets, we emphasise the need for aligning Artificial Intelligence (AI)-generated explanations with human judgements. This approach not only advances theoretical knowledge in XAI but also holds practical implications by enhancing the transparency, reliability and users’ trust in AI-driven fact-checking systems. Furthermore, the development of our metric learning models is a first step towards potentially increasing efficiency and reducing reliance on extensive manual assessment. Based on experimental results, our best performing generative model achieved a Recall-Oriented Understudy for Gisting Evaluation-1 ( ROUGE-1 ) score of 47.77 demonstrating superior performance in generating fact-checking explanations, particularly when provided with high-quality evidence. Additionally, the best performing metric learning model showed a moderately strong correlation with human judgements on objective dimensions such as (self)-contradiction and hallucination , achieving a Matthews Correlation Coefficient (MCC) of around 0.7.}, archive = {J_EAAI}, author = {Darius Feher and Abdullah Khered and Hao Zhang and Riza Batista-Navarro and Viktor Schlegel}, doi = {10.1016/j.engappai.2024.109492}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109492}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Learning to generate and evaluate fact-checking explanations with transformers}, volume = {139}, year = {2025}, }
• (2025). Multi-stage guided code generation for large language models. EAAI, 139, 109491. (www)

  @article{ , abstract = {Currently, although Large Language Models (LLMs) have shown significant performance in the field of code generation, their effectiveness in handling complex programming tasks remains limited. This is primarily due to the substantial distance between the problem description and the correct code, making it difficult to ensure accuracy when directly generating code. Human programmers, when faced with a complex programming problem, usually use multiple stages to solve it in order to reduce the difficulty of development. First, they analyze the problem and think about a solution plan, then they design a code architecture based on that plan, and finally they finish writing the detailed code. Based on this, we propose a multi-stage guided code generation strategy that aims to gradually shorten the transformation distance between the problem description and the correct code, thus improving the accuracy of code generation. Specifically, the approach consists of three stages: planning, design and implementation. In the planning phase, the Large Language Model (LLM) generates a solution plan based on the problem description; in the design phase, the code architecture is further designed based on the solution plan; and in the implementation phase, the previous solution plan and code architecture are utilized to guide the LLM in generating the final code. Additionally, we found that existing competition-level code generation benchmarks may overlap with the training data of the Chat Generative Pre-trained Transformer (ChatGPT), posing a risk of data leakage. To validate the above findings and circumvent this risk, we created a competition-level code generation dataset named CodeC, which contains data never used for training ChatGPT. Experimental results show that our method outperforms the most advanced baselines. On the CodeC dataset, our approach achieves a 34.7% relative improvement on the Pass@1 metric compared to the direct generation method of ChatGPT. We have published the relevant dataset at https://github.com/hcode666/MSG for further academic research and validation.}, archive = {J_EAAI}, author = {Yewei Han and Chen Lyu}, doi = {10.1016/j.engappai.2024.109491}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109491}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-stage guided code generation for large language models}, volume = {139}, year = {2025}, }
• (2025). Category knowledge-guided few-shot bearing fault diagnosis. EAAI, 139, 109489. (www)

  @article{ , abstract = {Real-time bearing fault diagnosis plays a vital role in maintaining the safety and reliability of sophisticated industrial systems. However, the scarcity of labeled data in fault diagnosis, due to the difficulty of collecting fault samples and the high cost of labeling, poses a significant challenge in learning discriminative fault features from limited and complex monitoring signals. Few-shot learning (FSL) emerges as a potent method for extracting and accurately classifying features from severe fault signals. Nonetheless, challenges such as data scarcity and environmental noise significantly impede the efficacy of existing FSL methods in diagnosing incipient faults effectively. These limitations are primarily due to the inadequate consideration of inter-class correlations within noisy contexts by current FSL strategies, which restricts their ability to extrapolate familiar features to new classes. Consequently, there is a pressing demand for an FSL approach that can exploit inter-class correlations to address the hurdles of data insufficiency and environmental complexities, thereby facilitating the diagnosis of incipient faults in few-shot settings. This paper proposes a novel category-knowledge-guided model tailored for few-shot multi-task scenarios. By leveraging attribute data from base categories and the similarities across new class samples, our model efficiently establishes mapping relations for unencountered tasks, significantly enhancing its generalization capabilities for early-stage fault diagnosis and multi-task applications. This model ensures swift and precise FSL fault diagnosis under uncharted operational conditions. Comparative analyses utilizing the Case Western Reserve University bearing dataset and the Early Mild Fault Traction Motor bearing dataset demonstrate our model’s superior performance against leading FSL and transfer learning approaches.}, archive = {J_EAAI}, author = {Feng Zhan and Lingkai Hu and Wenkai Huang and Yikai Dong and Hao He and Guanjun Wu}, doi = {10.1016/j.engappai.2024.109489}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109489}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Category knowledge-guided few-shot bearing fault diagnosis}, volume = {139}, year = {2025}, }
• (2025). Diverse policy generation for the flexible job-shop scheduling problem via deep reinforcement learning with a novel graph representation. EAAI, 139, 109488. (www)

  @article{ , abstract = {In scheduling problems common in the industry and various real-world scenarios, responding in real-time to disruptive events is important. Recent methods propose the use of deep reinforcement learning (DRL) to learn policies capable of generating solutions under this constraint. However, current DRL approaches struggle with large instances, which are common in real-world scenarios. The objective of this paper is to introduce a new DRL method for solving the flexible job-shop scheduling problem, with a focus on these type of instances. The approach is based on the use of heterogeneous graph neural networks to a more informative graph representation of the problem. This novel modeling of the problem enhances the policy’s ability to capture state information and improve its decision-making capacity. Additionally, we introduce two novel approaches to enhance the performance of the DRL approach: the first involves generating a diverse set of scheduling policies, while the second combines DRL with dispatching rules (DRs) constraining the action space, with a variable degree of freedom depending on the chosen policy. Experimental results on two public benchmarks show that our approach outperforms DRs and achieves superior results compared to three state-of-the-art DRL methods, particularly for large instances.}, archive = {J_EAAI}, author = {Imanol Echeverria and Maialen Murua and Roberto Santana}, doi = {10.1016/j.engappai.2024.109488}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109488}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Diverse policy generation for the flexible job-shop scheduling problem via deep reinforcement learning with a novel graph representation}, volume = {139}, year = {2025}, }
• (2025). Multi-stream feature aggregation network with multi-scale supervision for single image dehazing. EAAI, 139, 109486. (www)

  @article{ , abstract = {Single image dehazing is a challenge, as it requires to eliminate the haze while preserving image quality. Most existing models use encoder–decoder or single-scale structures. Owing to flaws of architectural design, these models fail to capture spatial details and semantic contexts in a complementary manner, resulting in inaccurately detailed recovery and suboptimal dehazing performance. To address these issues, we propose a novel framework known as the Multi-stream Feature Aggregation Network (MSFANet), which leverages and excavates the information from different scales of input images and fuses different levels of features. In addition, we design a Self-adjustable Complementary Features Selection Module (SCFSM) that efficiently selects and aggregates the features with different scales from multiple streams, enabling the full exchange of information between different levels of features within the network. Moreover, during the training stage, we introduce a Fast Fourier Transform (FFT) loss combined with pixel loss to simultaneously supervise the reconstruction process in the spatial and frequency domains with a multi-scale supervision strategy. Extensive experiments on homogeneous and nonhomogeneous datasets show that the proposed MSFANet achieves state-of-the-art dehazing performance. Specifically, our method dramatically boosts the Peak Signal-to-Noise Ratio (PSNR) metric to 42.47 dB on the commonly used Synthetic Objective Testing Set (SOTS) indoor dataset.}, archive = {J_EAAI}, author = {Junjiang Wu and Haibo Tao and Kai Xiao and Jun Chu and Lu Leng}, doi = {10.1016/j.engappai.2024.109486}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109486}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-stream feature aggregation network with multi-scale supervision for single image dehazing}, volume = {139}, year = {2025}, }
• (2025). A multi-criteria decision-making method based on discrete Z-numbers and aczel-alsina aggregation operators and its application on early diagnosis of depression. EAAI, 139, 109484. (www)

  @article{ , abstract = {In mental health diagnostics, the questionnaire is an effective and cost-effective method. However, the traditional questionnaire test methods for depression and anxiety have great ambiguity. The discrete Z-numbers (DZs) provide solutions for describing and resolving complex fuzzy issues in the intelligent multi-criteria decision-making (MCDM) process. However, large-scale datasets are not suited for the present MCDM techniques due to their extremely high computational cost. Additionally, these techniques are less stable and flexible. To address the above issues, a novel MCDM method is introduced, which is based on the DZs theory and the Aczel-Alsina (AA) aggregation operator (AO) for large-scale datasets. To begin with, centroid points are calculated for DZs, and a series of novel AOs are introduced. And then a score function with a parameter is introduced to balance the influence between the possibility restriction and the fuzzy restriction of DZs. Thirdly, a new MCDM method under DZs is presented based on the proposed AA AOs and score function. Finally, to support the early diagnosis of depression and anxiety, we apply our method to the real-life online Depression, Anxiety, and Stress Scale (DASS) which can be transformed into DZs by our proposed preprocessing method. According to experimental results, our method is applicable to large-scale datasets and has much lower complexity as well as higher flexibility and stability.}, archive = {J_EAAI}, author = {Dong Ren and Xiuqin Ma and Hongwu Qin and Siyue Lei and Xuli Niu}, doi = {10.1016/j.engappai.2024.109484}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109484}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A multi-criteria decision-making method based on discrete Z-numbers and aczel-alsina aggregation operators and its application on early diagnosis of depression}, volume = {139}, year = {2025}, }
• (2025). Evaluating the financial credibility of third-party logistic providers through a novel frank operators-driven group decision-making model with dual hesitant linguistic q-rung orthopair fuzzy information. EAAI, 139, 109483. (www)

  @article{ , abstract = {In the relevant literature, there is no study dealing with the financial credibility of third-party logistic providers with the help of decision-making frames. Further, there are no criteria to evaluate the third-party logistics providers' creditworthiness in practice, and decision-makers in the banks consider their judgments and experiences to assess the demand of the logistics firms. This study proposes a multi-criteria group decision-making framework through a dual hesitant linguistic q -rung orthopair fuzzy (DHL q -ROF) set to manage uncertainties more effectively and make a theoretical contribution to the academic literature. For ranking, the score function and accuracy function are defined. Additionally, some novel operational laws based on Frank t -norms and t -conorms are defined for DHL q -ROF numbers. A wide range of generalized aggregation operators, such as DHL q -ROF Frank weighted averaging, DHL q -ROF Frank weighted geometric, DHL q -ROF Frank generalized weighted averaging, and DHL q -ROF Frank generalized weighted geometric operators, are also investigated. Beyond that, several prominent characteristics of the proposed operators are studied. It is applied to a financial credibility problem for a multinational organization to demonstrate the introduced model's applicability. Considering the results obtained regarding the importance of the criteria, the most crucial criterion is market indebtedness, followed by fleet vehicle structure and current rate criteria, respectively. The results indicate that UPS, Kuhne & Nagel and DHL Deutsche Post are the best third-party logistic providers. The sensitivity analysis shows that the framework possesses favourable flexibility and effectiveness. Thanks to the framework's ability to produce practical solutions to challenging decision-making problems, it can be reliably preferred in engineering and other fields.}, archive = {J_EAAI}, author = {Arun Sarkar and Ömer Faruk Görçün and Fatih Ecer and Tapan Senapati and Hande Küçükönder}, doi = {10.1016/j.engappai.2024.109483}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109483}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Evaluating the financial credibility of third-party logistic providers through a novel frank operators-driven group decision-making model with dual hesitant linguistic q-rung orthopair fuzzy information}, volume = {139}, year = {2025}, }
• (2025). HK-MOEA/D: A historical knowledge-guided resource allocation for decomposition multiobjective optimization. EAAI, 139, 109482. (www)

  @article{ , abstract = {Decomposition-based multiobjective evolutionary algorithms is one of the prevailing algorithmic frameworks for multiobjective optimization. This framework distributes the same amount of evolutionary computing resources to each subproblems, but it ignores the variable contributions of different subproblems to population during the evolution. Resource allocation strategies (RAs) have been proposed to dynamically allocate appropriate evolutionary computational resources to different subproblems, with the aim of addressing this limitation. However, the majority of RA strategies result in inefficiencies and mistakes when performing subproblem assessment, thus generating unsuitable algorithmic results. To address this problem, this paper proposes a decomposition-based multiobjective evolutionary algorithm (HK-MOEA/D). The HK-MOEA/D algorithm uses a historical knowledge-guided RA strategy to evaluate the subproblem’s evolvability, allocate evolutionary computational resources based on the evaluation value, and adaptively select genetic operators based on the evaluation value to either help the subproblem converge or move away from a local optimum. Additionally, the density-first individual selection mechanism of the external archive is utilized to improve the diversity of the algorithm. An external archive update mechanism based on θ -dominance is also used to store solutions that are truly worth keeping to guide the evaluation of subproblem evolvability. The efficacy of the proposed algorithm is evaluated by comparing it with seven state-of-the-art algorithms on three types of benchmark functions and three types of real-world application problems. The experimental results show that HK-MOEA/D accurately evaluates the evolvability of the subproblems and displays reliable performance in a variety of complex Pareto front optimization problems.}, archive = {J_EAAI}, author = {Wei Li and Xiaolong Zeng and Ying Huang and Yiu-ming Cheung}, doi = {10.1016/j.engappai.2024.109482}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109482}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {HK-MOEA/D: A historical knowledge-guided resource allocation for decomposition multiobjective optimization}, volume = {139}, year = {2025}, }
• (2025). A geometric model with stochastic error for abnormal motion detection of portal crane bucket grab. EAAI, 139, 109481. (www)

  @article{ , abstract = {Abnormal swing angle detection of bucket grabs is crucial for efficient harbor operations. In this study, we develop a practically convenient swing angle detection method for crane operation, requiring only a single standard surveillance camera at the fly-jib head, without the need for sophisticated sensors or markers on the payload. Specifically, our algorithm takes the video images from the camera as input. Next, a fine-tuned ‘the fifth version of the You Only Look Once algorithm’ (YOLOv5) model is used to automatically detect the position of the bucket grab on the image plane. Subsequently, a novel geometric model is constructed, which takes the pixel position of the bucket grab, the steel rope length provided by the Programmable Logic Controller (PLC) system, and the optical lens information of the camera into consideration. The key parameters of this geometric model are statistically estimated by a novel iterative algorithm. Once the key parameters are estimated, the algorithm can automatically detect swing angles from video streams. Being analytically simple, the computation of our algorithm is fast, as it takes about 0.01 s to process one single image generated by the surveillance camera. Therefore, we are able to obtain an accurate and fast estimation of the swing angle of an operating crane in real-time applications. Simulation studies are conducted to validate the model and algorithm. Real video examples from Qingdao Seaport under various weather conditions are analyzed to demonstrate its practical performance.}, archive = {J_EAAI}, author = {Baichen Yu and Xiao Wang and Hansheng Wang}, doi = {10.1016/j.engappai.2024.109481}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109481}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A geometric model with stochastic error for abnormal motion detection of portal crane bucket grab}, volume = {139}, year = {2025}, }
• (2025). A dual-branch convolutional neural network with domain-informed attention for arrhythmia classification of 12-lead electrocardiograms. EAAI, 139, 109480. (www)

  @article{ , abstract = {The automatic classification of arrhythmia is an important task in the intelligent auxiliary diagnosis of an electrocardiogram. Its efficiency and accuracy are vital for practical deployment and applications in the medical field. For the 12-lead electrocardiogram, we know that the comprehensive utilization of lead characteristics is key to enhancing diagnostic accuracy. However, existing classification methods (1) neglect the similarities and differences between the limb lead group and the precordial lead group; (2) the commonly adopted attention mechanisms struggle to capture the domain characteristics in an electrocardiogram. To address these issues, we propose a new dual-branch convolutional neural network with domain-informed attention, which is novel in two ways. First, it adopts a dual-branch network to extract intra-group similarities and inter-group differences of limb and precordial leads. Second, it proposes a domain-informed attention mechanism to embed the critical domain knowledge of electrocardiogram, multiple RR (R wave to R wave) intervals, into coordinated attention to adaptively assign attention weights to key segments, thereby effectively capturing the characteristics of the electrocardiogram domain. Experimental results show that our method achieves an F1-score of 0.905 and a macro area under the curve of 0.936 on two widely used large-scale datasets, respectively. Compared to state-of-the-art methods, our method shows significant performance improvements with a drastic reduction in model parameters.}, archive = {J_EAAI}, author = {Rucheng Jiang and Bin Fu and Renfa Li and Rui Li and Danny Z. Chen and Yan Liu and Guoqi Xie and Keqin Li}, doi = {10.1016/j.engappai.2024.109480}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109480}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A dual-branch convolutional neural network with domain-informed attention for arrhythmia classification of 12-lead electrocardiograms}, volume = {139}, year = {2025}, }
• (2025). Interactive streaming feature selection based on neighborhood rough sets. EAAI, 139, 109479. (www)

  @article{ , abstract = {Feature streams refer to features that arrive continuously over time without changing the number of samples. Such data is commonly encountered in various practical application scenarios. Stream feature selection is a technique designed to select relevant features from high-dimensional stream data, thereby reducing its overall size. Feature interaction plays a crucial role in influencing the results of feature selection. Most existing methods address stream feature selection primarily by focusing on irrelevance and redundancy, often overlooking the important interactions between features. Additionally, these methods typically assume that all samples and features are known, which contradicts the fundamental nature of streaming data. This study introduces an interactive feature selection approach for stream feature selection, utilizing the neighborhood rough set. First, we provide a basic explanation of multi-neighbor entropy, which measures the amount of information related to neighborhood classes. It is used to measure how the amount of information about neighborhood classes. Next, we propose a feature evaluation method based on correlation, redundancy, and interaction analysis. Finally, we elaborate on functions for feature evaluation criteria, aiming to design streaming feature selection algorithms that integrate correlation, redundancy, and interactivity. The proposed algorithm is compared with six other representative feature selection algorithms across 14 public datasets. Experimental results demonstrate the validity of our proposed solution.}, archive = {J_EAAI}, author = {Gangqiang Zhang and Jingjing Hu and Jing Yang and Pengfei Zhang}, doi = {10.1016/j.engappai.2024.109479}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109479}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Interactive streaming feature selection based on neighborhood rough sets}, volume = {139}, year = {2025}, }
• (2025). Simulation-based genetic algorithm for optimizing a municipal cooperative waste supply chain in a pandemic. EAAI, 139, 109478. (www)

  @article{ , abstract = {The quantity of medical waste produced by municipalities is on the rise, potentially presenting significant hazards to both the environment and human health. Developing a robust supply chain network for managing municipal medical waste is important for society, especially during a pandemic like COVID-19. In supply chain network design, factors such as the collection of non-infectious waste, transporting infectious waste from hospitals to disposal facilities, revenue generation from waste-to-energy initiatives, and the potential for pandemic outbreaks are often overlooked. Hence, in this study, we design a model incorporating COVID-19 parameters to mitigate the spread of the virus while designing an effective municipal medical waste supply chain network during a pandemic. The proposed model is multi-objective, multi-echelon, multi-commodity and involves coalition-based cooperation. The first objective function aims to minimize total costs, while the second objective pertains to minimizing the risk of a COVID-19 outbreak. We identify optimal collaboration among municipal medical waste collection centers to maximize cost savings. The COVID-19 prevalence risk level by the waste in each zone is calculated pursuant to their inhabitants. Additionally, we analyze a system dynamic simulation framework to forecast waste generation levels amid COVID-19 conditions. A metaheuristic based on the Non-dominated Sorting Genetic Algorithm II is used to solve the problem and is benchmarked against exact solutions. To illustrate our approach, we present a case study focused on Tehran, Iran. The results show that an increase in the amount of generated waste leads to an increase in the total costs of the supply chain.}, archive = {J_EAAI}, author = {Peiman Ghasemi and Alireza Goli and Fariba Goodarzian and Jan Fabian Ehmke}, doi = {10.1016/j.engappai.2024.109478}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109478}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Simulation-based genetic algorithm for optimizing a municipal cooperative waste supply chain in a pandemic}, volume = {139}, year = {2025}, }
• (2025). A hybrid convolutional autoencoder training algorithm for unsupervised bearing health indicator construction. EAAI, 139, 109477. (www)

  @article{ , abstract = {Conventional Deep Learning (DL) methods for bearing health indicator (HI) adopt supervised approaches, requiring expert knowledge of the component degradation trend. Since bearings experience various failure modes, assuming a particular degradation trend for HI is suboptimal. Unsupervised DL methods are scarce in this domain. They generally maximise the HI monotonicity built in the middle layer of an Autoencoder (AE) trained to reconstruct the run-to-failure signals. The backpropagation (BP) training algorithm is unable to perform this maximisation since the monotonicity of HI subsections corresponding to input sample batches does not guarantee the monotonicity of the whole HI. Therefore, existing methods achieve this by searching AE hyperparameters so that its BP training to minimise the reconstruction error also leads to a highly monotonic HI in its middle layer. This is done using expensive search algorithms where the AE is trained numerous times using various hyperparameter settings, rendering them impractical for large datasets. To address this limitation, a small Convolutional Autoencoder (CAE) architecture and a hybrid training algorithm combining Particle Swarm Optimisation and BP are proposed in this work to enable simultaneous maximisation of the HI monotonicity and minimisation of the reconstruction error. As a result, the HI is built by training the CAE only once. The results from three case studies demonstrate this method’s lower computational burden compared to other unsupervised DL methods. Furthermore, the CAE-based HIs outperform the indicators built by equivalent and significantly larger models trained with a BP-based supervised approach, leading to 85% lower remaining useful life prediction errors.}, archive = {J_EAAI}, author = {Ali Eftekhari Milani and Donatella Zappalá and Simon J. Watson}, doi = {10.1016/j.engappai.2024.109477}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109477}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A hybrid convolutional autoencoder training algorithm for unsupervised bearing health indicator construction}, volume = {139}, year = {2025}, }
• (2025). Latent temporal smoothness-induced schatten-p norm factorization for sequential subspace clustering. EAAI, 139, 109476. (www)

  @article{ , abstract = {This paper presents an innovative latent temporal smoothness-induced Schatten- p norm factorization (SpFLTS) method aimed at addressing challenges in sequential subspace clustering tasks. Globally, SpFLTS employs a low-rank subspace clustering framework based on Schatten-2/3 norm factorization to enhance the comprehensive capture of the original data features. Locally, a total variation smoothing term is induced to the temporal gradients of latent subspace matrices obtained from sub-orthogonal projections, thereby preserving smoothness in the sequential latent space. To efficiently solve the closed-form optimization problem, a fast Fourier transform is combined with the non-convex alternating direction method of multipliers to optimize latent subspace matrix, which greatly speeds up computation. Experimental results demonstrate that the proposed SpFLTS method surpasses existing techniques on multiple benchmark databases, highlighting its superior clustering performance and extensive application potential.}, archive = {J_EAAI}, author = {Yuan Xu and Zhen-Zhen Zhao and Tong-Wei Lu and Wei Ke and Yi Luo and Yan-Lin He and Qun-Xiong Zhu and Yang Zhang and Ming-Qing Zhang}, doi = {10.1016/j.engappai.2024.109476}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109476}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Latent temporal smoothness-induced schatten-p norm factorization for sequential subspace clustering}, volume = {139}, year = {2025}, }
• (2025). Involution fused convolution for classifying eye-tracking patterns of children with autism spectrum disorder. EAAI, 139, 109475. (www)

  @article{ , abstract = {Autism Spectrum Disorder (ASD) is a neurological condition that is challenging to diagnose. Numerous studies demonstrate that children diagnosed with autism struggle with maintaining attention spans and have less focused vision. The eye-tracking technology has drawn special attention in the context of ASD since anomalies in gaze have long been acknowledged as a defining feature of autism. Deep Learning (DL) approaches coupled with eye-tracking sensors are exploiting additional capabilities to advance the diagnostic and its applications. DL architectures like convolution have been dominating this domain. However, convolutions alone may be insufficient to capture the important spatial information in eye-tracking patterns, as these patterns are more likely to be localized. The dynamic kernel-based process known as involutions can improve the classification efficiency. In this study, we utilize two image-processing operations to see how these processes learn eye-tracking patterns. Since these patterns are primarily based on spatial information, we employ a hybrid of involution and convolution. Our study shows that adding a few involution layers reduces size and computational cost while enhancing location-specific capabilities, maintaining performance comparable to pure convolutional models. However, excessive involution layers lead to weaker performance. For comparisons, we experiment with two datasets and a combined version of both in ablation studies. Our proposed model, featuring three involution layers and three convolution layers, achieved 99.43% accuracy on the first dataset and 96.78% on the second, with a size of only 1.36 megabytes. These results showcase the effectiveness of combining involution and convolution layers which outperforms previous literature.}, archive = {J_EAAI}, author = {Md. Farhadul Islam and Meem Arafat Manab and Joyanta Jyoti Mondal and Sarah Zabeen and Fardin Bin Rahman and Md. Zahidul Hasan and Farig Sadeque and Jannatun Noor}, doi = {10.1016/j.engappai.2024.109475}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109475}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Involution fused convolution for classifying eye-tracking patterns of children with autism spectrum disorder}, volume = {139}, year = {2025}, }
• (2025). Power transformer health index and life span assessment: A comprehensive review of conventional and machine learning based approaches. EAAI, 139, 109474. (www)

  @article{ , abstract = {Power transformers play a critical role within the electrical power system, making their health assessment and the prediction of their remaining lifespan paramount for the purpose of ensuring efficient operation and facilitating effective maintenance planning. This paper undertakes a comprehensive examination of existent literature, with a primary focus on both conventional and cutting-edge techniques employed within this domain. The merits and demerits of recent methodologies and techniques are subjected to meticulous scrutiny and explication. Furthermore, this paper expounds upon intelligent fault diagnosis methodologies and delves into the most widely utilized intelligent algorithms for the assessment of transformer conditions. Diverse Artificial Intelligence (AI) approaches, including Artificial Neural Networks (ANN) and Convolutional Neural Network (CNN), Support Vector Machine (SVM), Random Forest (RF), Genetic Algorithm (GA), and Particle Swarm Optimization (PSO), are elucidated offering pragmatic solutions for enhancing the performance of transformer fault diagnosis. The amalgamation of multiple AI methodologies and the exploration of time-series analysis further contribute to the augmentation of diagnostic precision and the early detection of faults in transformers. By furnishing a comprehensive panorama of AI applications in the field of transformer fault diagnosis, this study lays the groundwork for future research endeavors and the progression of this critical area of study.}, archive = {J_EAAI}, author = {Syeda Tahreem Zahra and Syed Kashif Imdad and Sohail Khan and Sohail Khalid and Nauman Anwar Baig}, doi = {10.1016/j.engappai.2024.109474}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109474}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Power transformer health index and life span assessment: A comprehensive review of conventional and machine learning based approaches}, volume = {139}, year = {2025}, }
• (2025). Enhanced cross layer refinement network for robust lane detection across diverse lighting and road conditions. EAAI, 139, 109473. (www)

  @article{ , abstract = {With the rapid development of autonomous driving technology, lane detection, a key component of intelligent vehicle systems, is crucial for ensuring road safety and efficient vehicle navigation. In this paper, a new lane detection method is proposed to address the problem of degraded performance of existing lane detection methods when dealing with complex road environments. The proposed method evolves from the original Cross Layer Refinement Network (CLRNet) by incorporating two of our carefully designed core components: the Global Feature Optimizer (GFO) and the Adaptive Lane Geometry Aggregator (ALGA). The GFO is a multi-scale attention mechanism that mimics the human visual focusing ability, effectively filtering out unimportant information and focusing on the image regions most relevant to the task. The ALGA is a shape feature-aware aggregation module that utilizes the shape prior of lanes to enhance the correlation of anchor points in an image, better fusing global and local information. By integrating both components into CLRNet, an enhanced version called Enhanced CLRNet (E-CLRNet) is presented, which exhibits higher performance stability in complex roadway scenarios. Experiments on the CULane dataset reveal that E-CLRNet demonstrates superior performance stability over the original CLRNet in complex scenarios, including curves, shadows, missing lines, and dazzling light conditions. In particular, in the curves, the F1 score of E-CLRNet is improved by almost 3% over the original CLRNet. This study not only improves the accuracy and performance stability of lane detection but also provides a new solution for the application of autonomous driving technology in complex environments, which promotes the development of intelligent vehicle systems.}, archive = {J_EAAI}, author = {Weilong Dai and Zuoyong Li and Xiaofeng Xu and Xiaobo Chen and Huanqiang Zeng and Rong Hu}, doi = {10.1016/j.engappai.2024.109473}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109473}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Enhanced cross layer refinement network for robust lane detection across diverse lighting and road conditions}, volume = {139}, year = {2025}, }
• (2025). Designing an adaptive learning framework for predicting drug-target affinity using reinforcement learning and graph neural networks. EAAI, 139, 109472. (www)

  @article{ , abstract = {In the field of biomedical engineering, predicting Drug-Target Affinities (DTA) is crucial. However, current affinity prediction models are predominantly manually designed, which is a complex, time-consuming process that may not effectively accommodate the diversity and complexity of datasets. To address this challenge, we propose an adaptive learning framework for predicting drug-target affinity, called Adaptive-DTA, which integrates reinforcement learning with graph neural networks to automate the design of affinity prediction models. Adaptive-DTA defines the architecture search space using directed acyclic graphs and employs an reinforcement learning algorithm to guide the architecture search, optimizing parameters based on the entropy of sampled architectures and model performance metrics. Additionally, we enhance efficiency with a two-stage training and validation strategy, incorporating low-fidelity and high-fidelity evaluations. Our framework not only alleviates the challenges associated with manual model design but also significantly improves model performance and generalization. To evaluate the performance of our method, we conducted extensive experiments on DTA benchmark datasets and compared the results with nine state-of-the-art methods. The experimental outcomes demonstrate that our proposed framework outperforms these methods, exhibiting outstanding performance in predicting drug-target affinities. Our innovative approach streamlines the design of affinity prediction model, reduces reliance on manual crafting, and enhances model generalization. Its ability to automatically optimize network architectures represents a major step forward in the automation of computational drug discovery processes.}, archive = {J_EAAI}, author = {Jun Ma and Zhili Zhao and Yunwu Liu and Tongfeng Li and Ruisheng Zhang}, doi = {10.1016/j.engappai.2024.109472}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109472}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Designing an adaptive learning framework for predicting drug-target affinity using reinforcement learning and graph neural networks}, volume = {139}, year = {2025}, }
• (2025). Implicit local–global feature extraction for diffusion sequence recommendation. EAAI, 139, 109471. (www)

  @article{ , abstract = {The existing research using diffusion model for item distribution modeling is a novel and effective recommendation method. However, the user interaction sequences contain multiple implicit features that reflect user preferences, and how to use implicit features to guide the diffusion process remains to be studied. Therefore, considering the dynamics of user preferences, we conduct fine-grained modeling of diffusion recommendation process. Specifically, we firstly define a sequence feature extraction layer that utilizes multi-scale convolutional neural networks and residual long short-term memory networks to learn local–global implicit features, and obtains implicit features through a weighted fusion strategy. Subsequently, the extracted output features are used as conditional inputs for the diffusion recommendation model to guide the denoising process. Finally, the items that meet user preferences are generated through the sampling and inference process to realize the personalized recommendation task. Through experiments on three publicly available datasets, the results show that the proposed model outperforms the strong baseline model in terms of performance. In addition, we conduct hyperparameter analysis and ablation experiments to verify the impact of model components on overall performance.}, archive = {J_EAAI}, author = {Yong Niu and Xing Xing and Zhichun Jia and Ruidi Liu and Mindong Xin}, doi = {10.1016/j.engappai.2024.109471}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109471}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Implicit local–global feature extraction for diffusion sequence recommendation}, volume = {139}, year = {2025}, }
• (2025). Multi-scale information sharing and selection network with boundary attention for polyp segmentation. EAAI, 139, 109467. (www)

  @article{ , abstract = {Polyp segmentation in colonoscopy images is essential in clinical practice, offering valuable information for the diagnosis of colorectal cancer and subsequent surgical procedures. Despite the relatively good performance of existing methods, polyp segmentation still faces the following challenges: (1) Varying lighting conditions in colonoscopy and differences in polyp locations, sizes, and morphologies. (2) The indistinct boundary between polyps and surrounding tissue. To tackle these challenges, we propose a Multi-scale Information Sharing and Selection Network (MISNet) for the polyp segmentation task. We have designed a Selectively Shared Fusion Module (SSFM) to facilitate information sharing and the active selection between low-level and high-level features, thus enhancing the model’s ability to capture comprehensive information. Subsequently, we have developed a Parallel Attention Module (PAM) to improve the model’s attention on boundaries, and a Balancing Weight Module (BWM) to support the continuous refinement of boundary segmentation through the bottom-up process. Extensive experiments on five benchmark datasets show competitive results compared to existing representative methods. Specifically, our method has reached the mean Dice coefficient of 0.903 and 0.918 on the Kvasir and CVC-ClinicDB datasets, 0.762 and 0.764 on the challenging CVC-ColonDB and ETIS datasets. These innovative modules in our proposed MISNet effectively address key challenges, providing a robust solution for accurate polyp segmentation in clinical diagnosis and treatment. The proposed model is available at https://github.com/q1216355254/MISNet.git .}, archive = {J_EAAI}, author = {Xiaolu Kang and Zhuoqi Ma and Kang Liu and Yunan Li and Qiguang Miao}, doi = {10.1016/j.engappai.2024.109467}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109467}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Multi-scale information sharing and selection network with boundary attention for polyp segmentation}, volume = {139}, year = {2025}, }
• (2025). Personalized hierarchical heterogeneous federated learning for thermal comfort prediction in smart buildings. EAAI, 139, 109464. (www)

  @article{ , abstract = {Federated Learning (FL) is gaining significant traction due to its ability to provide security and privacy. In the FL paradigm, the global model is learned at the cloud through the consolidation of local model parameters instead of collecting local training data at the central node. This approach mitigates privacy leakage caused by the collection of sensitive information. However, it poses challenges to the convergence of the global model due to system and statistical heterogeneity. In this study, we propose a two-fold Personalized Hierarchical Heterogeneous FL (PHHFL) approach. It leverages a hierarchical structure to handle statistical heterogeneity and a normal distribution-based client selection to control model divergence in FL environment. PHHFL aims to use a maximum number of local features of each client and assign specific level in the hierarchy. Furthermore, to address model divergence caused by the nodes’ statistical heterogeneity, we propose a novel client selection strategy based on the performance distribution of the nodes. Experiments are conducted on thermal comfort datasets and a synthetic dataset with 12 and 10 clients, respectively. The results show that the proposed PHHFL outperforms in terms of accuracy, F1 score, and class-wise precision on both thermal comfort and synthetic datasets. The source code of the PHHFL model and datasets is available on GitHub .}, archive = {J_EAAI}, author = {Atif Rizwan and Anam Nawaz Khan and Rashid Ahmad and Qazi Waqas Khan and Do Hyeun Kim}, doi = {10.1016/j.engappai.2024.109464}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109464}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Personalized hierarchical heterogeneous federated learning for thermal comfort prediction in smart buildings}, volume = {139}, year = {2025}, }
• (2025). GFML: Gravity function for metric learning. EAAI, 139, 109463. (www)

  @article{ , abstract = {Diverse machine learning algorithms rely on the distance metric to compare and aggregate the information. A metric learning algorithm that captures the relevance between two vectors plays a critical role in machine learning. Metric learning may become biased toward the major classes and not be robust to the minor ones, i.e., metric learning may be vulnerable in an imbalanced dataset. We propose a gravity function-based metric learning (GFML) that captures the relationship between vectors based on the gravity function. We formulate GFML with two terms, (1) mass of the given vectors and (2) distance between the query and key vector. Mass learns the importance of the object itself, enabling robust metric learning on imbalanced datasets. GFML is simple and scalable; therefore, it can be adopted in diverse tasks. We validate that GFML improves the recommender system and image classification.}, archive = {J_EAAI}, author = {Hoyoon Byun and Sungjun Lim and Kyungwoo Song}, doi = {10.1016/j.engappai.2024.109463}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109463}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {GFML: Gravity function for metric learning}, volume = {139}, year = {2025}, }
• (2025). A generative design method of airfoil based on conditional variational autoencoder. EAAI, 139, 109461. (www)

  @article{ , abstract = {The challenges in multi-objective and multi-dimensional optimization design of airfoils, marked by prolonged optimization cycles and low accuracy, call for an efficient solution to expedite airfoil design. This study presents an innovative airfoil generative design model based on a conditional variational autoencoder (CVAE). Initially, to overcome the limitation of insufficient training data, the model leverages the variational autoencoder (VAE) to learn the spatial distribution of University of Illinois at Urbana-Champaign (UIUC) airfoils, enabling the generation of a diverse set of airfoils with similar distributions. Subsequently, two CVAE-based airfoil generation models, the airfoil freedom design model and the airfoil precision design model, are proposed, which can realize diverse airfoil design under different conditions, such as shape and aerodynamic conditions. Furthermore, two measurements of roughness and diversity are introduced to evaluate the quality of the generated airfoils. The impact of different conditions and network parameters on the model’s generation performance is thoroughly analyzed. Results indicate that our proposed model achieves a 65% lower error compared to physics-guided conditional Wasserstein generative adversarial networks (PG-cWGAN) when generating airfoils that satisfy a specific lift coefficient and a 99.99% lower error compared to airfoil pressure distributions generative adversarial networks (Airfoil-Cp-GAN) when generating airfoils that satisfy specific pressure distributions. This method introduces a more creative and accurate approach for aircraft designers in the realm of airfoil design. The code used for this paper is available at https://github.com/liujun39/airfoilvae .}, archive = {J_EAAI}, author = {Xu Wang and Weiqi Qian and Tun Zhao and Hai Chen and Lei He and Haisheng Sun and Yuan Tian}, doi = {10.1016/j.engappai.2024.109461}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109461}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {A generative design method of airfoil based on conditional variational autoencoder}, volume = {139}, year = {2025}, }
• (2025). RMFDNet: Redundant and missing feature decoupling network for salient object detection. EAAI, 139, 109459. (www)

  @article{ , abstract = {Recently, many salient object detection methods have utilized edge contours to constrain the solution space. This approach aims to reduce the omission of salient features and minimize the inclusion of non-salient features. To further leverage the potential of edge-related information, this paper proposes a Redundant and Missing Feature Decoupling Network (RMFDNet). RMFDNet primarily consists of a segment decoder, a complement decoder, a removal decoder, and a recurrent repair encoder. The complement and removal decoders are designed to directly predict the missing and redundant features within the segmentation features. These predicted features are then processed by the recurrent repair encoder to refine the segmentation features. Experimental results on multiple Red–Green–Blue (RGB) and Red–Green–Blue-Depth (RGB-D) benchmark datasets, as well as polyp segmentation datasets, demonstrate that RMFDNet significantly outperforms previous state-of-the-art methods across various evaluation metrics. The efficiency, robustness, and generalization capability of RMFDNet are thoroughly analyzed through a carefully designed ablation study. The code will be made available upon paper acceptance.}, archive = {J_EAAI}, author = {Qianwei Zhou and Jintao Wang and Jiaqi Li and Chen Zhou and Haigen Hu and Keli Hu}, doi = {10.1016/j.engappai.2024.109459}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109459}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {RMFDNet: Redundant and missing feature decoupling network for salient object detection}, volume = {139}, year = {2025}, }
• (2025). Dynamically adaptive deformable feature fusion for multi-scale character detection in ancient documents. EAAI, 139, 109458. (www)

  @article{ , abstract = {Robust character detection of various sizes in ancient East-Asian handwritten documents is essential for accurate classification and translation. While prior studies have addressed various challenges associated with the writing styles and page layouts of historical documents, they struggle to detect small-sized characters, especially with an area below 32 square units due to their limited presence in the data corpus. Furthermore, the physical degradation, presence of artifacts, and inconsistencies in text density complicate character detection of historical documents across multiple scales. In this study, we propose a novel multi-scale character detection named Dynamically Adaptive Deformable Feature Fusion (DAF). This approach leverages deformable convolutions to improve feature extraction for the complex and irregular shapes found in ancient East-Asian manuscripts. We also present an innovative Adaptive Weight Module that dynamically adjusts top-down features across different levels by utilizing trainable weights. This enhances the detection of multi-scale characters and effectively identifies small-sized characters within documents. Further, we contribute to existing research by proposing a set of detection metrics, specifically designed to evaluate both general and scale-specific detection scenarios. Extensive experiments conducted on several datasets of ancient handwritten documents including the Nancho dataset, Multiple Tripitaka in Han dataset (MTHv2), and Kuzushiji dataset demonstrate the superior performance of our proposed DAF framework over existing multi-scale detection methods.}, archive = {J_EAAI}, author = {Mauricio Bermudez-Gonzalez and Amin Jalali and Minho Lee}, doi = {10.1016/j.engappai.2024.109458}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109458}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Dynamically adaptive deformable feature fusion for multi-scale character detection in ancient documents}, volume = {139}, year = {2025}, }
• (2025). Diophantine spherical vague sets and their applications for micro-technology robots based on multiple-attribute decision-making. EAAI, 139, 109447. (www)

  @article{ , abstract = {We introduce the concept of Diophantine spherical vague set approach to multiple-attribute decision-making. The Spherical vague set is a novel expansion of the vague set and interval valued spherical fuzzy set. Three new concepts have been introduce such as Diophantine spherical vague weighted averaging operator, Diophantine spherical vague weighted geometric operator, generalized Diophantine spherical vague weighted averaging operator and generalized Diophantine spherical vague weighted geometric operator. We provide a numerical example to show how Euclidean distance and Hamming distance interact. Applications of the Diophantine spherical vague number include idempotency, boundedness, commutativity and monotonicity in algebraic operations. They can determine the optimal option and are more well-known and reasonable. Our goal was to identify the optimal choice by comparing expert opinions with the criteria. As a result, the model’s output was more accurate as well as in the range of the natural number . The weighted averaging distance and weighted geometric distance operators are distance measure that is based on aggregating model. By comparing the models under discussion with those suggested in the literature, we hoped to show their worth and reliability. It is possible to find a better solution more quickly, simply, and practically. Our objective was to compare the expert evaluations with the criteria and determine which option was the most suitable. Because they yield more precise solutions, these models are more accurate and more related to models with . To show the superiority and the validity of the proposed aggregation operations, we compared it with the existing method and concluded from the comparison and sensitivity analysis that our proposed technique is more effective and reliable. This investigation yielded some intriguing results.}, archive = {J_EAAI}, author = {Murugan Palanikumar and Nasreen Kausar and Željko Stević and Sarfaraz Hashemkhani Zolfani}, doi = {10.1016/j.engappai.2024.109447}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109447}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Diophantine spherical vague sets and their applications for micro-technology robots based on multiple-attribute decision-making}, volume = {139}, year = {2025}, }
• (2025). Reconstruction error based implicit regularization method and its engineering application to lung cancer diagnosis. EAAI, 139, 109439. (www)

  @article{ , abstract = {The automatic diagnosis of lung cancer via artificial intelligence faces two hotspot issues: (1) insufficient data and (2) excessive redundant information, which make it difficult for convolutional neural networks (CNNs) to learn discriminative information of lung cancer. In this paper, we present the reconstruction error based implicit regularization method (REbIRM) that regularizes CNNs at the loss layer. During each training iteration, the reconstruction errors introduced by the two-stage discriminative auto-encoder are used to sharpen the generalization ability of deep CNNs by improving the decision boundary. In the application process, the trained deep CNN is used for completing computed tomography (CT) diagnostics. The main clinical benefit of our approach is that it is domain independent, requiring no specialized knowledge, and can therefore be applied to different types of datasets. To the best of our knowledge, this is the first attempt to implicitly regularize CNNs based on the reconstruction errors. Finally, experimental results on three CT image classification datasets show that REbIRM can achieve impressive results and that, in conjunction with Dropout, it obtains the state-of-the-art performance. REbIRM is also robust to the selection of hyper-parameters and only has the sublinear influence on the convergence of deep CNNs. Besides, empirical and theoretical evidence are provided to indicate that REbIRM prefers to converges in a constrained parameter space with flatter minima, which explains why it can generalize to new data. Finally, the nature of REbIRM is further explored through visualization techniques to analyze how it works in training deep CNNs.}, archive = {J_EAAI}, author = {Qinghe Zheng and Xinyu Tian and Mingqiang Yang and Shuang Han and Abdussalam Elhanashi and Sergio Saponara and Kidiyo Kpalma}, doi = {10.1016/j.engappai.2024.109439}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109439}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Reconstruction error based implicit regularization method and its engineering application to lung cancer diagnosis}, volume = {139}, year = {2025}, }
• (2025). Adaptive graph learning algorithm for incomplete multi-view clustered image segmentation. EAAI, 139, 109264. (www)

  @article{ , abstract = {There are problems of relying on data initialization and ignoring data structure in existing incomplete multi-view clustering algorithms, an adaptive graph learning incomplete multi-view clustering image segmentation algorithm is proposed. Firstly, the similarity matrix of each non missing view is adaptive learned, and the index matrix of the missing view is used to complete the similarity matrix and unify the dimensions,which ensure the authenticity of the data and revealing the data structure. Secondly, the low dimension representation of the complete similarity matrix under spectral constraints is calculated, and a discrete clustering index matrix is directly obtained through adaptive weighted spectral rotation, avoiding post-processing. The clustering index matrix is used to obtain clustering of multi-view features, thereby obtaining image segmentation results. Finally, an iterative algorithm optimization model is presented, which is compared with six existing algorithms using seven evaluation metrics on six datasets. The results show significant improvements in clustering performance and segmentation performance.}, archive = {J_EAAI}, author = {Junhui Cao and Jing Hu and Rongguo Zhang}, doi = {10.1016/j.engappai.2024.109264}, journal = {Engineering Applications of Artificial Intelligence}, month = {1}, pages = {109264}, shortjournal = {Eng. Appl. Artif. Intell.}, title = {Adaptive graph learning algorithm for incomplete multi-view clustered image segmentation}, volume = {139}, year = {2025}, }